SmartGridCity TM Pricing Pilot Program Pilot Evaluation and Final Report December 19 th, 2013

SmartGridCity TM Pricing Pilot Program Pilot Evaluation and Final Report December 19 th, 2013 This SmartGridCity TM (SGC) Pricing Pilot Program Pilot Evaluation and Final Report ( report ) is submitted as required by, and in full compliance with, the Stipulation and Settlement Agreement approved in Decision No. C10-0491 in Docket No. 09A-796E. The report includes the activities, participation, budget, full regression analysis and lessons learned from the Pricing Pilot Program. Per the settlement agreement the Public Service Company of Colorado (the Company or PSCo) has provided an interim report at the conclusion of each calendar year of the pilot. This is the final report. EnerNOC was contracted to provide the third party evaluation of this pilot. Their final report is attached and includes the complete evaluation of the pilot including a summary of key findings and recommendations should the Company decide to pursue a broader deployment of advanced rates in the future. The following sections provide a summary of the activities, participation, and budget and address the key research questions the Company set out to answer through this effort. Summary This report marks the completion of a 3 year, comprehensive rate pilot designed to determine the impact time differentiated pricing structures have on customer behavior. With nearly 4,000 customers (~20% of Boulder SmartGridCity) participating in this pilot, participation levels matched that of similar pilot programs deployed elsewhere in the industry. The pilot confirmed that these rates positively impact customer behavior resulting in energy savings with each of the three pilot rates. Most notably, the Critical Peak Pricing (CPP) rate provided greater load reductions than both the Peak Time Rebate (PTR) and Time of Use (TOU) rates. Customers on the CPP rate realized reduced peak demand approaching 30%, PTR participants realized load reductions of nearly 15% and TOU participants averaged 5-9%. Despite the fact that this pilot confirmed that these rates provide both load reduction and energy savings, it did not prove cost effective. Further details on how the cost effectiveness could be improved are discussed within this report. The report below provides additional detail on the insights gained through this effort. 1

Background In 2009, the Company proposed a dynamic pricing pilot within the City of Boulder to be enabled by the advanced metering deployed as part of SGC. Upon review of the proposal the Colorado Public Utilities Commission (Commission) recommended changes (per the Stipulation agreement) which were included in the program. The Company recruited participants into the pilot per the Commission s recommendations, and in October 2010 the pilot rates went into effect. The SmartGridCity Pricing Pilot Program offered residential customers three pricing structures (rates) to test customer understanding, acceptance and behavioral change resulting from the use of pricing signals. Participating customers could select from one of the following three rate options: 1. Critical Peak Pricing (CPP) The CPP rate included the on and off-peak aspects of the TOU while also having a third tier of usage. When system capacity or economic conditions warranted (Peak Energy Event days), this third tier became effective. Off-peak (year round) $0.04 On-peak (summer) $0.12 (non-summer) $0.05 Peak Energy Event Price per kwh (summer) $0.51 (non-summer) $0.33 2. Peak Time Rebate (PTR) The PTR rate allowed customers to pay the standard residential rate but are offered a rebate when they reduced their usage at critical peak times or during Peak Energy Event days Peak Energy Event Price per kwh (Summer rebate) $0.29 (non-summer rebate) $0.47 3. Time-of-Use (TOU) The TOU rate divided each day into two periods representing on and off peak with rates being higher during the on-peak period. This rate did not participate in Peak Energy Events. Off-peak (Year round) $0.04 On-peak (Summer) $0.17 (non-summer) $0.06 Below is an example of the marketing materials used during the recruitment phase of the pilot. The brochure was designed to explain each of the new pricing options as it compared to the standard tiered residential rate structure. 2

3

Study Objectives The purpose of the SmartGridCity Pricing Pilot was to test customer understanding, acceptance and the resulting behavioral change resulting from the use of pricing signals. Eight key objectives were enumerated in the Company s filing for this project. Those objectives along with the findings from the study are detailed bellow: 1. Does the rate reduce peak demand and energy consumption? The analysis of the pilot confirmed that yes, time differentiated rates do influence customer s usage during peak times as well as reducing their overall energy usage. As reported in the EnerNOC report (Page 58 Key Findings 3 rd bullet), CPP Phase I participants reduced loads by 22-29%, which represents average on-peak demand reductions of 0.22-0.28 kw, while PTR Phase I participants had load reductions of 8-14%, or 0.09-0.14 kw. TOU provided load reduction during the on-peak period. Load reductions for Phase I participants averaged 5-9% across the three years of the pilot, representing average on-peak demand reduction of 0.05-0.12 kw. (Page 58 Key Findings 5 th bullet). Other key observations across the life of the pilot include: Savings during the non-summer season were lower than the summer impacts due to reduced non-summer loads (Page viii 1 st bullet) Peak-event demand reductions remained consistent across the duration of the events Peak demand reductions lessened over time while annual energy savings increased Phase I ( opt-in ) participants outperformed Phase II ( opt-out ) 2. Does the rate reduce a carbon footprint? Yes, by reducing overall energy usage, which advanced rates have been shown to do, a customer s overall carbon footprint is reduced. However, this is purely a function of reducing energy use and not a function of when, on-peak or off-peak, that energy was used. This is because the overall carbon intensity of the Company's generation mix is a function of the incremental resources available for dispatch rather than the particular time of day. 3. Does the rate defer capital spending for distribution and transmission? The pilot rates as deployed did not defer capital spending for distribution or transmission as the number of participants and resulting load relief were small compared to overall system load. Should these types of rates be introduced across the service territory the aggregate impact could reduce the Company s overall capacity needs, possibly including the transmission requirements associated with that capacity. However, the extent to which any capital deferment is realized would be dependant upon overall penetration. Under any scenario there would be limited benefit on the local distribution level as customers load was only reduced on a limited number of peak event days. Customer s load was only reduced significantly on system peak days but these days 4

do not necessarily coincide with the customer s peak usage. Therefore the distribution system must still be built to accommodate the customer s full load on non-system peak days. In addition, distribution impacts are a function of the specific feeder routes for those customers participating and are difficult to determine the level of impact 4. How can the pilot be used to inform the Company and the Commission on whether these advanced rates can be offered cost effectively? At the conclusion of the pilot a benefit-cost analysis was performed. The analysis took into consideration the benefit and cost impacts on all ratepayers (Rate Impact Test) achieved during the three years of the pilot. The benefits were captured by summing the capacity and energy costs which might be avoided. The costs were those associated with delivering the pilot over the same time period.. This analysis excluded the costs of the smart meters and billing system enhancements as those costs were assumed to be part of overall system operations and not to be born solely by the advanced rates. A summary of benefits and costs is shown in the table below: Table 1 Cost-Benefit Analysis SGC Pricing Pilot Benefit Cost Analysis 2011 2012 2013 Total Benefits Avoided Capacity (kw) $ 10,940 $ 11,356 $ 7,787 $ 30,083 Avoided Energy (kwh) $ 6,824 $ 16,450 $ 22,149 $ 45,423 Total Benefits $ 17,764 $ 27,806 $ 29,936 $ 75,506 Pilot Costs Program Planning & Design $ 33,421 $ 28,818 $ 3,600 $ 65,839 Administration & Program Delivery $ 240,940 $ 251,730 $ 166,900 $ 659,570 Advertising/Promotion/Customer Ed $ 101,675 $ 3,213 $ 2,000 $ 106,887 Incentives (Rebates) $ 36,713 $ 11,179 $ 11,325 $ 59,217 Reduction in sales Revenue $ 20,956 $ 56,701 $ 58,174 $ 135,831 Total Costs $ 433,704 $ 351,642 $ 241,999 $ 1,027,345 Benefit cost ratio 1/ benefit cost 0.0735 13.6 The pilot results showed that the advanced rates tested are not cost effective as deployed, but the pilot does provide guidance as to how the cost effectiveness might be improved should programs be deployed in the future. Should advanced rates be offered on a broader basis several steps could be taken to improve the benefit to cost ratio, notably: Leverage economies of scale some of the costs are fixed and do not increase as the number of participants increase. Automation 1. Certain tasks which were performed manually because the pilot was of a fixed duration could be automated if advanced rates were offered to a larger population on an on-going basis 2. Administering the PTR rate accounted for nearly one-half of the labor costs. 5

Targeted customer recruiting Targeting only those customers who could benefit most from the rates, and thus provide the most benefit to the system could greatly increase overall benefits. These would primarily be customers with central air conditioning. Retain participants over the course of the pilot nearly 40% of the participants left, resulting in fewer participants contributing to the overall benefits each year. This attrition was primarily driven by moves. For evaluation purposes participants who moved could not rejoin the pilot at their new premise. As customer recruitment is one of the most expensive items it would greatly benefit a program if participants could retain the rate if they moved within the service territory. Perhaps the most important lesson learned from performing the benefit-cost analysis is that there are significant obstacles to overcome before pursuing a broader deployment of advanced rates: Smart meters - CPP and PTR rates require meters with interval data capability. The benefits from these rates alone cannot support justification for deployment of smart meters. Therefore our ability to offer these rates broadly is contingent upon the Company s metering strategy. Impact on other DR programs - The greatest on-peak reductions occurred from participants with central air conditioning. Targeting these customers would be most beneficial for advanced rates, but may have the effect of cannibalizing the Company s Saver s Switch program. This highly costeffective demand response program, Saver s Switch, already has strong penetration within our service territory and delivers considerably more kw per participant than was achieved through our Pricing Pilot. A careful analysis must be done to determine if the overall benefit of adding customers to pricing options could overcome any negative impacts to programs such as Saver s Switch. Impact on non-participants As previously mentioned a favorable or best case slant was taken when looking at the benefits achieved over the pilot period. Before deciding to pursue a broader offering of advanced rates a more in depth analysis should be performed taking into account lower value of capacity during years where the system is long on generation and taking into consideration a longer view of future energy prices. In spite of these findings there may be societal or long term strategic reasons that would suggest value in deploying such advanced rate structures in the future. If so it will be important to use this type of analysis to determine the true impact on those ratepayers not participating in the advanced rate structures if the total cost of those advanced rate structures are not borne by the participants. 6

5. What are the demographics and characteristics of the customers who provide the greatest response (load reduction) due to pricing signals? Answering this question is difficult as the analysis chosen for the Report leverages the difference of differences approach which looks at the aggregate. A customer with greater load and central A/C could benefit from these rates while a customer with minimal load may be best served by our standard rates. 6. How well do customers understand pricing signals, and how do these signals impact customers perceptions of PSCo? This is a difficult question to answer concerning the current environment in Boulder. Leveraging insights captured from the online panel deployed in 2012, it can be assumed from the panel that customer s perceptions of the Company have not changed. More specifically, when asked to Please rate your overall impressions of the SmartGridCity Pricing Pilot, the top three responses (Where 0 is Not at all satisfied and 10 is Very satisfied ), 30-40% of respondents rated this with high satisfaction (Scored 8 or higher). The Company did not specifically ask customers understanding of pricing signals but as mentioned by EnerNOC in their report (Page ix Key Benefits 15th bullet), Based upon the relative savings in the various periods, customers appeared to understand the price signals. 7. In conjunction with pricing signals, how does the voluntary use of in-home devices impact energy and demand relative to those without such devices? As was reported in each of the monthly attrition reports, the number of installed In Home Smart Devices (IHSDs) customers within the SmartGridCity Pricing Pilot was not statistically significant. With this in mind, leveraging their regression analysis, EnerNOC found that for those customers with IHSDs, there were higher savings during summer events (Page ix Key Benefits 13th bullet). 8. How do various combinations of in-home devices (or lack thereof) and SmartGridCity Pricing options impact energy and demand relative to other combinations? When the study was initially developed, the plan was to test and deploy multiple devices. Upon completion of our testing, only one device passed testing and was deployed. As such, the study was unable to assess this measure. Activities During the life of the Pricing Pilot a number of key accomplishments were made. While the majority of 2011 s activities consisted of standing-up the Pricing Pilot, 2012 shifted the focus to customer engagement. This same focus was carried into 2013. Below are highlights of those accomplishments by year: 7

2011 (Customer Recruitment and Operational Focus): Successfully deployed billing infrastructure with scalability and automation levels sufficient to support scope of pilot program. Greater levels of automation would be necessary to provide dynamic rates and additional customer pricing choices. Completed customer recruiting campaigns development and deployment. Developed and documented processes and procedures to support Peak Energy Events and the broader Pricing Pilot Program. Where applicable, each process was developed to include service level agreements (SLA s) as a means to ensure critical processes were executed in alignment with the tariff (e.g. customer communications deployed by 16:00 MT the day prior to a Peak Energy Event). 2012-2013 (Customer Engagement): Successfully deployed SMS (text) as an additional Peak Event notification medium providing customers further notification choices. With the addition of text, customers could now choose a combination of email, phone and text notification in support of Peak Energy Events. Customers were able to change these preferences at will throughout the life of the pilot. Prior to the 2012 event seasons, informational letters were provided to pilot participants. These communications reminded customers of their participation in the pilot, reviewed what takes place during Peak energy events; their purpose, frequency and how to best manage their communications preferences. Subsequent communications expanded on these messages by sharing best practices and peer experiences on how to best save energy and money. For 2011 s event season the Company provided customers with a year end summary of the pilot s achievements. For 2012 and 2013, a post event reporting framework was created. This framework allowed the Company to provide customers dynamic reports on their performance during the previous month s Peak Energy Events. In addition to providing their performance, these reports highlighted their performance against their peers on the same rate while providing dynamic energy saving tips. In 2012, a panel of 122 customers was recruited to take a series of four online surveys. These surveys were designed to provide the Company with greater insight into the needs of our pilot participants. Questions centered on communications (mediums, frequency), messaging, understanding of the pilot (rate plans, peak events), and energy conservation (typical strategies, best practices). These findings helped inform our on-going communications strategy. Responses to these surveys have confirmed that participants feel our communications strategy (content, medium, frequency) was appropriate. Participation Pricing Pilot customers were enrolled in two phases; In Phase I participants voluntarily signed up after responding to the Company s recruiting materials. In Phase II, customers were chosen based upon a random selection in which they were required to select one of the pilot rates or indicate they wished to remain on the standard residential rate. When 8

recruitment completed a total of 4,029 participants were enrolled in the pilot. Prior to recruitment a randomized control group was selected from eligible customers. Customer attrition from both the pricing program and the control group was tracked by the Company and reported to the Commission each month. Customer reasons for leaving the pilot were broken into four categories; Customer Move, Customer Choice, Conflict, and Other. Attrition from the control group reflects customers who have moved from their premise since the initial random selection. Details concerning customer enrollment and attrition were provided to the Commission monthly through the Pricing Pilot Attrition Report. The following table summarizes the number of participants and attrition through the conclusion of the pilot on September 30, 2013. Table 2 - Pricing Pilot Participation and Attrition as of September 30, 2013 9

By the conclusion of the pilot nearly 40% of the original participant base left as a result of attrition. The overwhelming reason for participants leaving the pilot was Customer Moves. A summary of attrition reasons and what insights can be gained from them is included below: Figure 1 Pricing Pilot Reasons for Attrition Customer Move was used to track attrition related to changes in customer residence. o Customer related moves were by far the leading reason for participants leaving the pilot, accounting for approximately 79% of the attrition. During the three years of the pilot 1,266 of the 4,029 participants moved, approximately 31% or roughly 10% per year. Customer Choice indicated participants request to be un-enrolled from the pilot rate. o Attrition from Customer Choice represents those customers who specifically requested to leave the trial rate and be return to the standard residential rate. Additional insight was gained from interaction with our call center agents when customers called to request being taken off the pilot. Three primary drivers were identified; 1) customer no longer wanted to receive event notifications associated with the pilots peak event days, 2) the customer did not believe they are saving enough money, and 3) customers were not home during event window/not able to participate. Conflict with Other Program was used to track customers desire to sign up for another Xcel Energy program that was not compatible with their rate. o Conflict with other programs indicates customers who could no longer be on the pilot because they requested to enroll in another of the Company s programs that were not compatible with the pilot rates. Specifically these programs would be Saver s Switch which PTR and CPP customers could not participate in or customers signing up for Solar*Rewards which requires net metering not compatible with the pilot rates. Of the 1% overall attrition associated with this category customers moving to Solar*Rewards made up the majority. 10

Other related attrition was applied when the reason the customer left the rate varied from those listed above. o The next largest category was Other. The majority of these issues were associated with non-communicating Smart Meters. If communications could not be restored after attempts at meter repair or replacement, a standard meter would have to be installed. The participant was then removed from the pilot and placed on the standard residential rate. Other issues captured in this category were minimal and didn t fit specifically into one of the other three segments One concern around attrition is whether there would be enough participants to assure statistically valid results could be reached at the end of the pilot. Throughout the pilot this was monitored by the Company and reported back to the commission. Even with overall pilot attrition approaching 40% there were still enough participants to demonstrate results within the desired range of 20% precision at a 90% confidence interval. Table 3 Confidence Interval by Rate Tariff Group Participants Predicted Relative Precision @ 90% Confidence Interval Phase 1 TOU Pricing Only 460 17.96% CPP Pricing Only 105 17.10% PTR Pricing Only 299 11.71% Phase 2 TOU Pricing Only 889 12.57% CPP Pricing Only 252 11.04% PTR Pricing Only 1122 6.52% The ranges of relative precision factors shown in the table above for the pricing only cells are within the original design criteria and no adverse impact on the ultimate evaluation of the Pricing Pilot is indicated based on the current participation levels. 11

Budget The table below shows the Pricing Pilot budget across the three years of the program. Table 4 - Pricing Pilot Budget/Actual/Variance * Reflect actual expenditures from January 1 through November 30, 2013. Further details on the major budget components are listed below: 1. As the pilot matured and processes solidified, less program design was necessary. 2. Administrative and program delivery time was significantly reduced by using dedicated contract labor and streamlining processes and procedures. 3. The 2012 and 2013 budget reflected the Company s commitment to continued customer education and engagement. The Company was able to continually meet these objectives at reduced costs by leveraging in-house resources and relying heavily on customer preferred less expensive, electronic communications. 4. Higher than normal temperatures combined with fewer participants due to attrition resulted in lower average rebate amounts. It is of note that incentives are not captured under this deferred subledger. 5. Expenses reflect equipment, installation, and on-going maintenance expenses for participants with In Home Devices. With all device installations completed, the ongoing maintenance and support costs were reduced. 6. Both the 2011 and 2012 filings include preliminary results. With the pilot s completion in September and the full regression analysis taking place, greater spend due to the increased level of effort took place. 7. Minimal costs such as mileage and other travel/employee related were only accrued early on in the pilot program. Lessons Learned 12

Most of the lessons learned would apply with a more broad deployment to a greater participant base. Greater automation of billing and customer engagement required for full-scale deployment. Based upon the scale of the SmartGridCity Pricing Pilot, automation was not necessary. A more broad deployment of the pilot rates would drive the need for automation. Automation would drive improvements in two key areas: o Customer Engagement With current levels of automation post event reporting was only able to be supported on a monthly basis. Additional automation, specifically in preparing customer billing data following Peak Events in near-real time would support timely post-event reporting. Improving the timeliness of these reports would enable the Company to provide timely, individualized reports to include energy savings tips and tricks following each event vs. a summary of the previous month s events. o Billing In the current environment, the billing process to support the RPTR rate is time consuming due to its manual nature. Automation would reduce the time to support the billing process by an estimated two thirds. Greater automation in support of the billing processes would also aide in providing near-real time post event reporting. Customer engagement continues to prove a necessary key to program success. o Ongoing Communications Beyond the post event reporting, ongoing customer communications are important. Customer education on the rate, how the customer can best benefit from it and why the utility has deployed them was provided during recruiting. Education and engagement is necessary beyond this initial decision point. Customers preferred a combination of both email and phone as their primary critical peak pricing event notification. o Public Service initially offered email, telephone (or a combination of both) notification for critical peak pricing events, adding text (SMS) capabilities in 2012. For those with an IHSD, notifications were also received via their device. Although adoption of text (SMS) capability (or a combination of text with either phone or email) was low, it is recommended that it remain as an additional customer contact choice and preference. With a full scale deployment, the rate should reside with the customer, not the premise. o During the pilot, the rate resided with the premise. More specifically, should a participating customer move they were not able to take the rate with them. In addition, this rate was not available for the new resident. This scenario, as was highlighted in the Participation section, accounted for the majority of the program attrition. In deploying dynamic rates more broadly while providing the customer the choice of rate should include the ability to maintain that rate and level of choice should the customer change residences. To allow this would require full deployment of smart meters to accommodate any rate for any premise within our jurisdiction. 13

The CPP rate provided the greatest amount of load reduction. These results are generally consistent with other pricing pilots. o As seen on page 4 of the EnerNOC report (Page iv Results and Key Findings 1 st bullet), the CPP rate provided greater load reduction than that of the PTR or TOU rates. These savings were experienced on both Summer on peak and event days. These results are consistent with other (EnerNOC report Pages 7 Key Findings 2 nd bullet and page 23 following figure 9) similar pilots in the industry. The peak period deployed for the pilot is the correct duration and would not need to change with a full scale deployment. o As commented on within the EnerNOC report (Page 47 Figure 24) the current event period of 14:00 20:00 MT (Weekdays, excluding holidays) is correct. With a broader deployment of dynamic rates, consideration should be taken relative to the minimum/maximum number of events per year specifically in the case of a system emergency. Final Report Attached is the final report as prepared by EnerNOC the third party evaluator contracted for this project. The report includes results of the full regression analysis of the three-year pilot. 14

SmartGridCity Pricing Pilot Program Impact Evaluation Results, 2011-2013 Prepared for Xcel Energy 1800 Larimer 15 th Floor Denver, CO 80202 Project Manager André Gouin Business Technology Consultant Tel. 303.294.2975 Andre.Gouin@xcelenergy.com Prepared by EnerNOC, Inc. 500 Ygnacio Valley Road Suite 450 Walnut Creek, CA 94596 Tel. 925.482.2000 www.enernoc.com Project Manager Craig Williamson Practice Lead Program Evaluation and Load Analysis Utility Solutions Consulting Tel. 720.233.1500 cwilliamson@enernoc.com December 6, 2013

Principal Investigators A. Nguyen C. Williamson K. Parmenter K. Marrin T. Williams

Executive Summary Pricing Pilot Background Public Service Company of Colorado (Public Service) carried out a SmartGridCity Pricing Pilot (Pricing Pilot) over the three year period from October 2010 through September 2013. A primary objective of the pilot was to assess the customers willingness to modify electricity use in response to price, thereby yielding peak demand reductions and energy savings. The Pricing Pilot consisted of three types of rates: 1) Critical Peak Pricing (CPP); 2) Peak Time Rebate (PTR); and 3) Time of Use (TOU). Event days were a key aspect of the Pricing Pilot. Customers on the CPP and PTR rates were notified a day ahead of each event day. Customers were encouraged to save energy between the hours of 2 p.m. and 8 p.m. on event days to avoid paying higher energy rates (CPP customers) or to earn a bill credit (PTR customers). Pilot participants were recruited in two phases. Phase I recruitment used direct mail to get voluntary or opt-in enrollments. Phase II recruitment attempted to simulate a mandatory choice between the three time-varying pricing rates and the standard tiered rates so as to have an opt-out enrollment strategy. Because of low initial participation, Phase II recruitment was modified and ended up being very similar to opt-in. Study Objectives The EnerNOC project team supported Public Service in the design of the pilot and provided ongoing support for calling events as well as monthly post-event processing throughout the pilot. Additionally, we conducted yearly impact evaluations as well as an overall evaluation of the pilot. There were five main objectives of our work: Support Public Service during design and initiation of the Pricing Pilot, including in the development of a protocol for calling pricing events Estimate impacts for each of the pricing alternatives in each year of the pilot, including on-peak and off-peak energy savings and peak demand reductions on event days and non-event days, during both non-summer and summer periods Develop a model to estimate actual impacts achieved during the pilot and to predict potential impacts under various scenarios Discuss findings from the pilot and provide comparisons to other similar programs Offer recommendations for program implementation considerations, including modifications to improve the likelihood of success for a wider scale deployment. Analysis Methodology We estimated impacts for the pilot program using two methods: difference of differences and regression modeling. Each method allows us to study a different aspect of the pilot. Difference of Differences: The difference of differences approach gives us an approximation of savings based on direct comparison of participant and control groups during the participation period ( treatment period ) and for a time before participation EnerNOC Utility Solutions iii

started ( pretreatment period ). It allows us to calculate the difference in energy use, corrected for any preexisting differences between the participant and control groups. We used this approach to estimate savings based on what actually happened during the analysis period ( ex post impacts) Regression Modeling: Though the difference of differences approach gives us valid estimates of ex post savings, it does not allow us to see the influence of independent variables on energy impacts. Regression modeling enables us to quantify the variability from other known sources affecting energy use so that we can estimate impacts as a function of these parameters. Specially, we can see how impacts depend on weather, presence of in-home smart devices (IHSDs), etc. This capability allows us to estimate energy use for an arbitrary day and participant type based on that day s weather and how impacts change with temperature. Therefore, with the regression models, we can provide estimates of the impacts achieved during the actual pilot weather conditions as well as predict impacts that would be achieved during other weather conditions. We used the regression model to predict savings for a normal weather year by applying typical meteorological year (TMY3) weather data from the National Oceanic and Atmospheric Administration. Results and Key Findings We can draw the following conclusions by comparing the results across the rates, pilot timeframe, participant phases, and presence of enabling technologies: Comparison of Rates: The CPP rate provides more load reduction during summer onpeak hours on event days and on non-event days than both the PTR and the TOU. TOU provides consistent savings on every day, but does not provide the extra decrease on event days, since there is no event pricing. The PTR, while attractive to customers because of its no-lose rebate, does not provide as much load reduction per customer as the CPP. The PTR and the CPP both also result in off-peak energy savings. Year-Over-Year Comparison: In nearly all cases, the difference of differences ex post event day load impacts diminished over time. The decreasing impacts were partly influenced by dramatic differences in the temperatures for the three years, but also may indicate that there was a drop off in persistence that affected the savings. The regression analysis using normal weather for all three years supports this notion since it showed lower event day load reductions in the second and third years. The overall energy savings did not show as clear a trend over time, which is not surprising as energy consumption over a longer time period is more stable; however energy savings tended to increase across the years in most cases. Comparison of Phases: The results indicate differences between the Phase I customers, who were recruited on an opt-in basis, and the Phase II customers, who were recruited in a pseudo-opt-out manner. Phase I load reductions were higher than Phase II, and tend to drop off less across the three years. Care must be taken, however, with these results, since the recruitment for Phase II was not consistent and was not truly opt-out. In addition, since the Phase I customers were recruited before the iv EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results Phase II customers, Phase I would include those customers most interested in a timevarying rate who may be more inclined to respond. Influence of Saver s Switch: The TOU SS customers are higher energy users than those without SS, both in the summer and non-summer. The summer usage is higher due to the fact that the TOU SS customers all have CAC, whereas the TOU NSS include a mix of customers with and without CAC. TOU NSS participants tended to have higher on-peak load reductions than SS participants during the winter, but both NSS and SS participants had load reductions of about the same magnitude during the summer onpeak. In terms of energy savings, SS participants saved less during the on-peak overall for the year, but did not increase kwh usage by as much as NSS participants during the off-peak. Impact of IHSDs: There were not enough IHSDs installed on pricing pilot customers to provide results that could be generalized to a broader population. However, using the regression analysis, we could estimate the impacts for those customers with IHSDs, few though they were. For those with IHSDs, there were higher savings during summer events, presumably driven by the control of the thermostat setpoint by the devices. During non-summer events, those with IHSDs increased their usage in some cases. This may have been due to an assumption by the customers that the device was taking care of things for them, when during the winter, there is not any CAC load for the device to reduce. This result should be considered anecdotal, and not specifically indicative of impacts for a broader roll-out. The first two of these conclusions are illustrated in Figure ES-1 and Figure ES-2 below, which compare the ex post event day on-peak load reduction and energy savings results for the different rates across the years and seasons. For clarity, we have included only Phase I ex post results. EnerNOC Utility Solutions v

Annual kwh Savings Average On-Peak kw Reduction Figure ES-1. Event Day On-Peak kw Reduction, Comparison by Rate 0.30 Event Day On-Peak kw Reduction Summer Events Non-Summer Winter Events CPP PTR 0.25 0.20 0.15 0.27 0.28 0.10 0.20 0.23 0.18 0.14 0.05 0.10 0.09 0.04 0.05 0.00 Summer 2011 Summer 2012 Summer 2013 Winter 2012 Winter 2013 Figure ES-2. Annual Energy Savings, Comparison by Rate 800 700 Annual kwh Savings CPP PTR TOU 695 600 574 500 400 385 413 300 200 312 217 100 0-100 -200 2011 2012 2013-33 -101 2 vi EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results The CPP provides both the highest load reduction during peak periods and the most overall kwh savings. Across the three years, the load reduction decreased somewhat, but the annual kwh savings increased. The PTR rate also results in overall savings, even though there is no price signal outside of event day on-peak hours. The TOU customers change their overall energy use the least. Key Findings The following is a summary of key findings from the Public Service Pricing Pilot: Participants Responded to Price: Residential customers participating in the Pricing Pilot responded to price signals, reducing their energy consumption during on-peak periods when prices were higher. Results Generally Consistent with Other Pricing Pilots: In most cases, the relative results found here were generally consistent with other pricing pilots in the industry. The impacts tended to be less than other pricing pilots, due to the lower saturation of central air-conditioning (CAC) in Public Service Company s territory and due to the low number of in-home smart devices. But the relationship between the impacts for different rates and for different time periods was consistent with other pricing pilots. On-Peak Summer Event Load Reductions Reached 31%: Event-driven rates, PTR and CPP, provided the highest load reduction of the pilot on summer event days. CPP Phase I participants reduced loads by 22-29%, which represents average on-peak demand reductions of 0.22-0.28 kw, while PTR Phase I participants had load reductions of 8-14%, or 0.09-0.14 kw. Summer Events Had Largest Impacts: For CPP, the summer event day load reductions were higher than the summer non-event day load reductions. This finding is consistent with similar event-driven rates offered elsewhere and is due to a combination of two factors: event days are hotter than non-event days, and prices are higher on event days. The summer event day impacts were also greater than the winter event day impacts, mainly due to the lack of CAC load during the winter. TOU Participants Saved Energy During On-Peak: TOU provided load reduction during the on-peak period on summer weekdays. Load reductions for Phase I participants averaged 5-9% across the three years of the pilot, representing average onpeak demand reductions of 0.05-0.12 kw. Off-peak energy use increased by a smaller percentage in most cases, as expected due to lower off-peak rates, and as seen in most TOU rates across the industry. CPP and PTR Participants Saved Energy at all Times: In general, the CPP and PTR participants saved energy at all times, not just during on-peak periods. This result is consistent with other findings in the industry for event-driven rates, as these rates usually lead to an increased awareness of overall energy use. When customers are more aware of their energy use, they use less energy. EnerNOC Utility Solutions vii

Non-Summer Impacts were Small: The impacts during the non-summer season, for both events and non-events, were consistently much lower than the summer impacts in almost all cases. This difference is due to lower loads and less discretionary energy use during the winter, spring, and fall. Impacts Were Consistent Through Duration of Event: The magnitude of savings was fairly consistent throughout events for CPP and PTR participants. This consistency is typical of pricing programs that are not primarily driven by an enabling technology, which is the case for nearly all customers in the pilot. When customer behavior is driving the load reduction, the actions taken tend to result in more consistent savings across the event. Performance of Phase I Tended to Exceed Phase II: The results indicated differences between the Phase I customers, who were recruited on an opt-in basis, and the Phase II customers, who were recruited in a pseudo-opt-out manner. Phase I load reductions were higher than Phase II, and tend to drop off less across the three years. Care must be taken, however, with these results, since the recruitment for Phase II was not consistent and was not truly opt-out. In addition, since the Phase I customers were recruited before the Phase II customers, Phase I would include those customers most interested in a time-varying rate who may be more inclined to respond. PTR, CPP, and TOU Customers Reduce Load during Events less than Saver s Switch Customers: Xcel s current Saver s Switch (SS) program yields a higher load reduction per customer than any of the time-varying rates in the pilot, at about 1.07 kw per participant during events, against a baseline of 3.00 kw for the average SS customer. 1 However, unlike the Pilot customers, which include a mix of those with and without central air-conditioning (CAC), SS customers all have CAC. Therefore, SS customers have dramatically higher baseline loads and have more discretionary load available. PTR Baseline Methodology Could be Improved: The PTR baseline used to estimate savings for the payment of the rebate resulted in biased savings estimates. It underestimated savings on hot event days and overestimated the savings on mild event days. A weather-adjusted PTR baseline based on the difference between the average temperature on the days used in the baseline and the event day temperature would provide a better estimate of savings and a more accurate rebate payment. Load Impacts Lessen over the Three Years: In nearly all cases, the load impacts for the second and third years were less than the load impacts for the first year. This may indicate that there is a drop off in persistence, but there were also dramatic differences in the temperatures for the three summers, which also influenced the savings. However, the regression analysis, using normal weather for all three years, showed lower savings in the second and third years. This drop in savings has not generally been seen in other studies in the industry, and may be due to a decrease in level of participant engagement. 1 http://www.xcelenergy.com/staticfiles/xe/marketing/files/co-dsm-2012-2013-biennial-plan-rev.pdf, page 284. viii EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results Devices Provide Anecdotal Results: There were not enough in-home smart devices (IHSDs) installed on pricing pilot customers to provide results that could be generalized to a broader population. However, using the regression analysis, we could estimate the impacts for those customers with IHSDs, few though they were. The results are reflective only of the individual customers who had IHSDs. For those with IHSDs, there were higher savings during summer events, presumably driven by the control of the thermostat setpoint by the devices. This result should be considered anecdotal, and not indicative of impacts for a broader roll-out. The Current Event Structure is Appropriate: The current event window of 2:00 to 8:00 in the afternoon is appropriate, and the correct duration. The restriction of calling 13-15 events is reasonable, but should be modified, if possible, to allow for additional events in the case of a system emergency. Customers Understand Price Signals: Based on the relative savings in the various periods, customers appeared to understand the price signals. The only exception is that CPP and PTR customers used less in off-peak periods, where lower prices would normally result in increased energy use. As explained above, this is likely due to increased awareness. Recommendations We have the following recommendations regarding the TOU Pilot if Public Service continues offering time varying rates. Keep Event Structure: We recommend that Public Service keep the current on-peak and event structure, with events and on-peak periods running from 2:00-8:00. 15 events seems an appropriate number, but if possible, we recommend a caveat on the maximum to be able to call additional events beyond 15 in case of a system emergency. Including a minimum number of events ensures that events are called, gives customers a chance to realize savings, minimizes free-ridership by some customers, and allows for estimation of impacts. Analyze Payout for New PTR Rate: Public Service should analyze the payout for random variation in the rate design process for a new PTR rate, and recover that cost from customers who are on the rate. Offer only Opt-In PTR: We recommend offering the PTR rate on a strictly opt-in basis, preferably combined with some sort of an enabling control technology such as a PCT. Improve PTR Baseline: We strongly recommend weather adjusting the PTR baseline based on the difference between the average temperature on the days used in the baseline and the event day. Continue Customer Engagement: We recommend continuing and enhancing efforts to engage customers who are on the rates, providing them feedback on their energy use changes (the same concept as the post-event reporting), and suggestions about how to EnerNOC Utility Solutions ix

benefit from the rate. The feedback must be credible and specific to the customer in order for it to be effective. Target Customers with Air Conditioning: If possible under current regulation, Public Service should consider offering time-varying rates only to those customers with air conditioning. x EnerNOC Utility Solutions

Table of Contents Chapter 1 Introduction... 1 Background... 1 Scope... 1 Report Organization... 1 Chapter 2 Overview of Pilot Program... 3 Pricing Pilot Description... 3 Target Market... 3 Pricing Options... 3 In-Home Smart Devices... 4 Events... 4 Participation... 5 Recruitment... 5 Engagement... 6 Attrition... 6 Chapter 3 Analysis Methodology... 9 Overall Analysis Approach... 9 Difference of Differences... 9 Methodology... 9 Control Groups...11 Considerations with the PTR Rate...12 Considerations with the TOU Rate...12 Regression Modeling...12 Methodology...12 Fixed Effect Model...14 Data Used in Analysis...15 Event Days...15 Estimating Impacts...15 Chapter 4 Impact Results...19 Organization of Results...19 Difference of Differences Analysis: Ex Post Impacts...19 CPP Results...20 EnerNOC Utility Solutions xi

PTR Results...26 TOU Results...30 Regression Analysis: Weather-Normalized Impacts...37 CPP Results...37 PTR Results...40 TOU Results...42 Discussion of Cross-Cutting Results...43 Chapter 5 Additional Analysis...47 Peak Event Duration Analysis...47 Utility Perspective of Event Duration...47 Customer Perspective of Event Duration...48 Number of Event Days...48 PTR Baseline Analysis...50 Opt-In Versus Opt-Out...54 Comparisons with Saver s Switch Customers...55 Program Comparison...56 Program Limitations...57 Chapter 6 Key Findings and Recommendations...59 Key Findings...59 Recommendations...61 Appendix A 2011 Difference of Differences Results... A-1 Impact Results... A-1 Load Profiles... A-5 Phase I Non-Summer Analysis... A-5 Phase I Summer Analysis... A-7 Phase II Summer Analysis... A-10 Appendix B 2012 Difference of Differences Results... B-1 Impact Results... B-1 Load Profiles... B-5 Phase I Non-Summer Analysis... B-6 Phase I Summer Analysis... B-9 Phase II Non-Summer Analysis... B-12 Phase II Summer Analysis... B-15 xii EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results Appendix C 2013 Difference of Differences Results... C-1 Impact Results... C-1 Load Profiles... C-4 Phase I Non-Summer... C-5 Phase I Summer... C-8 Phase II Non-Summer... C-12 Phase II Summer... C-15 Summary of 2013 Findings... C-18 EnerNOC Utility Solutions xiii

List of Figures Figure 1. Difference of Differences Approach... 10 Figure 2. Simplified Regression Modeling Approach... 14 Figure 3. Data Parameters Included in Difference of Differences Analysis: CPP & PTR Rates. 17 Figure 4. Data Parameters Included in Difference of Differences Analysis: TOU Rate... 18 Figure 5. CPP Ex Post Results: Event Day On-Peak Demand Impacts... 21 Figure 6. CPP Ex Post Results: Scatter Plot of Event Day On-Peak Demand Impacts... 22 Figure 7. CPP Ex Post Results: Non-Event Weekday On-Peak Demand Impacts... 23 Figure 8. CPP Ex Post Results: Energy Savings... 25 Figure 9. PTR Ex Post Results: Event Day On-Peak Demand Impacts... 27 Figure 10. PTR Ex Post Results: Scatter Plot of Event Day On-Peak Demand Impacts... 28 Figure 11. PTR Ex Post Results: Energy Savings... 29 Figure 12. TOU Ex Post Results: Weekday On-Peak Demand Impacts, with Saver s Switch... 32 Figure 13. TOU Ex Post Results: Weekday On-Peak Demand Impacts, without Saver s Switch... 33 Figure 14. TOU Ex Post Results: Energy Savings, with Saver s Switch... 34 Figure 15. TOU Ex Post Results: Energy Savings, without Saver s Switch... 35 Figure 16. CPP Weather-Normalized Results: Event Day On-Peak Demand Impacts, without IHSD... 38 Figure 17. CPP Weather-Normalized Results: Non-Event Weekday On-Peak Demand Impacts, without IHSD... 40 Figure 18. PTR Weather-Normalized Results: Event Day On-Peak Demand Impacts, without IHSD... 41 Figure 19. TOU Weather-Normalized Results: Weekday On-Peak Demand Impacts, without IHSD... 43 Figure 20. Event Day On-Peak kw Reduction, Comparison by Rate... 45 Figure 21. Annual Energy Savings, Comparison by Rate... 45 Figure 22. System Load on Annual System Peak Days for 2011, 2012, and 2013... 47 Figure 23. System Load Duration Curve Based on Daily Peaks... 49 Figure 24. Impact of Temperature on Energy Use... 50 Figure A1. Phase I, 2011, Non-Summer, Non-Event Weekday, CPP... A-6 Figure A2. Phase I, 2011, Non-Summer, Non-Event Weekday, PTR... A-6 Figure A3. Phase I, 2011, Non-Summer, Weekday, TOU... A-7 Figure A4. Phase I, 2011, Summer, Non-Event Weekday, CPP... A-8 Figure A5. Phase I, 2011, Summer, Non-Event Weekday, PTR... A-8 Figure A6. Phase I, 2011, Summer, Weekday, TOU... A-9 Figure A7. Phase I, 2011, Summer, Event Day, CPP... A-9 Figure A8. Phase I, 2011, Summer, Event Day, PTR... A-10 Figure A9. Phase II, 2011, Summer, Non-Event Weekday, CPP... A-11 Figure A10. Phase II, 2011, Summer, Non-Event Weekday, PTR... A-11 Figure A11. Phase II, 2011, Summer, Weekday, TOU... A-12 Figure A12. Phase II, 2011, Summer, Event Day, CPP... A-12 EnerNOC Utility Solutions xv

Figure A13. Phase II, 2011, Summer, Event Day, PTR... A-13 Figure B1. Phase I, 2012, Non-Summer, Non-Event Weekday, CPP... B-7 Figure B2. Phase I, 2012, Non-Summer, Non-Event Weekday, PTR... B-7 Figure B3. Phase I, 2012, Non-Summer, Weekday, TOU... B-8 Figure B4. Phase I, 2012, Non-Summer, Event Day, CPP... B-8 Figure B5. Phase I, 2012, Non-Summer, Event Day, PTR... B-9 Figure B6. Phase I, 2012, Summer, Non-Event Weekday, CPP... B-10 Figure B7. Phase I, 2012, Summer, Non-Event Weekday, PTR... B-10 Figure B8. Phase I, 2012, Summer, Weekday, TOU... B-11 Figure B9. Phase I, 2012, Summer, Event Day, CPP... B-11 Figure B10. Phase I, 2012, Summer, Event Day, PTR... B-12 Figure B11. Phase II, 2012, Non-Summer, Non-Event Weekday, CPP... B-13 Figure B12. Phase II, 2012, Non-Summer, Non-Event Weekday, PTR... B-13 Figure B13. Phase II, 2012, Non-Summer, Weekday, TOU... B-14 Figure B14. Phase II, 2012, Non-Summer, Event Day, CPP... B-14 Figure B15. Phase II, 2012, Non-Summer, Event Day, PTR... B-15 Figure B16. Phase II, 2012, Summer, Non-Event Weekday, CPP... B-16 Figure B17. Phase II, 2012, Summer, Non-Event Weekday, PTR... B-16 Figure B18. Phase II, 2012, Summer, Weekday, TOU... B-17 Figure B19. Phase II, 2012, Summer, Event Day, CPP... B-17 Figure B20. Phase II, 2012, Summer, Event Day, PTR... B-18 Figure C1. Phase I, 2013, Non-Summer, Non-Event Weekday, CPP... C-6 Figure C2. Phase I, 2013, Non-Summer, Non-Event Weekday, PTR... C-6 Figure C3. Phase I, 2013, Non-Summer, Weekday, TOU... C-7 Figure C4. Phase I, 2013, Non-Summer, Event Day, CPP... C-7 Figure C5. Phase I, 2013, Non-Summer, Event Day, PTR... C-8 Figure C6. Phase I, 2013, Summer, Non-Event Weekday, CPP... C-9 Figure C7. Phase I, 2013, Summer, Non-Event Weekday, PTR... C-10 Figure C8. Phase I, 2013, Summer, Weekday, TOU... C-10 Figure C9. Phase I, 2013, Summer, Event Day, CPP... C-11 Figure C10. Phase I, 2013, Summer, Event Day, PTR... C-11 Figure C11. Phase II, 2013, Non-Summer, Non-Event Weekday, CPP... C-12 Figure C12. Phase II, 2013, Non-Summer, Non-Event Weekday, PTR... C-13 Figure C13. Phase II, 2013, Non-Summer, Weekday, TOU... C-13 Figure C14. Phase II, 2013, Non-Summer, Event Day, CPP... C-14 Figure C15. Phase II, 2013, Non-Summer, Event Day, PTR... C-14 Figure C16. Phase II, 2013, Summer, Non-Event Weekday, CPP... C-16 Figure C17. Phase II, 2013, Summer, Non-Event Weekday, PTR... C-16 Figure C18. Phase II, 2013, Summer, Weekday, TOU... C-17 Figure C19. Phase II, 2013, Summer, Event Day, CPP... C-17 Figure C20. Phase II, 2013, Summer, Event Day, PTR... C-18 xvi EnerNOC Utility Solutions

List of Tables Table 1. Definition of Analysis Years... 1 Table 2. SmartGridCity Pricing Pilot Rate Options... 3 Table 3. Pricing Pilot IHSD Installations... 4 Table 4. Pricing Pilot Event Days and Corresponding Summer High Temperature by Analysis Year... 5 Table 5. Pricing Pilot Participation and Attrition as of Pilot Completion... 7 Table 6. Cooling and Heating Degree Days by Analysis Year... 20 Table 7. CPP Ex Post Results: Event Day On-Peak Demand Impacts... 20 Table 8. CPP Ex Post Results: Non-Event Weekday On-Peak Demand Impacts... 23 Table 9. CPP Ex Post Results: Changes in Energy Consumption... 24 Table 10. PTR Ex Post Results: Event Day On-Peak Demand Impacts... 26 Table 11. PTR Ex Post Results: Changes in Energy Consumption... 29 Table 12. TOU Ex Post Results: Weekday On-Peak Demand Impacts... 31 Table 13. TOU Ex Post Results: Changes in Energy Consumption, with Saver s Switch... 34 Table 14. TOU Ex Post Results: Changes in Energy Consumption, without Saver s Switch... 35 Table 15. CPP Weather-Normalized Results: Event Day On-Peak Demand Impacts... 38 Table 16. CPP Weather-Normalized Results: Non-Event Weekday On-Peak Demand Impacts... 39 Table 17. PTR Weather-Normalized Results: Event Day On-Peak Demand Impacts... 41 Table 18. TOU Weather-Normalized Results: Weekday On-Peak Demand Impacts... 42 Table 19. Comparison of Event Day Temperatures with Baseline Average Temperature... 53 Table A1. kw Reduction by Rate, 2011, Phase I... A-2 Table A2. kw Reduction by Rate, 2011, Phase II... A-3 Table A3. Annual kwh Savings by Rate, 2011, Phase I... A-4 Table B1. kw Reduction by Rate, 2012, Phase I... B-2 Table B2. kw Reduction by Rate, 2012, Phase II... B-3 Table B3. Annual kwh Savings by Rate, 2012, Phase I... B-4 Table B4. Annual kwh Savings by Rate, 2012, Phase II... B-5 Table C1. kw Reduction by Rate, 2013, Phase I... C-1 Table C2. kw Reduction by Rate, 2013, Phase II... C-2 Table C3. Annual kwh Savings by Rate, 2013, Phase I... C-3 Table C4. Annual kwh Savings by Rate, 2013, Phase II... C-4 EnerNOC Utility Solutions xvii

Chapter 1 Introduction Background Public Service Company of Colorado (Public Service) carried out a SmartGridCity Pricing Pilot (Pricing Pilot) over the three year period from October 2010 through September 2013. The Pricing Pilot consisted of three types of rates: 1) Critical Peak Pricing (CPP); 2) Peak Time Rebate (PTR); and 3) Time of Use (TOU). A primary objective of the pilot was to assess the customers willingness to modify electricity use in response to price, thereby yielding peak demand reductions and energy savings. The EnerNOC project team supported Public Service in the design of the pilot and provided on-going support for calling event as well as monthly postevent processing throughout the pilot. Additionally, we conducted yearly impact evaluations using the difference of differences analysis approach and completed a regression analysis based on all three years of the pilot to model program impacts. Scope In this report, we describe the pilot program, explain our analysis methodology, present detailed impact results for the three years of analysis, and discuss findings and recommendations from the pilot in comparison with best practices observed in similar programs nationwide. Our evaluation spans three analysis years, which we refer to as 2011, 2012, and 2013, based on the periods used for the interim analysis, which also served to keep the non-summer periods together as consecutive periods. Table 1 shows how the analysis years relate to calendar months. Table 1. Definition of Analysis Years Analysis Year Calendar Time 2011 October 1, 2010 to September 30, 2011 2012 October 1, 2011 to September 30, 2012 2013 October 1, 2012 to September 30, 2013 Report Organization The report is organized as follows: Chapter 2 summarizes the key features of the pilot to provide context for the impact evaluation. Chapter 3 describes the analysis methodology, including the approaches we followed for the difference of differences analysis and the regression modeling. Chapter 4 presents results from the impact analysis for each rate across the life of the pilot. The appendices include supplemental graphs and information. Chapter 5 discusses additional analysis we completed related to peak event duration, PTR baseline, comparison with other programs, and program limitations. Chapter 6 summarizes key findings from our analysis and provides recommendations for program modifications to improve the likelihood of success for a wider scale deployment. EnerNOC Utility Solutions 1

Chapter 2 Overview of Pilot Program Public Service carried out the Pricing Pilot over the three year period from October 2010 through September 2013. The discussion below summarizes key characteristics of the pilot. Pricing Pilot Description Defining aspects of the Pricing Pilot pertain to its target market, pricing options, use of enabling technologies for a subset of customers, and peak event notifications. Target Market The target market for the Pricing Pilot was Public Service s residential customers with smart meters in Boulder, CO. Phase I and Phase II targeted slightly different populations. Phase I targeted all customers with smart meters; although in a few cases, customers in the Smart Grid footprint that did not already have smart meters volunteered and subsequently had a smart meter installed and were allowed onto the pilot. The target population for the Phase II customers was restricted to a subset of the smart meter population. The recruiting section below describes the recruiting process and reasoning for the different phases in more detail. Pricing Options Pilot participants were offered the choice of three alternatives to the standard tiered rate. Table 2 compares the rates for each pricing plan. The names we use in this report for each of the rate alternatives differ from the plan names used in the customer-facing marketing material. From hereafter, we refer to the three pricing alternatives as time-of-use (TOU), critical peak pricing (CPP), and peak time rebate (PTR). Table 2. SmartGridCity Pricing Pilot Rate Options Rate Type Time of Use (TOU) Critical Peak Pricing (CPP) Peak Time Rebate (PTR) Standard Plan Name Shift & Save Peak Plus Plan Reduce- Your- Use Rebate Tiered Rate Off- Peak Price a On-Peak Price b 4 Non- Summer: 6 Summer: 17 4 Non- Summer: 5 Summer: 12 Peak Energy Event Price c None Non- Summer: 33 Summer: 51 Summer Price All prices are per kwh a Midnight to 2 p.m. and 8 p.m. to midnight weekdays, weekends, and holidays b 2 p.m. to 8 p.m. weekdays c 2 p.m. to 8 p.m. up to 15 weekdays per year 0-500 kwh: 5 500+ kwh: 9 0-500 kwh: 5 500+ kwh: 9 Non- Summer Price Rebate Price 5 Non-Summer Rebate: 29 Summer Rebate: 47 5 None EnerNOC Utility Solutions 3

In-Home Smart Devices One of the objectives of the Pricing Pilot was to determine whether or not the availability of inhome smart devices (IHSDs) impacted a participant s savings. We included consideration of IHSDs in the regression analysis to assess how these participants performed compared to the larger group. Public Service encountered some challenges to the successful deployment of IHSDs, resulting in lower than hoped for installation numbers. As of completion of the pilot, 62 IHSDs remained in the homes of Pricing Pilot participants (see Table 3). The small number of customers with IHSDs renders the results less representative of what might happen given a broader roll-out, but they do provide some useful insights. Public Service also implemented a separate much larger IHSD pilot involving customers on the standard tiered rate, but because that did not involve pricing customers, the results are outside the scope of this report. Table 3. Pricing Pilot IHSD Installations IHSD Deployment TOU CPP PTR Total Phase I 22 4 8 34 Phase II 18 3 7 28 Total 62 Events Event days were a key aspect of the Pricing Pilot. Customers on the CPP and PTR rates were notified a day ahead of each event day. Customers were encouraged to save energy between the hours of 2 p.m. and 8 p.m. on event days to avoid paying higher energy rates (CPP customers) or to earn a bill credit (PTR customers). The tariffs for these rates mandated that there would be at least 13 events and no more than 15 events in each year. Table 4 lists event days called during the analysis period for each analysis year of the pilot. Customers were notified of events by email, SMS (text), or telephone. Those with IHSDs were also notified via the devices. 4 EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results Table 4. Pricing Pilot Event Days and Corresponding Summer High Temperature by Analysis Year 2011 2012 2013 Event Date High T ( F) Event Date High T ( F) Event Date High T ( F) 7/14/2011 88 10/4/2011 11/2/2012 7/19/2011 95 11/1/2011 1/29/2013 7/22/2011 96 12/1/2011 2/21/2013 8/8/2011 92 5/18/2012 6/10/2013 97 8/10/2011 88 6/18/2012 100 6/12/2013 95 8/16/2011 90 6/22/2012 102 6/13/2013 96 8/22/2011 97 6/25/2012 105 6/19/2013 95 8/23/2011 98 6/27/2012 97 6/21/2013 87 8/25/2011 99 7/2/2012 101 6/27/2013 98 8/26/2011 92 7/13/2012 96 7/16/2013 84 9/1/2011 96 7/20/2012 101 7/19/2013 81 9/8/2011 76 7/23/2012 100 7/22/2013 97 8/1/2012 94 8/1/2013 93 8/8/2012 96 8/19/2013 96 9/14/2012 77 8/21/2013 87 Participation Participation features of the Pricing Pilot relate to recruitment methods, customer engagement, and customer attrition over time. Recruitment Pilot participants were recruited in two phases. For Phase I, voluntary participants were recruited between October and December 2010 using direct mail. The purpose of Phase I was to study the impacts of opt-in enrollment. For Phase II, the recruitment process took place from January through June 2011. The purpose of Phase II was to study the impacts of opt-out enrollment as compared with the Phase I opt-in enrollment. Ultimately, Phase II recruitment ended up being only pseudo opt-out, as described below. For Phase II, we assigned all eligible customers a random number. We then selected a random control group from these eligible customers, and these were excluded from the Phase II recruitment. We also randomly split the remaining customers into three waves for recruitment into the pilot. An attempt was made to simulate a mandatory choice between the three timevarying pricing rates and the standard tiered rates. Unfortunately, initial participation was low, so to increase participation, enrollment was opened to all three waves concurrently. However, opening enrollment to customers outside of the initial randomly selected wave meant Phase II was not truly a random sample; this affected the original 2011 analysis, as we explain further in Chapter 3. We made one additional assumption during the Phase II recruiting. Because the Phase I customers had already volunteered to be on the rate, we assumed that all Phase I EnerNOC Utility Solutions 5

customers, if they received the Phase II recruitment letter, would sign up for the same rate that they already chose when they volunteered for Phase I. Because of this, all Phase I customers that were not originally assigned to the Phase II control group were also considered Phase II customers. This means that among participants, there were some customers that were included as part of both phases, as well as some that were only in Phase I or only in Phase II. Table 5 in the Attrition section below lists the numbers in each of these groups. Engagement Public Service used a variety of communication strategies to engage customers. Examples are as follows: Informational Letters: Letters were sent to pilot participants prior to the event season to remind customers of their participation in the pilot and to review what occurs during peak energy events. Additional communications provided best practices and peer experiences on ways to save energy and money. Post-Event Reporting: After the 2011 event season, Public Service provided participants with a year-end summary of the pilot s achievements in aggregate. For the 2012 and 2013 analysis years, Public Service provided customer-specific monthly reports summarizing the pilot community s achievement for the past month and the customer s individual achievements, as well as suggestions on how to shift and save energy. Online Panel: Public Service recruited an online panel to collect participant feedback through a series of four online surveys. The primary purpose was to inform the communication strategy carried out during the pilot. Attrition Customer attrition occurred throughout the life of the pilot program. The most notable reasons for customers leaving the pilot include: 1) customers moving to a new residence; and 2) customers simply changing their minds and deciding not to participate. Public Service tracked attrition from both the Pricing Program participants and the control group each month. Table 5 shows the number of pilot participants by Phase and Rate as of the end of the pilot. (Note: attrition from the control group is based solely on customer moves, to ensure consistency between participant and control groups). 6 EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results Table 5. Pricing Pilot Participation and Attrition as of Pilot Completion PARTICIPANTS AS OF: CHANGE FROM: 7/1/2011 9/15/2013 7/1/2011 Phase I Total 1,458 890 568 Phase I Assigned to Phase II 1,238 753 485 Incremental Phase II 2,571 1,536 1,035 Total Phase II 3,809 2,289 1,520 Total Discreet PP Participants 4,029 2,426 1,603 Phase I Control Group 1,292 890 402 Phase II Control Group 3,543 2,035 1,508 PARTICIPATION BY TARIFF Phase I Tariff Date Phase I Assigned to Phase II CPP TOU PTR Incremental Phase II Total Phase II Total Change 7/1/2011 223 188 320 508 543 9/15/2013 108 90 165 255 273 Attrition 115 98 155 253 270 7/1/2011 767 647 831 1,478 1,598 9/15/2013 477 404 503 907 980 Attrition 290 243 328 571 618 7/1/2011 468 403 1,420 1,823 1,888 9/15/2013 305 259 868 1,127 1,173 Attrition 163 144 552 696 715 Overall Attrition 568 485 1,035 1,520 1,603 Reasons for Attrition SINCE JULY 1, 2011 Phase I Phase II Customer Move 437 829 Customer Choice 56 68 Conflict with Other Program 4 11 Other 71 127 Total 568 1,035 EnerNOC Utility Solutions 7

Chapter 3 Analysis Methodology Overall Analysis Approach The analysis covers three years of the Pricing Pilot, spanning October 1, 2010 through September 30, 2013. Though not required by commission order, 2 Public Service included preliminary results for the first two years of the pilot in December filings to the Commission. The December 2011 filing covered the period October 1, 2010 through September 30, 2011 for Phase I and June 1, 2011 through September 30, 2011 for Phase II. We refer to this period as analysis year 2011. The December 2012 filing covered the period October 1, 2011 through September 30, 2012 for both Phase I and Phase II, and we refer to this period as analysis year 2012. This report contains the results for the final year, October 1, 2012 to September 30, 2013, analysis year 2013, as well as the complete results for the entire pilot. The EnerNOC team estimated the impacts of the Pricing Pilot using two methods: 1) differencing approach ( difference of differences ), and 2) regression modeling approach. Each method allows us to study a different aspect of the pilot. We used the difference of differences approach to estimate the actual impacts for each year of the Pricing Pilot; these are called ex post impacts. We then used the regression modeling approach to integrate all the pilot data together to estimate how different factors influence impacts, thereby allowing us not only to estimate actual impacts achieved during the pilot, but to also predict potential impacts under various scenarios. We focused our regression analysis on estimating weather-normalized impacts and on studying the influence of IHSDs on savings. In essence, the difference of differences method allows us to study what happened in the past, while the regression modeling approach allows us to look at potential future scenarios. Difference of Differences Methodology We used the difference of differences approach to estimate impacts achieved by rate for each year of the pilot. This method compares load shapes of participating customers with a control group of similar, but non-participating customers, both during the participation period ( treatment period ) and for a time before participation started ( pretreatment period ). Comparison during the treatment period gives an unadjusted estimate of the impacts. This estimate is then corrected using the difference during the pretreatment period to adjust for any preexisting differences between the participant and control groups. Therefore, the difference of differences method provides a robust savings estimation that is particularly useful for situations where there may be preexisting differences between the participants and the customers in the control group. The difference of differences method consists of the following six steps (Figure 1 illustrates the approach): 1. Start with 15 minute interval data for the treatment and pretreatment periods for participating customers and a control group. 2 Stipulation and Settlement Agreement approved in Decision No. C10-0491 in Docket No. 09A-796E. EnerNOC Utility Solutions 9

2. Divide the analysis year into day types. Day types include CPP and PTR event days, an average non-event weekday and an average weekend day by month, with holidays included as weekend days. Create average day types for each customer in the study. 3. Calculate the average load shape for the participant and control groups for each rate and day type. 4. Calculate the difference between the control group average load and the participant group average load for each day type, across the analysis year (treatment period) and for the year before the participants went on the rate (pretreatment period). The result of the difference during the analysis (treatment) period is the first difference, which represents the unadjusted impact. 5. Subtract the pre-participation (pretreatment) difference for each day type from the unadjusted impact to get the adjusted or corrected impact. 6. This second difference represents the estimated savings impacts corrected for the preparticipation differences between the two groups. Figure 1. Difference of Differences Approach Input is interval data Divide analysis year into day types Create average load shapes for participant and control groups Calculate difference between control and participant average loads Subtract preparticipation difference Output is preliminary estimate of savings impacts Equation 1 shows a simplified form of the mathematical calculations used in the difference of differences analysis to estimate energy savings for each day type. Where Savings = (Cntl after Tx after ) (Cntl before Tx before ) (1) Cntl after is the average control group customer energy use in the treatment (after) period Tx after is the average participant group (also referred to as the treatment group) customer energy use in the treatment (after) period 10 EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results Cntl before is the average control group customer energy use in the pre-treatment (before) period Tx before is the average participant group customer energy use in the pre-treatment (before) period This formula can easily be rewritten as shown in Equation 2, which allows for comparison of the actual participant group load in the analysis period with an adjusted treatment period control group. This is the way data is displayed on the load profile graphs in the appendices. Visually, this depicts the savings as the difference between the two lines, in comparison with the actual participant group load shape. Savings = [Cntl after (Cntl before Tx before )] Tx after (2) The term in the square brackets is the adjusted control group load and the final term is the actual participant group load during the analysis period. Control Groups Initially, the control groups for the two phases of the Pricing Pilot were selected differently. For Phase I, we selected the control group after participants enrolled by matching the participant customers with a similar customer from the larger pool of non-participating customers with smart meters, based on geography and similarity of seasonal on-peak and off-peak energy. By matching the individual customers, the control group could be split into rate-specific control groups, pairing each participant rate group with a control group that was similar to the customers that selected that rate. For the initial Phase II control group, we randomly assigned customers to the control group before recruiting began. If everyone in the participant group that was also randomly assigned were recruited, this would result in a very well-matched control group. However, since recruitment numbers were far below what was anticipated, and all customers not in the randomly assigned control group were eventually targeted for recruitment, there was a significant difference between the control group and the participants. This difference introduced what is known as self-selection bias. The differences between the participant and control groups represent both the behavior change resulting from the rate and the inherent differences between the people who are likely to sign up for the rate and those who are not. If the differences are not too dramatic, the difference of differences can account for this since the first difference in the pre-participation period should account for the preexisting differences. However, with something as complex as a load shape, if the difference in shape is too extreme, the results may not be reasonable or stable. In addition, the choice of which rate to sign up for was made by the participant, so there is no way to divide the randomized control group into subgroups that match each of the rate-specific participant groups. In our 2011 analysis for the first pilot year, we used the randomly assigned control group for the Phase II analysis and the matched control group for the Phase I analysis. However, in the two subsequent years of the pilot, we have used a matched control group for both Phases I and II to provide a more accurate estimation of energy savings. In this report, we also include revised difference of differences impacts results for 2011 that are based on the matched control group in the place of the randomly assigned control group previously used. EnerNOC Utility Solutions 11

Considerations with the PTR Rate One important point to clarify is how we calculated savings for the PTR rate for this analysis. For billing purposes, a customer s energy savings is calculated by comparing actual usage during a pricing event to a customer-specific baseline usage calculated for each event day. The rebate is the difference between that baseline and the customer s actual usage, as long as the difference is positive. This is appropriate and correct for customer billing. However, to estimate the impact PTR participants have on the electric system, it is important to use a method that more accurately estimates the impacts and is consistent with the method used for the other rates. This differs from the calculation of the PTR credit because it is not one-sided (i.e., it considers both positive and negative differences), and also because it is based on a comparison with a control group load on event days, not with a customer s own load on non-event days. Considerations with the TOU Rate TOU participants include customers with and without Saver s Switch (SS). 3 For this reason, we analyzed TOU participants as two groups, those with SS and those without. This distinction is most important on SS event days, but it is also important on other days, since the load shapes and possibly the behavior of those with and without SS are different both before and after they go on the TOU rate. One additional effect this has is that, while customers with and without Central Air-Conditioning (CAC) were eligible for all the rates, the TOU subgroup with SS was made up of only CAC customers, whereas the other groups were a mixture of those with and without CAC. As a result, the loads for the TOU with SS groups are significantly higher during the summer, when there is more CAC load. Regression Modeling Methodology In the regression analysis, we used statistical models to estimate how different factors (independent variables) influence energy use (dependent variable) for each rate. In this case, we have three dependent energy use variables, and therefore three regression models, for each rate: 1) on-peak weekday; 2) off-peak weekday; and 3) weekend. The independent variables investigated are as follows: Pricing events Weather (cooling degree days and heating degree days) Pilot program year to assess any changes in customer response and in general energy use across the pilot period Season Presence of in-home smart device (IHSD) 3 SS customers were not allowed on the CPP and PTR rates, unless they left the SS program before enrolling on one of these rates. 12 EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results SS events Each model looks at the dependent variable as a function of the other independent variables and estimates the coefficients of the dependent variables in that function. Because there was such a large overlap between the Phase I and Phase II customers, the phases were pooled and results reflect the impacts for all customers. Given the similarity of the actual recruiting that was done, it is reasonable to treat them consistently. Our regression analysis included the following steps and assumptions for each rate (Figure 2 illustrates the approach): Use all participant and control customers in a fixed effect model by rate. Run models for the combined group of Phase I and Phase II customers and identify changes in response by year. For the on-peak model, calculate the on-peak energy for each customer by summing the energy for each interval during the on-peak period for each day. Similarly, for the offpeak weekday model, calculate the off-peak energy by summing the energy for each interval during the off-peak period for each weekday. Do the same for the weekend model, during which all hours in the day are off-peak. Create variables and indicators in the database. Note: The independent variables investigated include some that are related to participation, and others that are not. Conceptually, information not related to participation goes into the model to estimate the baseline energy use based on all customers (including participants and control groups), while the participation variables (in some cases interacted with weather data) estimate the program impacts. We used the number of degree days for each day in the model, both with and without participation, so that the model quantifies the relationship between energy use and temperature as well as the relationship between savings and temperature. Test all the individual variables for statistical significance, and include only variables that actually influence energy use significantly. To make the savings estimates more stable and consistent, endeavor to keep the structure of the model consistent across variables related to participation and savings. Based on this analysis, select the most appropriate model for each dependent variable and each rate. After selection of the models, estimate the savings impacts for the actual analysis period and for different scenarios using the model coefficients. Sum the results across seasons and day types to get the total energy impacts. EnerNOC Utility Solutions 13

Figure 2. Simplified Regression Modeling Approach Input includes interval data, weather data, participant features Use fixed effect model Sum energy for each interval during the period in question Create variables and indicators in the database Test for statistical significance Output is savings impacts for different scenarios Fixed Effect Model We used a fixed effect model in the regression analysis. A fixed effect model introduces indicator variables for each participant to capture and control for unobservable customerspecific effects. Equation 3 is a somewhat simplified version of the type of model specification we used. (3) Where the variables and their coefficients are defined as: = consumption of customer i in period j (on-peak or off-peak) on day t = a fixed effect for each customer i = a vector of seasonal indicator variables, i.e., month, year, and day of week = several different weather related variables and interactions between weather and seasonal indicator variables; CDD = cooling degree days = an indicator variable that takes on a value of one after customer i enrolls in the program = an indicator variable that takes on a value of one after customer i enrolls in the program with an IHSD = an interaction term between the indicator variable vector P(x) which takes on a value of one or zero at each of the price levels and CDD 14 EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results = an interaction term between the indicator variable vector P(x) which takes on a value of one or zero at each of the price levels and the indicator variable IHSD and CDD = the error for participant i on day t Data Used in Analysis We used 15 minute interval energy usage data collected from the smart meters installed on participant and control group customers within SmartGridCity TM to conduct the analysis. Note that we did not include all pricing participants or control group customers in the full duration of the analysis periods. Data were excluded for a variety of reasons: For the difference in differences analysis, we only used customers with complete data to determine each month s impact. o Customers who left the rates were only included for the time they were participants. o Customers were excluded if they did not have a smart meter installed a full year prior to participation. o We excluded data for customers on days with more than half of the data for day missing. For the regression analysis, we excluded energy data for the days after a customer left the rate. We also excluded days for specific customers from individual models for those days when a customer did not have complete data for the period. Extracting data from the data warehouse required special handling for the month a customer was added to the pricing pilot. We excluded that month s data for the preliminary analysis but added it in this final analysis. We excluded other clearly erroneous data, such as spikes that were more than 20 times adjacent intervals. For the difference in differences analysis, we excluded 0.42% of the interval data. For the on-peak regression, we excluded 0.59% of the data, and for the off-peak regression, we excluded 1.14% of the data. Event Days As noted in Chapter 2, event days were a key aspect of the Pricing Pilot. Customers on the CPP and PTR rates were notified a day ahead of each event day. Customers were encouraged to save energy between the hours of 2 p.m. and 8 p.m. on event days to avoid paying higher energy rates (CPP customers) or to earn a bill credit (PTR customers). Table 4 in the previous chapter lists event days called during the analysis period for each analysis year of the pilot. Estimating Impacts As described above, we estimated impacts for the pilot program using two methods: 1) difference in differences, and 2) regression modeling. The difference of differences approach EnerNOC Utility Solutions 15

gives us an approximation of savings based on direct comparison of participant and control groups during the pre-treatment and treatment periods. It allows us to calculate the difference in energy use, corrected for any preexisting differences between the participant and control groups. Because this method allows us to estimate the impacts directly, we can easily sum those and report on the annual ex post energy impacts. However, while the impact estimates are valid estimates of savings based on what actually happened during the analysis period, the difference of differences does not allow us to see the influence of independent variables on energy impacts. Regression modeling, on the other hand, enables us to quantify the variability from other known sources affecting energy use so that we can estimate impacts as a function of these parameters. Specially, we can see how impacts depend on weather, presence of IHSD, etc. This capability allows us to estimate energy use for an arbitrary day and participant type based on that day s weather and how impacts change with temperature. Therefore, with the regression models, we can provide estimates of the impacts achieved during the actual pilot weather conditions and predict impacts that would be achieved during other weather conditions, such as a normal weather year We looked at a variety of characteristics in estimating the impacts from the Pilot Program for each rate. Figure 3 and Figure 4 on the following two pages map the characteristics analyzed for the CPP/PTR and TOU rates, respectively, during the difference of differences ex post analysis. There are two primary differences in characteristics between the CPP/PTR and TOU rates: 1) The CPP/PTR rates have pricing events, and 2) the TOU rate includes some customers who are also enrolled in Saver s Switch (SS). The regression model considers additional characteristics not shown in these figures, including weather and presence of an inhome device (IHD). 16 EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results Figure 3. Data Parameters Included in Difference of Differences Analysis: CPP & PTR Rates Rate CPP/PTR Timeframe Oct 10-Sep 11 Oct 11-Sep 12 Oct 12-Sep 13 Phase Phase I Phase II Season Weekday Non- Summer Summer Event Day Status Non- Event Event Day of Week Period Weekend Off- Peak On- Peak EnerNOC Utility Solutions 17

Figure 4. Data Parameters Included in Difference of Differences Analysis: TOU Rate Rate TOU Timeframe Oct 10-Sep 11 Oct 11-Sep 12 Oct 12-Sep 13 Phase Phase I Phase II Saver Switch Yes No Season Non- Summer Summer Day of Week Period Weekday Weekend Off- Peak On- Peak 18 EnerNOC Utility Solutions

Chapter 4 Impact Results Organization of Results This chapter presents summary impact results from the difference of differences ex post analysis and the weather-normalized regression modeling. We categorize the results by rate and present a comparison of impacts for each analysis year of the pilot. Analysis year 2011 covers the period of October 1, 2010 through September 30, 2011. Similarly, analysis year 2012 section covers October 1, 2011 to September 30, 2012 and analysis year 2013 section covers October 1, 2012 to September 30, 2013. The chapter concludes with a summary of crosscutting Pilot Program impacts, including comparisons of estimated savings by rate, analysis year, participant phase, and presence of enabling technologies. We followed standard conventions for presenting the data throughout the report and in the appendices. The convention we followed in the tables is that the load (kw) impacts are labeled as reductions, so a positive value means lower energy demand. Since the energy (kwh) impacts are expected to go both directions, we refer to them as changes, with negative values meaning lower energy use during the period and higher values meaning higher energy use. In the figures, we also show kw reductions as positive values; but, for kwh, we plot energy savings instead of energy changes, therefore energy savings are shown as positive values. Preliminary difference of differences impact results were included in the 2011 and 2012 reports for those years. The 2011 and 2012 differences of differences impact results were updated for this report in several ways, in order to make the results consistent across years, to correctly account for participation in Public Service s Saver s Switch program before and during the pilot, and to stabilize the estimates. More details about these changes are described in the appendices. The Appendices contain detailed results and load profiles organized by analysis year. Difference of Differences Analysis: Ex Post Impacts Here we summarize the ex post impacts estimated with the difference of differences approach for each rate, each participant phase, and each analysis year. Since impacts are a dependent on the weather during each year of the pilot, it is important to consider the results in the context of the cooling degree days (CDDs) and heating degree days (HDDs) associated with each analysis year. Table 6 shows how the CDDs and HDDs varied considerably by year, with the 2012 analysis year having considerably more CDDs than in 2011 and 2013, and 2013 having more HDDs than 2011 or 2012. The table also includes data for a normal weather year, with events defined consistently with the way Public Service called events for the pilot. EnerNOC Utility Solutions 19

Table 6. Cooling and Heating Degree Days by Analysis Year Analysis Year 2011 2012 2013 Normal Year Average Summer Cooling Degree Days Weekday Non-Event 4.6 6.5 5.6 3.9 Weekday Event 9.0 12.8 8.8 10.3 Weekend 5.8 6.9 4.8 4.6 Average Non-Summer Heating Degree Days Weekday Non-Event 14.2 13.2 15.2 13.5 Weekday Event 12.8 27.3 27.0 Weekend 13.5 13.2 15.5 14.3 CPP Results Event Day On-Peak Demand Impacts Table 7 and Figure 5 summarize the ex post load reduction results on event days for CPP participants. The data include Phases I and II and analysis years 2011 through 2013. The kw reduction table lists baseline average on-peak kw, average on-peak kw reduction, and percent reduction for average summer and non-summer event days. The figure plots average on-peak kw reduction for summer and non-summer events and includes trend lines for the data. Note that there were no non-summer events in 2011. Table 7. CPP Ex Post Results: Event Day On-Peak Demand Impacts CPP Difference of Differences Results Event Day On-Peak Demand Impacts Average Summer Event Day Analysis Year 2011 2012 2013 Phase Ph. I Ph. II Ph. I Ph. II Ph. I Ph. II Baseline Avg. On-Peak kw 0.93 1.03 1.06 1.22 0.93 1.02 Avg. On-Peak kw Reduction 0.27 0.27 0.28 0.29 0.20 0.13 Percent Reduction 29% 26% 26% 23% 22% 13% Average Non-Summer Event Day Baseline Avg. On-Peak kw 0.98 0.97 1.08 1.12 Avg. On-Peak kw Reduction 0.23 0.13 0.18 0.09 Percent Reduction 24% 14% 16% 8% 20 EnerNOC Utility Solutions

Average On-Peak kw Reduction SmartGridCity Pricing Pilot Impact Evaluation Results Figure 5. CPP Ex Post Results: Event Day On-Peak Demand Impacts 0.35 0.30 Ex Post Event Day On-Peak kw Reduction, CPP Summer Events Phase I Ex Post Phase II Ex Post Non-Summer Winter Events 0.25 0.20 0.15 0.10 0.27 0.27 0.28 0.29 0.20 0.23 0.18 0.05 0.13 0.13 0.09 0.00 Summer 2011 Summer 2012 Summer 2013 Winter 2012 Winter 2013 The highest load reduction for the pilot in each year was for the CPP customers on summer event days. On these most extreme days, when prices are highest, customers respond. This is true for both phases, however here, as nearly everywhere else, the Phase I reductions are greater than the Phase II reductions on a percentage basis, and they are either greater, or in one case about the same, on a kw basis. Across the three years, the percent savings decrease, while 2012, the hottest year, has the highest kw savings. The decrease in percentage savings may represent some loss of interest or commitment on the part of the customers, but it is difficult to say based on the load data. The non-summer event day load reductions are much less than the summer event day load reductions, both because the baseline usage is less, and because the percent reduction is smaller. Air conditioning (AC) load is generally the easiest electric load to reduce in a home, and the non-summer period has little or no AC usage, resulting in lower savings. Figure 6 below shows the on-peak kw load reduction for each event across the entire pilot period. These events are on different days of the week, with different temperatures, and in different seasons. The wide variation in load reductions across the events is clear, even within one season. The highest load reductions are on the 2012 events, most likely due to those being EnerNOC Utility Solutions 21

7/1/2011 8/30/2011 10/29/2011 12/28/2011 2/26/2012 4/26/2012 6/25/2012 8/24/2012 10/23/2012 12/22/2012 2/20/2013 4/21/2013 6/20/2013 8/19/2013 10/18/2013 Average On-Peak kw Reduction the days with the hottest temperature. The non-summer events are lower in general. In most cases, but not all, the Phase I load reductions are greater than the Phase II load reductions. The results for individual events are generally consistent with the results for the averages by season. Figure 6. CPP Ex Post Results: Scatter Plot of Event Day On-Peak Demand Impacts 0.45 0.40 0.35 CPP - Ex Post Event Day On-Peak kw Reduction By Event Day CPP Phase I CPP Phase II 0.30 0.25 0.20 0.15 0.10 0.05 0.00 Non-Event Weekday On-Peak Demand Impacts Table 8 and Figure 7 summarize the ex post load reduction results on non-event days for CPP participants. The data include Phases I and II and analysis years 2011 through 2013. The kw reduction table lists baseline average on-peak kw, average on-peak kw reduction, and percent reduction for average summer and non-summer non-event weekdays. The figure plots average on-peak kw reduction for summer and non-summer non-event weekdays and includes trend lines for the data. Because the winter of 2010-2011 was when Phase II customers were being recruited, we do not include results for that period for Phase II. 22 EnerNOC Utility Solutions

Average On-Peak kw Reduction SmartGridCity Pricing Pilot Impact Evaluation Results Table 8. CPP Ex Post Results: Non-Event Weekday On-Peak Demand Impacts CPP Difference of Differences Results Non-Event Weekday On-Peak Demand Impacts Analysis Year 2011 2012 2013 Phase Ph. I Ph. II Ph. I Ph. II Ph. I Ph. II Average Summer Non-Event Weekday Baseline Avg. On-Peak kw 0.86 0.92 0.86 0.98 0.90 1.00 Avg. On-Peak kw Reduction 0.14 0.11 0.14 0.10 0.14 0.08 Percent Reduction 17% 12% 16% 10% 16% 8% Average Non-Summer Non-Event Weekday Baseline Avg. On-Peak kw 0.87 0.88 0.92 0.94 1.01 Avg. On-Peak kw Reduction 0.04 0.09 0.03 0.09 0.03 Percent Reduction 4% 10% 3% 10% 3% Figure 7. CPP Ex Post Results: Non-Event Weekday On-Peak Demand Impacts 0.16 0.14 Ex Post Non-Event Weekday On-Peak kw Reduction, CPP Summer Events Phase I Ex Post Phase II Ex Post Non-Summer Winter Events 0.12 0.10 0.08 0.06 0.04 0.14 0.11 0.14 0.14 0.10 0.08 0.09 0.09 0.02 0.03 0.03 0.00 Summer 2011 Summer 2012 Summer 2013 Winter 2012 Winter 2013 The load reductions for CPP customers on non-event summer weekdays are less than on event weekdays, but interestingly do not drop off as much as the event day load reductions across the three years, especially for Phase I. Excluding the first winter, when customers may have just been learning about the rate and had not yet been subjected to any events, the winter savings EnerNOC Utility Solutions 23

are consistent year to year. This may be because actions that people take to save energy on all weekdays, such as changing programmable thermostats or making permanent equipment changes, tend to persist over time. Changes made only on event days tend to be more behavioral, and may decrease over time. The Phase I load reductions are higher than the Phase II load reductions for all years and both seasons. And as with the event days, the non-summer savings are much lower than the summer savings. It is interesting to note that even though the CPP non-event on-peak price is lower than the TOU on-peak price, the CPP customers save more. Other pricing pilots have found similar results, driven by the fact that CPP customers do things to reduce load on event days, but in many cases they do those things more broadly, and save energy on other days as well. Energy Impacts Table 9 and Figure 8 present the ex post energy results for CPP participants. The data include Phases I and II and analysis years 2011 through 2013. The kwh table lists baseline kwh, kwh change (negative numbers reflect energy savings), and percent change in consumption for onpeak hours, off-peak hours, and overall for the year. The figure plots on-peak, off-peak, and overall kwh savings (displayed as positive values) by phase and year. Because Phase II began in June 2011, we only have data for a partial year. Therefore, we have not included 2011 kwh values for Phase II. Table 9. CPP Ex Post Results: Changes in Energy Consumption CPP Difference of Differences Results Energy Impacts On-Peak Consumption Analysis Year 2011 2012 2013 Phase Ph. I Ph. II Ph. I Ph. II Ph. I Ph. II Baseline kwh 1,302 1,333 1,433 1,394 1,515 kwh Change -112-172 -94-169 -73 Percent Change -9% -13% -7% -12% -5% Off-Peak Consumption Baseline kwh 5,435 5,621 6,032 5,908 6,412 kwh Change -200-402 -27-526 -6 Percent Change -4% -7% 0% -9% 0% Overall Consumption Baseline kwh 6,738 6,954 7,465 7,302 7,927 kwh Change -312-574 -121-695 -79 Percent Change -5% -8% -2% -10% -1% 24 EnerNOC Utility Solutions

kwh Savings SmartGridCity Pricing Pilot Impact Evaluation Results Figure 8. CPP Ex Post Results: Energy Savings 800 700 600 500 Ex Post kwh Savings, CPP 2011 2012 2013 Off-Peak 526 574 695 Annual 400 300 On-Peak 402 312 200 100 0 200 172 169 112 121 94 73 79 27 6 Phase I Phase II Phase I Phase II Phase I Phase II The on-peak energy savings for the CPP customers are consistent with the non-event weekday on-peak load reductions, with the first year small, and a small decrease between 2012 and 2013. Phase II has less savings than Phase I during on-peak and during off-peak. The off-peak savings are somewhat surprising for two reasons. First, there are savings in the off-peak period, instead of an increase in energy use, which would be expected based on the lower price. However, in many pricing pilots across the country, customers on time-varying rates often reduce their consumption overall, even in off-peak periods. This is generally attributed to an increased awareness of and focus on energy use. People think about what they do that uses energy, and make changes to reduce their energy use. Some of these changes may be during on-peak times, but many reduce energy use across all hours. While the total energy reduction during off-peak is higher than the total energy reduction during peak hours, there are many more off-peak hours in a year, meaning that the average hourly reduction during off-peak hours is much less than the reduction during on-peak hours. This effect is less pronounced in the Phase II customers, which supports the idea that Phase I customers, being the first to volunteer for the pilot, probably are more engaged and aware of energy use, and so take more actions than the Phase II customers. The load shapes (included in the appendices) also show a drop in load at around 9:00 AM, implying that some customers are using a programmable thermostat to save energy when they EnerNOC Utility Solutions 25

leave for the day. This drop is not as pronounced as the drop at the beginning of the on-peak period, but still results in savings during the off-peak hours leading up to the beginning of the on-peak period. The second surprising thing is that the off-peak savings increase each year for Phase I, and because off-peak hours are more numerous than the on-peak hours, the total energy savings increases as well. This could be simply due to weather variations and timing, or some other behavioral reason that we don t understand. It could be due to a cumulative effect of multiple actions taken over time. PTR Results Event Day On-Peak Demand Impacts Table 10 and Figure 9 summarize the ex post load reduction results for PTR participants. The data include Phases I and II and analysis years 2011 through 2013. The kw reduction table lists baseline average on-peak kw, average on-peak kw reduction, and percent reduction for average summer and non-summer event days. The figure plots average on-peak kw reduction for summer and non-summer events and includes trend lines for the data. Note that there were no non-summer events in 2011. Table 10. PTR Ex Post Results: Event Day On-Peak Demand Impacts PTR Difference of Differences Results Event Day On-Peak Demand Impacts Average Summer Event Day Analysis Year 2011 2012 2013 Phase Ph. I Ph. II Ph. I Ph. II Ph. I Ph. II Baseline Avg. On-Peak kw 1.05 1.00 1.24 1.20 1.13 1.11 Avg. On-Peak kw Reduction 0.14 0.12 0.10 0.10 0.09 0.09 Percent Reduction 14% 12% 8% 8% 8% 8% Average Non-Summer Event Day Baseline Avg. On-Peak kw 0.88 0.87 0.94 0.93 Avg. On-Peak kw Reduction 0.04 0.03 0.05 0.02 Percent Reduction 5% 3% 5% 2% 26 EnerNOC Utility Solutions

Average On-Peak kw Reduction SmartGridCity Pricing Pilot Impact Evaluation Results Figure 9. PTR Ex Post Results: Event Day On-Peak Demand Impacts 0.16 0.14 Ex Post Event Day On-Peak kw Reduction, PTR Summer Events Phase I Ex Post Phase II Ex Post Non-Summer Winter Events 0.12 0.10 0.08 0.14 0.06 0.12 0.04 0.10 0.10 0.09 0.09 0.02 0.00 0.05 0.04 0.03 0.02 Summer 2011 Summer 2012 Summer 2013 Winter 2012 Winter 2013 Figure 10 below shows the on-peak kw load reduction for each event across the entire pilot period. These events are on different days of the week, with different temperatures, and in different seasons. The wide variation in load reductions across the events is clear, even within one season. Unlike the CPP, the highest load reductions here are in the first summer. The non-summer events are lower in general. In most cases, but not all, the Phase I load reductions are greater than the Phase II load reductions. The results for individual events are generally consistent with the results for the averages by season. EnerNOC Utility Solutions 27

7/1/2011 8/30/2011 10/29/2011 12/28/2011 2/26/2012 4/26/2012 6/25/2012 8/24/2012 10/23/2012 12/22/2012 2/20/2013 4/21/2013 6/20/2013 8/19/2013 10/18/2013 Average On-Peak kw Reduction Figure 10. PTR Ex Post Results: Scatter Plot of Event Day On-Peak Demand Impacts 0.25 0.20 PTR - Ex Post Event Day On-Peak kw Reduction by Event Day PTR Phase I PTR Phase II 0.15 0.10 0.05 0.00 PTR customers reduce load during on-peak periods on event days, but reduce less than the CPP customers. This is consistent to what others in the industry have found in more recent pilots. The general consensus is that the stick is more powerful than the carrot, meaning that when customers face higher prices, as in a CPP, they are more compelled to reduce load than if they are given a chance for a rebate, with no loss if they do nothing. However, the load reduction is still there, and greater than the TOU average weekday load reduction. The load reductions during the summer are higher than during the winter, but unlike with other groups, the Phase I and Phase II load reductions are the same in summer 2012 and summer 2013. This may be just random chance that the Phase I and Phase II customers in this group are more similar, or are taking similar actions during the peak period, or it could be something that can t easily be measured. There is a small drop in savings across the three years. This may be due to weather differences, other factors, or even just random variation, or could be a sign of decreased engagement. Energy Impacts Table 11 and Figure 11 present the ex post energy results for PTR participants. The data include Phases I and II and analysis years 2011 through 2013. The kwh table lists baseline kwh, kwh change (negative numbers reflect energy savings), and percent change in consumption for on-peak hours, off-peak hours, and overall for the year. The figure plots onpeak, off-peak, and overall kwh savings (displayed as positive values) by phase and year. Because Phase II began in June 2011, we only have data for a partial year. Therefore, we have not included 2011 kwh values for Phase II. 28 EnerNOC Utility Solutions

kwh Savings SmartGridCity Pricing Pilot Impact Evaluation Results Table 11. PTR Ex Post Results: Changes in Energy Consumption PTR Difference of Differences Results Energy Impacts Analysis Year 2011 2012 2013 Phase Ph. I Ph. II Ph. I Ph. II Ph. I Ph. II On-Peak Consumption Baseline kwh 1,317 1,361 1,330 1,435 1,413 kwh Change -45-58 -28-73 -43 Percent Change -3% -4% -2% -5% -3% Off-Peak Consumption Baseline kwh 5,313 5,462 5,337 5,652 5,569 kwh Change -172-326 -165-340 -227 Percent Change -3% -6% -3% -6% -4% Overall Consumption Baseline kwh 6,630 6,823 6,667 7,087 6,982 kwh Change -217-385 -193-413 -270 Percent Change -3% -6% -3% -6% -4% Figure 11. PTR Ex Post Results: Energy Savings 500 450 400 350 Ex Post kwh Savings, PTR 2011 2012 2013 340 326 Off-Peak 385 413 Annual 300 270 250 200 150 On-Peak 172 165 227 217 193 100 50 45 58 73 28 43 0 Phase I Phase II Phase I Phase II Phase I Phase II EnerNOC Utility Solutions 29

The energy savings for the PTR are similar to the energy savings for the CPP, though smaller in magnitude. Customers are more aware of their energy use, and are taking actions that reduce energy use across all hours, not only in the on-peak hours. With the energy impacts, the difference between the Phase I and Phase II customers is more pronounced than with the demand impacts. TOU Results The TOU results include the following abbreviations: TOU SS Indicates TOU customers who also participate in the Saver s Switch program. Note that these customers all have central air conditioning (CAC), which is a requirement for the SS program, and so have higher energy use on average than those without SS. TOU NSS Indicates TOU customers not enrolled in the Saver s Switch program. This group, like the PTR and CPP groups, includes a mix of those with CAC and those without. Weekday On-Peak Demand Impacts Table 12 and Figure 12 summarize the ex post load reduction results on average weekdays for TOU participants. The data include Phases I and II and analysis years 2011 through 2013. The kw reduction table lists baseline average on-peak kw, average on-peak kw reduction, and percent reduction for average summer and non-summer weekdays. The figure plots average on-peak kw reduction for summer and non-summer weekdays and includes trend lines for the data. Because the winter of 2010-2011 was when Phase II customers were being recruited, we do not include results for that period for Phase II. TOU customers who are also SS customers reduce energy more on summer weekdays than those who are not SS customers. This is most likely because all the SS customers have CAC, whereas the NSS customers are a mixture of those with and without CAC. Customers with CAC have more load available to reduce, and AC load is easier to reduce in response to price than other loads. In winter, however, TOU customers with SS have virtually no savings, whereas TOU customers without SS show load reduction. The load reductions for TOU customers, both with and without SS, are less in general than the savings for CPP and PTR on event days, and even less than CPP on non-event weekdays. This is common in the industry TOU rates that do not have a dynamic event component do not engage customers as much as rates with events, which serve to remind customers that they are on a time-varying rate. With lower awareness of the rate, the savings tend to be less. 30 EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results Table 12. TOU Ex Post Results: Weekday On-Peak Demand Impacts TOU Difference of Differences Results Weekday On-Peak Demand Impacts Analysis Year 2011 2012 2013 Phase Ph. I Ph. II Ph. I Ph. II Ph. I Ph. II TOU-SS Average Summer Weekday Baseline Avg. On-Peak kw 1.52 1.52 1.63 1.60 1.69 1.63 Avg. On-Peak kw Reduction 0.12 0.14 0.10 0.10 0.12 0.08 Percent Reduction 8% 9% 6% 7% 7% 5% TOU-SS Average Non-Summer Weekday Baseline Avg. On-Peak kw 1.01 0.98 0.99 1.04 1.03 Avg. On-Peak kw Reduction 0.02-0.01 0.01 0.01 0.01 Percent Reduction 2% -1% 1% 1% 1% TOU-NSS Average Summer Weekday Baseline Avg. On-Peak kw 0.89 1.04 0.95 1.15 0.99 1.23 Avg. On-Peak kw Reduction 0.08 0.06 0.07 0.06 0.05 0.04 Percent Reduction 9% 6% 7% 5% 5% 3% TOU-NSS Average Non-Summer Weekday Baseline Avg. On-Peak kw 0.86 0.87 0.95 0.93 1.01 Avg. On-Peak kw Reduction 0.01 0.04 0.03 0.04 0.03 Percent Reduction 1% 4% 3% 4% 3% EnerNOC Utility Solutions 31

Average On-Peak kw Reduction Figure 12. TOU Ex Post Results: Weekday On-Peak Demand Impacts, with Saver s Switch 0.16 0.14 Ex Post Weekday On-Peak kw Reduction, TOU SS Summer Events Phase I Ex Post Phase II Ex Post Non-Summer Winter Events Events 0.12 0.10 0.08 0.06 0.04 0.12 0.14 0.10 0.10 0.12 0.08 0.02 0.00-0.02-0.01 0.01 0.01 0.01 Summer 2011 Summer 2012 Summer 2013 Winter 2012 Winter 2013-0.04 32 EnerNOC Utility Solutions

Average On-Peak kw Reduction SmartGridCity Pricing Pilot Impact Evaluation Results Figure 13. TOU Ex Post Results: Weekday On-Peak Demand Impacts, without Saver s Switch 0.16 0.14 Ex Post Weekday On-Peak kw Reduction, TOU NSS Summer Events Phase I Ex Post Phase II Ex Post Non-Summer Winter Events 0.12 0.10 0.08 0.06 0.04 0.02 0.08 0.07 0.06 0.06 0.05 0.04 0.04 0.04 0.03 0.03 0.00 Summer 2011 Summer 2012 Summer 2013 Winter 2012 Winter 2013 Energy Impacts Table 13 and Figure 14 present the ex post energy results for TOU SS participants, and Table 14 and Figure 15 present the results for TOU NSS participants. The data include Phases I and II and analysis years 2011 through 2013. The kwh table lists baseline kwh, kwh change (negative numbers reflect energy savings), and percent change in consumption for on-peak hours, off-peak hours, and overall for the year. The figure plots on-peak, off-peak, and overall kwh savings (displayed as positive values) by phase and year. Because Phase II began in June 2011, we only have data for a partial year. Therefore, we have not included 2011 kwh values for Phase II. EnerNOC Utility Solutions 33

kwh Savings Table 13. TOU Ex Post Results: Changes in Energy Consumption, with Saver s Switch TOU SS Difference of Differences Results Energy Impacts Analysis Year 2011 2012 2013 Phase Ph. I Ph. II Ph. I Ph. II Ph. I Ph. II On-Peak Consumption Baseline kwh 1,789 1,810 1,805 1,891 1,847 kwh Change -82-36 -68-66 -57 Percent Change -5% -2% -4% -4% -3% Off-Peak Consumption Baseline kwh 6,721 6,674 6,639 6,854 6,699 kwh Change 93 134 39 24 80 Percent Change 1% 2% 1% 0% 1% Overall Consumption Baseline kwh 8,510 8,484 8,444 8,744 8,546 kwh Change 11 98-29 -43 23 Percent Change 0% 1% 0% 0% 0% Figure 14. TOU Ex Post Results: Energy Savings, with Saver s Switch Ex Post kwh Savings, TOU SS 150 On-Peak 2011 2012 2013 Annual 100 50 82 36 66 68 57 Off-Peak 43 29 0-50 Phase I Phase II Phase I Phase II Phase I -11 Phase II -24-23 -39-100 -93-80 -98-150 -134 34 EnerNOC Utility Solutions

kwh Savings SmartGridCity Pricing Pilot Impact Evaluation Results Table 14. TOU Ex Post Results: Changes in Energy Consumption, without Saver s Switch TOU NSS Difference of Differences Results Energy Impacts Analysis Year 2011 2012 2013 Phase Ph. I Ph. II Ph. I Ph. II Ph. I Ph. II On-Peak Consumption Baseline kwh 1,314 1,353 1,537 1,426 1,628 kwh Change -50-72 -62-65 -50 Percent Change -4% -5% -4% -5% -3% Off-Peak Consumption Baseline kwh 5,305 5,437 6,159 5,656 6,418 kwh Change 178 83-101 78-150 Percent Change 3% 2% -2% 1% -2% Overall Consumption Baseline kwh 6,619 6,790 7,697 7,082 8,047 kwh Change 128 11-163 13-200 Percent Change 2% 0% -2% 0% -2% Figure 15. TOU Ex Post Results: Energy Savings, without Saver s Switch Ex Post kwh Savings, TOU NSS 250 200 150 On-Peak 2011 2012 2013 Off-Peak 150 Annual 163 200 100 50 50 72 65 62 50 101 0-50 -100-150 Phase I Phase II Phase I Phase II Phase I Phase II -11-13 -83-78 -128-200 -178 EnerNOC Utility Solutions 35

The TOU-SS customers and the Phase I TOU-NSS customers increased their energy use offpeak, and decreased it on-peak, indicating that the customers understood the rate. The overall impacts were very close to zero for these customers, meaning that the increase in the off-peak approximately offset the decrease in the on-peak. Phase II TOU-NSS customers, however, decreased energy use during both on-peak and off-peak periods, and as a result, decreased energy use overall. 36 EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results Regression Analysis: Weather-Normalized Impacts The regression model developed was applied to two types of weather data: 1) actual weather data for the three analysis years; and 2) normal year weather data. For the normal weather, we used typical meteorological year (TMY3) weather data from the National Oceanic and Atmospheric Administration based on 1976-2005. Here we present the results using normal weather data to show the year over year changes with the effects of different weather across the years removed. By comparing these results with the ex post findings, we can investigate the changes in impacts across the years that are not weather-related. The regression model was created based on all customers, combining both Phase I and Phase II, and does not differentiate between the two. This was done to allow for consistent estimation by the model, to leverage the maximum value by using all data available, and because the nature of the recruiting ended up very similar between the two groups. Because of this, care should be taken when making direct comparisons of the ex post results for Phase I and Phase II with the regression results. We also modeled the effects of IHSDs on participant energy use and response to pricing signals. Because there were only 62 Pricing Pilot participants with IHSDs remaining in their homes by the end of the pilot, the results for customers with IHSDs should be considered anecdotal, and should not be generalized to a broader population. Even though the results apply only to this small group of customers with IHSDs, they do provide a general sense of the affect of IHSDs on energy use. Comparing the regression model results and the ex post difference of differences results presents challenges. They are estimates of the same load reduction, but the assumptions of the methods are different, and the regression model, by its nature, makes assumptions about the nature of relationships between variables in order to quantify that relationship. The regression results provide the most value when used to assess those relationships, and to look at differences between scenarios and time periods. The ex post difference of differences provide the most robust direct estimates of savings for a specific time period or year. Because of this, the focus on the regression results should be how they differ between years and scenarios, and how they help us understand how the savings relate to different variables. CPP Results Event Day On-Peak Demand Impacts Table 15 and Figure 16 summarize the weather-normalized load reduction results for CPP participants on event days. The data include analysis years 2011 through 2013. The table lists baseline average on-peak kw, average on-peak kw reduction, and percent reduction for average summer and non-summer event days. It also shows estimates for both participants with IHSDs and participants without IHSDs. 4 As mentioned above, the IHSD results should be considered anecdotal, and not representative of the broader population. The figure plots the weather-normalized average on-peak kw reduction for summer and non-summer events as an 4 The vast majority of participants did not have IHSDs. Only four Phase I and three Phase II CPP participants had IHSDs remaining in their homes at the end of the pilot. EnerNOC Utility Solutions 37

Average On-Peak kw Reduction overlay on top of the ex post results shown previously. To simplify the graph, we only include non-ihsd regression data since that represents that majority of participants. Note that there were no non-summer events in 2011. Table 15. CPP Weather-Normalized Results: Event Day On-Peak Demand Impacts CPP Regression Results Event Day On-Peak Demand Impacts Analysis Year 2011 2012 2013 In-Home Smart Device Yes No Yes No Yes No Average Summer Event Day Baseline Avg. On-Peak kw 1.00 1.46 0.99 1.46 0.99 Avg. On-Peak kw Reduction 0.28 0.67 0.18 0.66 0.18 Percent Reduction 27% 46% 18% 45% 18% Average Non-Summer Event Day Baseline Avg. On-Peak kw 0.92 0.98 0.92 0.98 Avg. On-Peak kw Reduction -0.06 0.08-0.09 0.09 Percent Reduction -7% 8% -9% 9% Figure 16. CPP Weather-Normalized Results: Event Day On-Peak Demand Impacts, without IHSD 0.35 0.30 Event Day On-Peak kw Reduction, CPP Summer Events Phase I Ex Post Phase II Ex Post Weather-Normalized Non-Summer Winter Events 0.25 0.20 0.15 0.28 0.10 0.18 0.18 0.05 0.08 0.09 0.00 Summer 2011 Summer 2012 Summer 2013 Winter 2012 Winter 2013 38 EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results The weather-normalized regression results show that the summer on-peak event day load reduction dropped after the first year, but was the same in the second and third year. The winter on-peak event day load reduction was much less than the summer, but increased slightly between the second and third year, though this difference is so small as to be considered negligible. Non-Event Weekday On-Peak Demand Impacts Table 16 and Figure 17 summarize the weather-normalized load reduction results for CPP participants on non-event days. The data include analysis years 2011 through 2013. The table lists baseline average on-peak kw, average on-peak kw reduction, and percent reduction for average summer and non-summer weekdays. It also shows estimates for both participants with IHSDs and participants without IHSDs. 5 As mentioned above, the IHSD results should be considered anecdotal, and not representative of the broader population. The overlay of weathernormalized and ex post data allows us to directly compare results based on normal weather with those based on the actual weather that occurred during the pilot. To simplify the graph, we only include non-ihsd regression data since that represents that majority of participants. Note that there were no non-summer events in 2011. Table 16. CPP Weather-Normalized Results: Non-Event Weekday On-Peak Demand Impacts CPP Regression Results Non-Event Weekday On-Peak Demand Impacts Analysis Year 2011 2012 2013 In-Home Smart Device Yes No Yes No Yes No Average Summer Non-Event Weekday Baseline Avg. On-Peak kw 0.86 1.02 0.84 1.02 0.84 Avg. On-Peak kw Reduction 0.10 0.20 0.09 0.18 0.08 Percent Reduction 12% 20% 10% 18% 10% Average Non-Summer Non-Event Weekday Baseline Avg. On-Peak kw 0.83 0.86 0.82 0.86 Avg. On-Peak kw Reduction -0.10 0.02-0.14-0.01 Percent Reduction -12% 2% -17% -2% 5 The vast majority of participants did not have IHSDs. Only four Phase I and three Phase II CPP participants had IHSDs remaining in their homes at the end of the pilot. EnerNOC Utility Solutions 39

Average On-Peak kw Reduction Figure 17. CPP Weather-Normalized Results: Non-Event Weekday On-Peak Demand Impacts, without IHSD 0.16 0.14 0.12 Non-Event Weekday On-Peak kw Reduction, CPP Summer Summer Events Events Phase I Ex Post Phase II Ex Post Weather-Normalized Non-Summer Winter Events Events 0.10 0.08 0.06 0.04 0.10 0.09 0.08 0.02 0.00-0.02 0.02-0.01 Summer 2011 Summer 2012 Summer 2013 Winter 2012 Winter 2013 The summer non-event weekday load reductions drop off somewhat across the three years of the pilot, though not as dramatically as the summer event day load reductions. The winter event day load reductions are very small, and basically disappear in the third year, showing a slight increase in usage but it is important to note that there were only 7 non-summer events during the pilot, so the results are not as generalizable as the summer results, which were based on over 30 events. PTR Results Event Day On-Peak Demand Impacts Table 17 and Figure 18 summarize the weather-normalized load reduction results for PTR participants on event days. The data include analysis years 2011 through 2013. The table lists baseline average on-peak kw, average on-peak kw reduction, and percent reduction for average summer and non-summer event days. It also shows estimates for both participants with IHSDs and participants without IHSDs. 6 As mentioned above, the IHSD results should be considered anecdotal, and not representative of the broader population. The figure plots the weather-normalized average on-peak kw reduction for summer and non-summer events as an 6 Only eight Phase I and seven Phase II PTR participants had IHSDs remaining in their homes at the end of the pilot. 40 EnerNOC Utility Solutions

Average On-Peak kw Reduction SmartGridCity Pricing Pilot Impact Evaluation Results overlay on top of the ex post results shown previously. To simplify the graph, we only include non-ihsd regression data since that represents that majority of participants. Note that there were no non-summer events in 2011. Table 17. PTR Weather-Normalized Results: Event Day On-Peak Demand Impacts PTR Regression Results Event Day On-Peak Demand Impacts Analysis Year 2011 2012 2013 In-Home Smart Device Yes No Yes No Yes No Average Summer Event Day Baseline Avg. On-Peak kw 1.00 1.55 1.00 1.55 1.00 Avg. On-Peak kw Reduction 0.14 0.58 0.06 0.31 0.05 Percent Reduction 14% 38% 6% 20% 5% Average Non-Summer Event Day Baseline Avg. On-Peak kw 0.87 0.87 0.87 0.87 Avg. On-Peak kw Reduction 0.27 0.04 0.32 0.10 Percent Reduction 31% 5% 37% 11% Figure 18. PTR Weather-Normalized Results: Event Day On-Peak Demand Impacts, without IHSD 0.16 0.14 Event Day On-Peak kw Reduction, PTR Summer Events Phase I Ex Post Phase II Ex Post Weather-Normalized Non-Summer Winter Events 0.12 0.1 0.08 0.14 0.06 0.10 0.04 0.02 0.06 0.05 0.04 0 Summer 2011 Summer 2012 Summer 2013 Winter 2012 Winter 2013 EnerNOC Utility Solutions 41

The summer event day load reductions show a decrease across the three years of the pilot, with a big decrease after the first year, and a smaller decrease after the second year. The nonsummer event day load reductions show an increase after the first year, but it is important to note that there were only 7 non-summer events during the pilot, so the results are not as generalizable as the summer results, which were based on over 30 events. TOU Results Weekday On-Peak Demand Impacts Table 17 and Figure 18 summarize the weather-normalized load reduction results for TOU participants on weekdays. The data include analysis years 2011 through 2013. The table lists baseline average on-peak kw, average on-peak kw reduction, and percent reduction for average summer and non-summer event days. It also shows estimates for both participants with IHSDs and participants without IHSDs. 7 As mentioned above, the IHSD results should be considered anecdotal, and not representative of the broader population. The figure plots the weather-normalized average on-peak kw reduction for summer and non-summer events as an overlay on top of the ex post results. Because the ex post results were split between SS and non-ss, we have combined the exp post results using weights based on the number of SS and non-ss customers in the TOU group. The regression analysis accounted for the presence of SS, but did not create separate estimates for the two groups. To simplify the graph, we only include non-ihsd regression data since that represents that majority of participants. Note that there were no non-summer events in 2011. Table 18. TOU Weather-Normalized Results: Weekday On-Peak Demand Impacts TOU Regression Results Weekday On-Peak Demand Impacts Analysis Year 2011 2012 2013 In-Home Smart Device Yes No Yes No Yes No Average Summer Weekday Baseline Avg. On-Peak kw 0.98 1.25 1.03 1.26 1.03 Avg. On-Peak kw Reduction 0.08 0.19 0.08-0.08 0.06 Percent Reduction 8% 15% 7% -6% 6% Average Non-Summer Weekday Baseline Avg. On-Peak kw 0.89 0.88 0.87 0.88 0.88 Avg. On-Peak kw Reduction 0.01 0.03 0.01 0.03 0.01 Percent Reduction 2% 4% 2% 4% 2% 7 Only forty TOU participants had IHSDs remaining in their homes at the end of the pilot. 42 EnerNOC Utility Solutions

Average On-Peak kw Reduction SmartGridCity Pricing Pilot Impact Evaluation Results Figure 19. TOU Weather-Normalized Results: Weekday On-Peak Demand Impacts, without IHSD 0.10 0.09 0.08 Weekday On-Peak kw Reduction, TOU Summer Events Phase I Ex Post Phase II Ex Post Weather-Normalized Non-Summer Winter Events Events 0.07 0.06 0.05 0.04 0.03 0.08 0.08 0.06 0.02 0.01 0.00 0.01 0.01 Summer 2011 Summer 2012 Summer 2013 Winter 2012 Winter 2013 The summer on-peak weekday load reductions drop off somewhat in the third year, but the winter reductions don t change between 2012 and 2013. The TOU group appears to have less decay in on-peak load reduction across the three years of the pilot than CPP and PTR. Discussion of Cross-Cutting Results We can draw several conclusions by comparing the results across the rates, pilot timeframe, participant phases, and presence of enabling technologies: Comparison of Rates: The CPP rate provides more load reduction during summer onpeak hours on event days and on non-event days than both the PTR and the TOU. TOU provides consistent savings on every day, but does not provide the extra decrease on event days, since there is no event pricing. The PTR, while attractive to customers because of its no-lose rebate, does not provide as much load reduction per customer as the CPP. The PTR and the CPP both also result in off-peak energy savings. Year-Over-Year Comparison: In nearly all cases, the difference of differences ex post event day load impacts diminished over time. The decreasing impacts were partly influenced by dramatic differences in the temperatures for the three years, but also may indicate that there was a drop off in persistence that affected the savings. The EnerNOC Utility Solutions 43

regression analysis using normal weather for all three years supports this notion since it showed lower event day load reductions in the second and third years. The overall energy savings did not show as clear a trend over time, which is not surprising as energy consumption over a longer time period is more stable; however energy savings tended to increase across the years in most cases. Comparison of Phases: The results indicate differences between the Phase I customers, who were recruited on an opt-in basis, and the Phase II customers, who were recruited in a pseudo-opt-out manner. Phase I load reductions were higher than Phase II, and tend to drop off less across the three years. Care must be taken, however, with these results, since the recruitment for Phase II was not consistent and was not truly opt-out. In addition, since the Phase I customers were recruited before the Phase II customers, Phase I would include those customers most interested in a timevarying rate who may be more inclined to respond. Influence of Saver s Switch: The TOU SS customers are higher energy users than those without SS, both in the summer and non-summer. The summer usage is higher due to the fact that the TOU SS customers all have CAC, whereas the TOU NSS include a mix of customers with and without CAC. TOU NSS participants tended to have higher on-peak load reductions than SS participants during the winter, but both NSS and SS participants had load reductions of about the same magnitude during the summer onpeak. In terms of energy savings, SS participants saved less during the on-peak overall for the year, but did not increase kwh usage by as much as NSS participants during the off-peak. Impact of IHSDs: There were not enough IHSDs installed on pricing pilot customers to provide results that could be generalized to a broader population. However, using the regression analysis, we could estimate the impacts for those customers with IHSDs, few though they were. For those with IHSDs, there were higher savings during summer events, presumably driven by the control of the thermostat setpoint by the devices. During non-summer events, those with IHSDs increased their usage in some cases. This may have been due to an assumption by the customers that the device was taking care of things for them, when during the winter, there is not any CAC load for the device to reduce. This result should be considered anecdotal, and not specifically indicative of impacts for a broader roll-out. The first two cross-cutting impacts are apparent in Figure 20 and Figure 21, which compare the ex post event day on-peak load reduction and energy savings results for the different rates across the years and seasons. For clarity, we have included only Phase I ex post results. 44 EnerNOC Utility Solutions

Annual kwh Savings Average On-Peak kw Reduction SmartGridCity Pricing Pilot Impact Evaluation Results Figure 20. Event Day On-Peak kw Reduction, Comparison by Rate 0.30 Event Day On-Peak kw Reduction Summer Events Non-Summer Winter Events CPP PTR 0.25 0.20 0.15 0.27 0.28 0.10 0.20 0.23 0.18 0.14 0.05 0.10 0.09 0.04 0.05 0.00 Summer 2011 Summer 2012 Summer 2013 Winter 2012 Winter 2013 Figure 21. Annual Energy Savings, Comparison by Rate 800 700 Annual kwh Savings CPP PTR TOU 695 600 574 500 400 385 413 300 200 312 217 100 0-100 -200 2011 2012 2013-33 -101 2 EnerNOC Utility Solutions 45

Chapter 5 Additional Analysis Peak Event Duration Analysis To determine the most appropriate peak period for any time-varying rates, we should consider the utility needs and customer perspectives. We first look at Public Service s needs, and then we discuss the customer perspective. Because most of the savings are during summer events and because Public Service is summer-peaking, we focus this analysis on the summer. Utility Perspective of Event Duration Figure 22 shows the system load (net firm obligation) for Public Service for the annual system peak days for the three years of the pilot. Note that the shapes, while not identical, are fairly similar. The peak period for the pilot is indicated by the dashed vertical lines. The actual system peak in each case was during the hour ending 17:00. Figure 22. System Load on Annual System Peak Days for 2011, 2012, and 2013 8,000 7,000 6,000 5,000 System Peak Day Comparison, 2011-2013 MW 4,000 3,000 2,000 1,000 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour July 18, 2011 June 25, 2012 July 11, 2013 From the utility perspective, the goal of time-varying pricing is to reduce the system peak. The peak period should be set to cover the period when the highest demands tend to occur on system peak days. But hitting the single highest peak hour is not sufficient while this could reduce that peak hour, if the hours just before or after the peak hour are nearly the same, reducing only the peak hour would not reduce the overall system peak by much at all, it would just shift it. So the peak period must be big enough to cover enough high demand hours that the peak doesn t simply shift without being reduced. Looking across the three annual system peak days, the peak period starting at 14:00 and going through 20:00 appears to capture the peak and the appropriate number of high load hours on either side. For these three days, the peak period includes all those hours that are within 5% of EnerNOC Utility Solutions 47

the peak demand, meaning that load during the peak period could be reduced by up to 5% without any issues around shifting without reducing the system peak. From the utility perspective, the peak period of 14:00 through 20:00 is appropriate. Based on the system load shape, the duration of the peak period is appropriate. Customer Perspective of Event Duration The customer perspective is less easy to pin down objectively. Customers tend to prefer shorter peak periods, since that reduces the effort required to save money, and the time that they are potentially uncomfortable. The six hour peak period that Public Service has used is comparable to the peak periods used by other utilities for time-varying pricing, most of which are 5, 6, or 7 hours long. Other utilities have not encountered significant customer dissatisfaction with this peak period length. Unless Public Service has customer research results that indicate customer dissatisfaction with the length of the peak period, we conclude that the peak period from 14:00 to 20:00 is reasonable from the customer perspective. Peak periods of similar duration have not caused significant customer dissatisfaction for other utilities. It is important to note that the customers enrolled in this pilot understood that the peak period would be the length that it is. There may be some potential additional market for a time-varying rate with a shorter peak period. While this may result in greater customer enrollment, it could be problematic from an operational standpoint. As discussed above, reducing load for only, say, 3 hours, might cause the peak to shift but not be reduced much at all. By comparison, Salt River Project (SRP) in Arizona, offers an alternate to their longstanding TOU rate that has only a 3 hour peak period. The cost savings to the customer are less, but so is the inconvenience. In order to address the operational issues, SRP is offering the 3-hour TOU rate for several different blocks of time, and limits the number of customers on each block. For instance, one group has a peak period from 14:00-17:00, another from 15:00-18:00, and another from 16:00-19:00. As long as all three groups are fully subscribed, this creates an overall load reduction which tends to match the shape of the system load from 14:00-19:00, maximizing the savings during the usual system peak hour of 16:00-17:00. Public Service might consider this in order to expand time-varying rates to more customers, but because of its complexity, it would be prudent to implement this after getting a few years of experience with a simpler time-varying rate design. A more complex rate design with multiple shorter blocks of on-peak time may be worth considering in the future. Number of Event Days Another important consideration is the number of event days allowed. The Public Service program is required to call at least 13 events and no more than 15 events. In order to understand if the maximum of 15 events is enough operationally, we can look at a load duration curve of daily system peak demands. This is a graph of the highest system demand on each day, sorted from the highest to the lowest for the year. In a fully operational program, the best 48 EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results that we can hope for is to have events on the 15 highest demand days. This will reduce the overall capacity requirement for the utility by the total program demand reduction, unless the difference between the 16 th highest day and the highest day is less than that amount. Figure 23 shows the load duration curve based on the daily system peaks. The difference between the highest day and the 16 th highest day is 570 MW in 2011, 716 MW in 2012, and 445 MW in 2013, for an average of 577 MW. This is the upper limit on the potential for capacity reduction for any program based on a maximum of 15 event days. This is well beyond the possible total load reduction for this program, given the per-customer impacts that we ve seen and any reasonable assumption about recruiting. We conclude that a maximum of 15 event days for this program is reasonable. The requirement of a minimum of 13 events was put in place to ensure that events are called. Because the PTR rate uses rebates to provide a price signal, there is no opportunity to save money if events are not called. Our understanding is that this requirement was put in place to ensure that PTR customers would have an opportunity for savings each year. This requirement will ensure that the program is used, and is not unduly burdensome, and so is reasonable. Range of 13 to 15 event days is reasonable. Figure 23. System Load Duration Curve Based on Daily Peaks 8,000 7,000 6,000 5,000 4,000 3,000 2011 2012 2013 2,000 1,000 0 However, beyond the operational considerations, even given a careful planning process to ensure that event days are called on only the hottest days, it is always possible that there may be an unexpected heat wave late in the season. With a firm maximum of 15 event days, this can create issues around the availability of the resource. While we recommend including a maximum number of event days under normal circumstances, which helps recruiting, we recommend that, if possible, Public Service include some sort of contingency for system emergencies. Language such as a maximum of 15 event days, unless additional event days EnerNOC Utility Solutions 49

Average kwh Consumption are needed for system emergencies, if acceptable to Public Service and to the Commission, would alleviate the risk of a system emergency late in the season. Consider allowing more than 15 event days in case of a system emergency. PTR Baseline Analysis One issue that has the potential to create customer dissatisfaction is the calculation of the PTR baseline, as it could be perceived as unfair. The goal of the PTR baseline is to estimate what the customer s peak period energy use would have been on the event day if an event had not been called. Each customer s rebate is then calculated based on the difference between this baseline and the customer s actual peak period energy use on the event day. There has been a great deal of controversy and discussion about the effects of different types of PTR baselines on the estimated savings. The biggest challenge is that residential customer energy use is highly variable across days, driven in part by temperature, but also driven by variation in customer activities. Because of this variability, estimating what the customer would have used in the absence of an event is difficult. Many utilities have taken a similar approach to that of Public Service, using a baseline that is the average of the 5 highest energy use days in the previous 10 non-event weekdays. Event days tend to be called on the hottest days, which are also days with higher use, so by averaging the highest use days, the baseline is more representative of usage on a hot day. Figure 24 illustrates the average on-peak energy use for the control group customers across a broad range of high temperatures. The smoothed line follows the rough trend of the average energy use for each temperature. Figure 24. Impact of Temperature on Energy Use 11 Residential On-peak Consumption and Temperature 10 9 8 7 6 5 4 3 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 105 110 Temperature ( F) 50 EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results There are two important things to note in this graph: Higher Temperatures Yield Higher Energy Use: Days with higher temperatures clearly tend to have higher energy use. There is Notable Variation in Energy Use at each Temperature: Even with this clear trend, notable variation exists between average energy use on days with the same high temperature. In addition to the variation shown here between days for an average customer, there is also more variation in individual customer energy use across days. There are two implications to these trends: Mismatch Between Rebates and Action Taken: The wide variation in energy use means that some customers will have lower energy use on event days without taking any action, and some will have higher energy use, and even if they take action, may not reduce their usage to below the baseline amount. This presents the possibility that some customers who take no action will receive a rebate, and some customers who take action will not get a rebate. This is an inherent characteristic of a PTR rate, created by customer variation and the one-sided nature of the rate. Across many events, if the differences are random, the variation will average out, but the one-sided nature means that the revenue does not average out. At times, the baseline will be too high, and sometimes too low. Event Day Baselines Consistently Low: Since most event days are on hot days, some or all of the 5 highest days in the previous 10 will almost certainly have lower temperatures than the event day, and therefore have lower on-peak energy use. The average of these five days will therefore be lower, with the estimated savings being less or even non-existent. Unlike the variation, this will not average out across days the baseline will consistently be too low, and the customer energy savings will consistently be underestimated. As mentioned above, differences due to customer variation are random, and will average out over the course of the summer. However, the rebates will be paid if usage is lower, but this is not offset if usage is higher. There are three points Public Service should consider if they intend to offer the PTR rate to a broad population: Recover the Cost by the Rate itself: There is a cost, created by the combination of the asymmetry of the rebate and the random variation in customer use. This cost should be recovered through the rate itself, by including analysis of the payout for random variation in the rate design process. Consider Offering the PTR Rate on a Strictly Opt-In Basis: Customers who opt-in to a PTR rate are more likely to take action to reduce energy use, and those who do take action will cause less revenue loss due to random variation. Tying participation to the use of an enabling technology such as an IHSD, a programmable communicating EnerNOC Utility Solutions 51

thermostat (PCT), or an AC cycling switch has also been shown to increase load response, and so would also minimize revenue loss. Consider Making the Rebates Based on Cumulative Energy Savings across the Summer: This would alleviate much of the revenue loss due to variation in energy use, as that variation would tend to average out across multiple events. However, this has the disadvantage that customers would not receive any actual rebate until the end of the summer, which may discourage people from taking actions to reduce energy use during events, especially later in the summer. Table 19 shows each of the event days from 2011 to 2013, the actual high temperature on that event day, the average temperature for the baseline period, and the difference between those two temperatures. When the baseline temperature is higher, savings tend to be overestimated. When the baseline temperature is lower than the event day, the savings tend to be underestimated. This is most notable early in the summer when the first hot days are being compared to a baseline of mild days. Notice that for the first three June events in 2012, all with temperatures over 100 F, the baselines are based on days with an average temperature at least 9 F lower than the event day. This resulted in PTR rebates for these days being smaller even for customers who reduced their use. When events are called on mild days, such as 9/8/2011, 9/14/2012, 7/16/2013, and 7/19/2013, the opposite is true the baseline is higher, so the PTR rebate amounts tend to be higher than would be expected. The closer the baseline temperature approaches the event day temperature the better the rebate represents savings attributable to the participant s actions rather than inherent differences in consumption based on weather. When baseline temperatures are closer to event day temperatures, rebates align better with actions taken by participants to reduce usage. 52 EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results Table 19. Comparison of Event Day Temperatures with Baseline Average Temperature Temperature ( F) Event Date Average of 5 Actual Event Day Baseline Days Difference 14-Jul-11 88 90.6 (2.6) 19-Jul-11 95 90.4 4.6 22-Jul-11 96 91.8 4.2 8-Aug-11 92 92.6 (0.6) 10-Aug-11 88 92.8 (4.8) 16-Aug-11 90 94.0 (4.0) 22-Aug-11 97 95.2 1.8 23-Aug-11 98 95.2 2.8 25-Aug-11 99 96.4 2.6 26-Aug-11 92 96.4 (4.4) 1-Sep-11 96 96.8 (0.8) 8-Sep-11 76 96.2 (20.2) 2011 Summer 92.3 94.0 (1.8) 18-Jun-12 100 90.6 9.4 22-Jun-12 102 90.4 11.6 25-Jun-12 105 90.4 14.6 27-Jun-12 97 94.0 3.0 2-Jul-12 101 97.6 3.4 13-Jul-12 96 99.0 (3.0) 20-Jul-12 101 97.0 4.0 23-Jul-12 100 97.0 3.0 1-Aug-12 94 97.4 (3.4) 8-Aug-12 96 97.0 (1.0) 14-Sep-12 77 93.0 (16.0) 2012 Summer 97.2 94.9 2.3 10-Jun-13 97 83.8 13.2 12-Jun-13 95 86.6 8.4 13-Jun-13 96 86.6 9.4 19-Jun-13 95 89.8 5.2 21-Jun-13 87 91.2 (4.2) 27-Jun-13 98 94.8 3.2 16-Jul-13 84 94.6 (10.6) 19-Jul-13 81 94.6 (13.6) 22-Jul-13 97 94.6 2.4 1-Aug-13 93 90.8 2.2 19-Aug-13 96 88.6 7.4 21-Aug-13 87 90.0 (3.0) 2013 Summer 92.2 90.5 1.7 As a result of this phenomenon, we also recommend that Public Service consider using a temperature adjustment to the 5-in-10-day baseline. This would involve an adjustment of the baseline up or down based on the difference between the event day temperature and the average temperature of the 5 days that go into the baseline. We would recommend creating a table of adjustment factors based on class average peak period energy use (as illustrated in EnerNOC Utility Solutions 53

Figure 24 above) and high temperature. These adjustment factors would be specific to the combination of event day and baseline average high temperatures. This would correct the quantifiable difference between the baseline as an estimate of event day energy use and the actual event day energy use, and improve the accuracy of the savings estimate. Consider a temperature adjustment to the 5-in-10-day baseline. Opt-In Versus Opt-Out The two phases of this pricing pilot were initially attempted to compare results between a pure voluntary opt-in approach for Phase I, and a mandatory opt-out approach for Phase II. During Phase II recruiting, customers were offered a choice between the time-varying rates and the standard tiered rates. However, due to implementation concerns, the opt-out approach was only in the messaging customers who did not make a choice were left on the tiered rate, meaning that in effect, Phase II was really opt-in as well. When deciding whether to offer a pricing program on an opt-in basis or on a default basis with an opt-out provision, there are several things to consider. They fall under two general areas, as follows: Customer Service Considerations: When customers are defaulted to a time-varying pricing rate, the vast majority of them do not actively opt-out, but simply stay on the new rate. Opt-out rates in the industry generally are less than 20%. Because of this, many customers will be on a rate that they did not choose and probably do not understand. This can cause customer service problems, especially with event-based pricing, when a customer may end up paying much more for their energy use. Savings Considerations: Customers who opt-in to a pricing program do so because they believe that they can save money. They are much more likely to actively change their energy use to shift load to less expensive times, and so will generally provide better average load reduction than customers who are defaulted to time-varying pricing. Of course, when customers are defaulted to a time-varying rate, some of those who go on the rate would have opted-in in a voluntary program. Those customers who would have opted in would likely provide most of the savings in an opt-out program, with the majority of the customers not taking any action, since they may not even know they are on a different rate. However, some customers that are defaulted on the rate may then take action to shift energy use, and if so, the total load shifted for an opt-out program may be greater than the load shifted for an opt-in program. Sacramento Municipal Utility District (SMUD) recently completed a pricing pilot where they compared opt-in and opt-out approaches in a robustly designed statistical experiment. They segmented their customer population, and implemented a pricing pilot with one segment offered the rates on an opt-in basis, and another segment defaulted onto the rates, but allowed to optout if they chose to. This was truly a default offering customers in the default segment were moved onto the time-varying rate unless they specifically asked not to be moved. The results of 54 EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results the first summer of the pilot were presented at the National Town Meeting on Demand Response in Washington, DC, on July 10, 2013. The SMUD pilot realized opt-in rates of 16%-19%, and opt-out rates of 4%-11% over the first year. The opt-in rates were higher than expected, and the opt-out rates were lower than expected, when compared with other studies across the industry. While the average savings for the opt-in customers were significantly higher than the average savings for the defaulted customers, the total savings were greater for the defaulted customer group, since they represented a much higher percentage of the customers recruited. The SMUD pilot is the only pilot that we are aware of that has done this direct comparison of an opt-in and a default/opt-out recruitment. It is important to note that SMUD offered only a CPP and a TOU rate they did not offer a PTR rate. As discussed above under PTR Baseline Considerations, PTR rates perform better and provide more real load reductions when offered on an opt-in basis. Public Service should carefully consider whether to offer any future pricing programs on an opt-in or a default opt-out basis. Comparisons with Saver s Switch Customers Public Service also offers another program, Saver s Switch (SS), which is designed to reduce customer peak energy use at the time of the system peak. This program involves an AC cycling switch that can be activated remotely, which limits the time that the customer s AC compressor operates. Customers who participate receive an annual payment for participation. The SS program has been evaluated independently, and provides a load reduction of 1.07 kw, or about 35% of the average SS customer load of 3.00 kw at the time of the peak. The SS program provides a much higher kw reduction per customer than any of the rates in the Pricing Pilot. There are several reasons for this: 1. All SS Customers have Central AC: The baseline load for SS customers is 3.00 kw, meaning there is much more load available to reduce for these customers. The pricing pilot rates include a mix of customers with and without Central AC. 2. SS Involves Direct Control: The customer does not need to take any action, or even be aware of SS event, in order for savings to be realized. The customer does not have control or choice in their response. For the Pricing Pilot customers, only the few customers with IHSDs have the possibility of an automated response, and with that, they still could choose to override the event, and pay more to stay cool during an event. Direct Load Control (DLC) programs such as SS are fundamentally different from pricing programs, both in how they operate and in what types of customers find them attractive. DLC programs appeal to customers who do not want any involvement in savings energy they are willing to give up some control in exchange for a payment. Pricing programs appeal to more involved customers, those that want to better understand how their energy use affects their bill, customers who want to be able to choose the right combination of comfort and cost. EnerNOC Utility Solutions 55

Because of the fundamental differences between DLC and pricing programs, there is value in offering both. Customers who want control will sign up for a pricing program, but would be unlikely to participate in DLC. Other customers who want to avoid any hassles would be more likely to participate in a DLC program, but would be less interested in a pricing program. In a service territory like that of Public Service, customers without Central AC have no option to participate in the SS program, so offering a pricing program gives those customers an opportunity to save money and provide a resource to Public Service. There is value in offering pricing programs in addition to Saver s Switch. Program Comparison In general, the results of the Pricing Pilot are consistent with what we have seen in other timevarying pricing pilots in the industry. The largest load reductions were for the CPP rate, on summer event days, ranging from 22%-30% for Phase I. PTR customers also reduced during summer events, though at a lower percentage than CPP customers across the three years. On summer non-event days, CPP customers also reduced on-peak energy the most, a bit more than the TOU group. The PTR customers also reduced, even though they had no price incentive to do so. This is consistent with other pricing pilots across the country. Customers on a CPP rate with an underlying TOU tend to save more than customers on a TOU without any sort of event calling, even given similar prices, such as we have here. This is probably due to increased awareness of energy use based on the frequent reminders of events, and because customers make behavior or thermostat setting changes to avoid the very high CPP prices, and some of those changes carry through to all weekdays. Relative savings by rate type are consistent with other pricing pilots. While it is useful and valid to compare the relative level of savings between this pilot and others, the actual kwh and percentage savings are not necessarily comparable. One big difference between the Public Service Pricing Pilot and other recent pricing experiments, such as Oklahoma Gas & Electric (OG&E), Baltimore Gas & Electric (BG&E), and Sacramento (SMUD) is that those pilots were based on customers that all or nearly all had CAC, and were in much hotter climates than Boulder. This affects the baseline load, which is much higher, and the load available for reduction. This makes it impractical to expect the same level of savings. Characteristics of the rate make direct comparisons difficult, including the structure of the rate and particularly the price levels both of the standard rate customers paid before going on the time-varying rate, and the price differentials between the time periods. Less influential but still important are variable such as income and other customer demographics, and the prevalence of certain things such as swimming pools and spas, electric heat, and electric water heat. Lastly, the use of an enabling technology in conjunction with event-based pricing can have a profound influence on the savings. With so few IHSDs installed on pricing customers, it is difficult to compare with other programs that have much higher percentages with enabling technologies. Magnitudes of savings are difficult to compare. 56 EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results That said, as expected, the savings for this pricing pilot are generally lower than other recent pilots with enabling technologies, but in some cases higher than those without enabling technologies. OG&E reported seeing impacts of 40-50% for the highest priced period with PCTs, but only 15-17% without PCTs. The SMUD pilot found comparable percentage savings to Public Service, with 22-26% for CPP, and 10-13% for TOU, though the presence of an IHD made only a small difference in load impact. Savings are generally lower than other recent pilots with enabling technologies. Program Limitations The time-varying rates studied in the SmartGridCity Pricing Pilot require measurement of energy use by time period. Interval data, as collected by smart meters, is sufficient to bill any of the rates. However, standard billing meters, read monthly, are not sufficient. The requirements for data collection are different for the TOU rate as compared with CPP and PTR. The TOU schedule is predetermined. While it may vary by season, it is not dynamic. The CPP and PTR have a dynamic component events are called when needed, and cannot be anticipated. Because of this, CPP and PTR require on-peak and off-peak energy be measured for every weekday separately, so that data for billing are available for whichever days are called as event days. This can only be realistically met by a meter that collects interval data, such as a smart meter, or a meter with a built in recorder under glass (RUG), commonly used for load research. However, there are other options for the TOU rate. Multiple register meters can be programmed to collect data for different predetermined time periods, and that data can then be collected in a conventional way. This tends to work well, but does have potential drawbacks. Because the meters must be preprogrammed and then installed in the field, the accuracy of the data is dependent on the accuracy of the internal clock. Should that clock drift, as they have potential to do, the data for peak periods will not be accurate. Also, any changes, such as the change to Daylight Savings Time that happened a few years back, or any holidays that are added or changed, will require that all meters in the field be reprogrammed. If there are many customers on the TOU rate, this reprogramming effort could be herculean. So while there are other options for TOU implementation, there are risks and potential problems associated with those options. EnerNOC Utility Solutions 57

Chapter 6 Key Findings and Recommendations Key Findings The following is a summary of key findings from the Public Service Pricing Pilot: Participants Responded to Price: Residential customers participating in the Pricing Pilot responded to price signals, reducing their energy consumption during on-peak periods when prices were higher. Results Generally Consistent with Other Pricing Pilots: In most cases, the relative results found here were generally consistent with other pricing pilots in the industry. The impacts tended to be less than other pricing pilots, due to the lower saturation of central air-conditioning (CAC) in Public Service Company s territory and due to the low number of in-home smart devices. But the relationship between the impacts for different rates and for different time periods was consistent with other pricing pilots. On-Peak Summer Event Load Reductions Reached 31%: Event-driven rates, PTR and CPP, provided the highest load reduction of the pilot on summer event days. CPP Phase I participants reduced loads by 22-29%, which represents average on-peak demand reductions of 0.22-0.28 kw, while PTR Phase I participants had load reductions of 8-14%, or 0.09-0.14 kw. Summer Events Had Largest Impacts: For CPP, the summer event day load reductions were higher than the summer non-event day load reductions. This finding is consistent with similar event-driven rates offered elsewhere and is due to a combination of two factors: event days are hotter than non-event days, and prices are higher on event days. The summer event day impacts were also greater than the winter event day impacts, mainly due to the lack of CAC load during the winter. TOU Participants Saved Energy During On-Peak: TOU provided load reduction during the on-peak period on summer weekdays. Load reductions for Phase I participants averaged 5-9% across the three years of the pilot, representing average onpeak demand reductions of 0.05-0.12 kw. Off-peak energy use increased by a smaller percentage in most cases, as expected due to lower off-peak rates, and as seen in most TOU rates across the industry. CPP and PTR Participants Saved Energy at all Times: In general, the CPP and PTR participants saved energy at all times, not just during on-peak periods. This result is consistent with other findings in the industry for event-driven rates, as these rates usually lead to an increased awareness of overall energy use. When customers are more aware of their energy use, they use less energy. Non-Summer Impacts were Small: The impacts during the non-summer season, for both events and non-events, were consistently much lower than the summer impacts in almost all cases. This difference is due to lower loads and less discretionary energy use during the winter, spring, and fall. EnerNOC Utility Solutions 59

Impacts Were Consistent Through Duration of Event: The magnitude of savings was fairly consistent throughout events for CPP and PTR participants. This consistency is typical of pricing programs that are not primarily driven by an enabling technology, which is the case for nearly all customers in the pilot. When customer behavior is driving the load reduction, the actions taken tend to result in more consistent savings across the event. Performance of Phase I Tended to Exceed Phase II: The results indicated differences between the Phase I customers, who were recruited on an opt-in basis, and the Phase II customers, who were recruited in a pseudo-opt-out manner. Phase I load reductions were higher than Phase II, and tend to drop off less across the three years. Care must be taken, however, with these results, since the recruitment for Phase II was not consistent and was not truly opt-out. In addition, since the Phase I customers were recruited before the Phase II customers, Phase I would include those customers most interested in a time-varying rate who may be more inclined to respond. PTR, CPP, and TOU Customers Reduce Load during Events less than Saver s Switch Customers: Xcel s current Saver s Switch (SS) program yields a higher load reduction per customer than any of the time-varying rates in the pilot, at about 1.07 kw per participant during events, against a baseline of 3.00 kw for the average SS customer. 8 However, unlike the Pilot customers, which include a mix of those with and without central air-conditioning (CAC), SS customers all have CAC. Therefore, SS customers have dramatically higher baseline loads and have more discretionary load available. PTR Baseline Methodology Could be Improved: The PTR baseline used to estimate savings for the payment of the rebate resulted in biased savings estimates. It underestimated savings on hot event days and overestimated the savings on mild event days. A weather-adjusted PTR baseline based on the difference between the average temperature on the days used in the baseline and the event day temperature would provide a better estimate of savings and a more accurate rebate payment. Load Impacts Lessen over the Three Years: In nearly all cases, the load impacts for the second and third years were less than the load impacts for the first year. This may indicate that there is a drop off in persistence, but there were also dramatic differences in the temperatures for the three summers, which also influenced the savings. However, the regression analysis, using normal weather for all three years, showed lower savings in the second and third years. This drop in savings has not generally been seen in other studies in the industry, and may be due to a decrease in level of participant engagement. Devices Provide Anecdotal Results: There were not enough in-home smart devices (IHSDs) installed on pricing pilot customers to provide results that could be generalized to a broader population. However, using the regression analysis, we could estimate the impacts for those customers with IHSDs, few though they were. The results are 8 http://www.xcelenergy.com/staticfiles/xe/marketing/files/co-dsm-2012-2013-biennial-plan-rev.pdf, page 284. 60 EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results reflective only of the individual customers who had IHSDs. For those with IHSDs, there were higher savings during summer events, presumably driven by the control of the thermostat setpoint by the devices. This result should be considered anecdotal, and not indicative of impacts for a broader roll-out. The Current Event Structure is Appropriate: The current event window of 2:00 to 8:00 in the afternoon is appropriate, and the correct duration. The restriction of calling 13-15 events is reasonable, but should be modified, if possible, to allow for additional events in the case of a system emergency. Customers Understand Price Signals: Based on the relative savings in the various periods, customers appeared to understand the price signals. The only exception is that CPP and PTR customers used less in off-peak periods, where lower prices would normally result in increased energy use. As explained above, this is likely due to increased awareness. Recommendations We have the following recommendations regarding the TOU Pilot if Public Service continues offering time varying rates. Keep Event Structure: We recommend that Public Service keep the current on-peak and event structure, with events and on-peak periods running from 2:00-8:00. 15 events seems an appropriate number, but if possible, we recommend a caveat on the maximum to be able to call additional events beyond 15 in case of a system emergency. Including a minimum of 13 events ensures that events are called, gives customers a chance to realize savings, minimizes free-ridership by some customers, and allows for estimation of impacts. Analyze Payout for New PTR Rate: Public Service should analyze the payout for random variation in the rate design process for a new PTR rate, and recover that cost from customers who are on the rate. Offer only Opt-In PTR: We recommend offering the PTR rate on a strictly opt-in basis, preferably combined with some sort of an enabling control technology such as a PCT. Improve PTR Baseline: We strongly recommend weather adjusting the PTR baseline based on the difference between the average temperature on the days used in the baseline and the event day. Continue Customer Engagement: We recommend continuing and enhancing efforts to engage customers who are on the rates, providing them feedback on their energy use changes (the same concept as the post-event reporting), and suggestions about how to benefit from the rate. The feedback must be credible and specific to the customer in order for it to be effective. EnerNOC Utility Solutions 61

Target Customers with Air Conditioning: If possible under current regulation, Public Service should consider offering time-varying rates only to those customers with air conditioning. 62 EnerNOC Utility Solutions

Appendices

A. Appendix A 2011 Difference of Differences Results Impact Results Table A1 through Table A3 summarize the difference of differences results for 2011. Table A1 and Table A2 present the on-peak kw reduction results for Phase I and Phase II, respectively, while Table A3 presents the kwh changes results for Phase I. (Because Phase II began in June 2011, we only have data for a partial year. Therefore, we have not included annual kwh changes for Phase II.) The kw reduction tables include baseline average on-peak kw, average on-peak kw reduction, and percent reduction for each rate and day type. Similarly, the kwh changes tables include baseline kwh, kwh change, and percent change for each rate and day type. The convention we followed in the tables is that the kw impacts are reductions, so a positive value means less energy use. Since the kwh impacts can go both directions, they are shown as changes, with negative values meaning lower energy use during the period and higher values meaning higher energy use. The Phase II results presented in these 2011 tables differ from those presented in the report filed with the Commission on December 16, 2011. These new tables reflect a matched control group, whereas the previous results were developed with a randomly assigned control group. The revised results are more accurate and consistent with the other years. We also changed the analysis of all customers, both Phase I and Phase II, across all three years, to provide more stable estimates and to account for Saver s Switch (SS) participation both before and after the pilot. We made the following changes to the analysis: We removed SS event days from the pre-treatment period. Because some of the participants had been SS customers prior to signing up for the pilot, the load reduction during a SS event could inappropriately affect the estimate of savings for the pilot. Instead of using a single similar day from the pre-treatment period for each event day, we used the average of several similar days, grouped by temperature range. This made the impact estimates less sensitive to variations in load on particular days. We included all days in the pre-treatment period to cover a broader range of temperatures. These improvements to the analysis did change the results somewhat from what was described in the preliminary results, but did not dramatically alter the nature of the results. EnerNOC Utility Solutions A-1

Table A1. kw Reduction by Rate, 2011, Phase I Baseline On-Peak Average kw Average On-Peak kw Reduction Percent Reduction CPP Average Non-Summer Non-Event Weekday 0.87 0.04 4.33% Average Summer Non-Event Weekday 0.86 0.14 16.72% Average Summer Event Day 0.93 0.27 28.78% PTR Average Non-Summer Non-Event Weekday 0.85 0.02 1.82% Average Summer Non-Event Weekday 0.93 0.05 5.35% Average Summer Event Day 1.05 0.14 13.60% TOU NSS Average Non-Summer Weekday 0.86 0.01 0.67% Average Summer Weekday 0.89 0.08 9.48% TOU SS Average Non-Summer Weekday 1.01 0.02 2.42% Average Summer Weekday 1.52 0.12 7.61% The estimates in Table A1 show the following trends for Phase I on-peak kw reduction in 2011: Impacts are higher in the summer Impacts are higher on event days (during 2011, all event days were during the summer) Impacts are higher for CPP participants than for PTR participants For summer non-event weekdays, CPP impacts are higher than TOU impacts; even though there is no price difference on non-event days, there are small PTR savings on those days For non-summer non-event weekdays, TOU impacts are lower than for CPP participants Impacts are slightly lower in the summer for SS participants than for non-ss (NSS) participants, and negligible in the winter for the NSS participants A-2 EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results Table A2. kw Reduction by Rate, 2011, Phase II Baseline On-Peak Average kw Average On-Peak kw Reduction Percent Reduction CPP Average Non-Summer Non-Event Weekday 0.90-0.03-3.17% Average Summer Non-Event Weekday 0.92 0.11 11.62% Average Summer Event Day 1.03 0.27 26.32% PTR Average Non-Summer Non-Event Weekday 0.82 0.00-0.51% Average Summer Non-Event Weekday 0.88 0.02 2.66% Average Summer Event Day 1.00 0.12 12.46% TOU NSS Average Non-Summer Weekday 0.94 0.01 0.76% Average Summer Weekday 1.04 0.06 5.77% TOU SS Average Non-Summer Weekday 1.02 0.04 3.88% Average Summer Weekday 1.52 0.14 9.13% The estimates in Table A2 show the following trends for Phase II on-peak kw reduction in 2011: Impacts are higher in the summer Impacts are higher on event days Impacts are higher for CPP participants than for PTR participants For summer non-event weekdays, TOU SS impacts are higher than for PTR participants, but lower than for CPP participants For non-summer non-event weekdays, TOU impacts are higher than for CPP and PTR participants EnerNOC Utility Solutions A-3

Table A3. Annual kwh Savings by Rate, 2011, Phase I Baseline kwh kwh Change Percent Change CPP On-Peak Consumption 1,302-112 -8.61% Off-Peak Consumption 5,435-200 -3.68% Overall Consumption 6,738-312 -4.63% PTR On-Peak Consumption 1,317-45 -3.44% Off-Peak Consumption 5,313-172 -3.24% Overall Consumption 6,630-217 -3.28% TOU NSS On-Peak Consumption 1,314-50 -3.77% Off-Peak Consumption 5,305 178 3.35% Overall Consumption 6,619 128 1.94% TOU SS On-Peak Consumption 1,789-82 -4.60% Off-Peak Consumption 6,721 93 1.38% Overall Consumption 8,510 11 0.12% The estimates in Table A3 show the changes in kwh energy use across the whole year. Not that for consistency, here and in the other kwh tables, we summarize by off-peak and on-peak consumption across all weekdays, including event days. This means that the CPP and PTR onpeak consumption includes both event weekdays and non-event weekdays, even though the pricing structures of those rates are difference on event days and non-event days. We see the following trends for Phase I annual kwh savings in 2011: Impacts in both periods and in total are largest for CPP participants TOU participants use more energy during off-peak periods relative to the baseline, more than offsetting their reduction during peak periods Overall consumption for TOU participants is slightly higher than the baseline On-peak impacts for PTR and TOU participants are comparable Impacts for NSS and SS participants are comparable Based on pricing, we would expect that TOU and CPP participants decrease energy use during on-peak periods when prices are higher, and increase energy use during off-peak periods when prices are lower. We would also expect PTR customers not to change their energy use other than during events, when they can receive a rebate. As is commonly seen in the industry, however, when customers become engaged and aware of their energy use, they often decrease usage during all periods, including those when they pay less. The CPP and PTR customers, A-4 EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results who are reminded about being in the program by event notification, appear to be showing this awareness reduction. The TOU customers do not. Load Profiles The subsections below contain the sample load profiles from the 2011 analysis. We divide the profiles into three groups: 1) Phase I Non-Summer; 2) Phase I Summer; and 3) Phase II Summer. (Note: Phase II includes only the summer data, as recruitment was in progress during the winter months.) In each group, we include separate graphs for each rate type and for nonevent and event days. Because TOU participants do not receive special event day pricing or notification, we do not include event day graphs for them. Each graph shows two lines, one being the adjusted average control group load shape, and the other being the average participant group load shape. (The abbreviation CTL on the graphs refers to the control group.) The adjusted savings estimate is the difference between these two lines. For the non-event days, we created the non-summer weekday graphs by averaging data across the non-summer period. Similarly, we created the non-event summer weekday graphs by averaging data across the summer period. The event-day graphs represent the average data for all event days in the given period (non-summer or summer). The graphs for 2011 differ from those presented in the report filed with the Commission on December 16, 2011 in two ways. First, in the graphs included in the former report we displayed one representative non-summer weekday (January), one representative summer weekday (July), and one example of a summer event day (August 23, 2011) for each rate, rather than presenting average load profiles across the seasons and events as we have done here. Second, as in the Phase II tables presented above, the data in the Phase II graphs now reflect analysis with a matched control group instead of the original randomly-assigned control group. Phase I Non-Summer Analysis As the graphs in Figure A1 through Figure A3 indicate, the control group and participant group for each rate appear well matched. This is demonstrated by the similarity in load shape between the control group curves (designated CTL) and the pricing plan curves. The CPP participant curve is sometimes below and sometimes above that of the control group, showing only very small load reductions during part of the on-peak period. The PTR participant curve is similar, showing negligible on-peak load reductions, energy savings overnight, and higher energy usage during the morning hours. The TOU customers with SS appear to be higher energy users than those without SS, even in winter, though the general shape of the load is the same. The TOU customers with SS decrease usage during the on-peak hours more than TOU customers without SS. The differences between the participant and control groups in all of these load graphs is very small. EnerNOC Utility Solutions A-5

kw kw Figure A1. Phase I, 2011, Non-Summer, Non-Event Weekday, CPP 1.2 Phase I: 2011 Non-Summer Non-Event Weekday - CPP 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment Figure A2. Phase I, 2011, Non-Summer, Non-Event Weekday, PTR 1.2 Phase I: 2011 Non-Summer Non-Event Weekday - PTR 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment A-6 EnerNOC Utility Solutions

kw SmartGridCity Pricing Pilot Impact Evaluation Results Figure A3. Phase I, 2011, Non-Summer, Weekday, TOU 1.4 Phase I: 2011 Non-Summer Weekday - TOU 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending SS - Control SS - Treatment non-ss - Control non-ss - Treatment Phase I Summer Analysis The summer graphs in Figure A4 through Figure A8 show a clearer price response than the non-summer graphs. This is probably due to the higher price differential as well as the presence of CAC for some customers. The CPP customers appear to be reducing load at all times, with higher reductions during on-peak times. On the event day, the CPP customers reduce load dramatically during the on-peak period, with higher savings than at any other time and for any other rate group. The PTR customers show a very small reduction in all hours on the non-event weekdays, which may be the result of increased awareness of energy use. However, they show a much larger reduction on the event days, when the rebate applies. TOU customers, both with and without SS, show a reduction during the on-peak period on summer weekdays. EnerNOC Utility Solutions A-7

kw kw Figure A4. Phase I, 2011, Summer, Non-Event Weekday, CPP 1.2 Phase I: 2011 Summer Non-Event Weekday - CPP 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment Figure A5. Phase I, 2011, Summer, Non-Event Weekday, PTR 1.2 Phase I: 2011 Summer Non-Event Weekday - PTR 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment A-8 EnerNOC Utility Solutions

kw kw SmartGridCity Pricing Pilot Impact Evaluation Results Figure A6. Phase I, 2011, Summer, Weekday, TOU 1.8 1.6 1.4 1.2 1.0 Phase I: 2011 Summer Weekday - TOU 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending SS - Control SS - Treatment non-ss - Control non-ss - Treatment Figure A7. Phase I, 2011, Summer, Event Day, CPP 1.2 Phase I: 2011 Summer Event Day - CPP 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment EnerNOC Utility Solutions A-9

kw Figure A8. Phase I, 2011, Summer, Event Day, PTR 1.4 Phase I: 2011 Summer Event Day - PTR 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment Phase II Summer Analysis As Figure A9 through Figure A13 indicate, the summer results for Phase II are fairly similar to the results for Phase I. A-10 EnerNOC Utility Solutions

kw kw SmartGridCity Pricing Pilot Impact Evaluation Results Figure A9. Phase II, 2011, Summer, Non-Event Weekday, CPP 1.2 Phase II: 2011 Summer Non-Event Weekday - CPP 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment Figure A10. Phase II, 2011, Summer, Non-Event Weekday, PTR 1.2 Phase II: 2011 Summer Non-Event Weekday - PTR 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment EnerNOC Utility Solutions A-11

kw kw Figure A11. Phase II, 2011, Summer, Weekday, TOU 1.8 1.6 1.4 1.2 1.0 Phase II: 2011 Summer Weekday - TOU 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending SS - Control SS - Treatment non-ss - Control non-ss - Treatment Figure A12. Phase II, 2011, Summer, Event Day, CPP 1.4 Phase II: 2011 Summer Event Day - CPP 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment A-12 EnerNOC Utility Solutions

kw SmartGridCity Pricing Pilot Impact Evaluation Results Figure A13. Phase II, 2011, Summer, Event Day, PTR 1.2 Phase II: 2011 Summer Event Day - PTR 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment EnerNOC Utility Solutions A-13

B. Appendix B 2012 Difference of Differences Results Impact Results Table B1 through Table B4 summarize the difference of differences results for 2012. Table B1 and Table B2 present the on-peak kw reduction results for Phase I and Phase II, respectively, while Table B3 and Table B4 present the kwh savings results for Phase I and Phase II. The format for the tables is consistent with the format discussed in Appendix A for the 2011 results. The only difference is that Table B1 and Table B2 include results for non-summer event days; there were no non-summer events during the 2011 analysis year. We changed the analysis of all customers, both Phase I and Phase II, across all three years, to provide more stable estimates and to account for Saver s Switch (SS) participation both before and after the pilot. We made the following changes to the analysis: We removed SS event days from the pre-treatment period. Because some of the participants had been SS customers prior to signing up for the pilot, the load reduction during a SS event could inappropriately affect the estimate of savings for the pilot. Instead of using a single similar day from the pre-treatment period for each event day, we used the average of several similar days, grouped by temperature range. This made the impact estimates less sensitive to variations in load on particular days. We included all days in the pre-treatment period to cover a broader range of temperatures. These improvements to the analysis did change the results somewhat from what was described in the preliminary results, but did not dramatically alter the nature of the results. EnerNOC Utility Solutions B-1

Table B1. kw Reduction by Rate, 2012, Phase I Baseline On-Peak Average kw Average On-Peak kw Reduction Percent Reduction CPP Average Non-Summer Non-Event Weekday 0.88 0.09 10.23% Average Summer Non-Event Weekday 0.86 0.14 15.92% Average Summer Event Day 1.06 0.28 26.16% Average Non-Summer Event Day 0.98 0.23 23.77% PTR Average Non-Summer Non-Event Weekday 0.86 0.04 4.11% Average Summer Non-Event Weekday 0.96 0.04 3.85% Average Summer Event Day 1.24 0.10 8.09% Average Non-Summer Event Day 0.88 0.04 4.81% TOU NSS Average Non-Summer Weekday 0.87 0.04 4.25% Average Summer Weekday 0.95 0.07 7.20% TOU SS Average Non-Summer Weekday 0.98-0.01-1.48% Average Summer Weekday 1.63 0.10 5.88% The estimates in Table B1 show the following trends for Phase I on-peak kw reduction in 2012: Impacts are higher in the summer Impacts are higher on event days CPP event day impacts are comparable for summer and non-summer events, though they tend to be slightly higher for summer events Impacts are higher for CPP participants than for PTR participants For non-summer non-event weekdays, TOU NSS impacts are higher than for PTR and TOU SS participants, but lower than for CPP participants For summer non-event weekdays, TOU SS and NSS impacts are comparable B-2 EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results Table B2. kw Reduction by Rate, 2012, Phase II Baseline On-Peak Average kw Average On-Peak kw Reduction Percent Reduction CPP Average Non-Summer Non-Event Weekday 0.92 0.03 2.97% Average Summer Non-Event Weekday 0.98 0.10 10.34% Average Summer Event Day 1.22 0.29 23.37% Average Non-Summer Event Day 0.97 0.13 13.63% PTR Average Non-Summer Non-Event Weekday 0.84 0.01 0.99% Average Summer Non-Event Weekday 0.95 0.03 2.99% Average Summer Event Day 1.20 0.10 8.44% Average Non-Summer Event Day 0.87 0.03 2.94% TOU NSS Average Non-Summer Weekday 0.95 0.03 3.20% Average Summer Weekday 1.15 0.06 5.39% TOU SS Average Non-Summer Weekday 0.99 0.01 1.30% Average Summer Weekday 1.60 0.10 6.52% The estimates in Table B2 show the following trends for Phase II on-peak kw reduction in 2012: Impacts are higher in the summer Impacts are higher on event days Impacts are higher for CPP participants than for PTR participants For summer non-event weekdays, TOU impacts are higher than for PTR participants, but lower than for CPP participants For non-summer non-event weekdays, TOU NSS impacts are higher than for PTR participants Non-summer impacts for SS participants are lower than for NSS participants On the whole, the 2012 Phase I and Phase II impacts are comparable. However, Phase I impacts appear to be a little larger for CPP and PTR participants. EnerNOC Utility Solutions B-3

Table B3. Annual kwh Savings by Rate, 2012, Phase I Baseline kwh kwh Change Percent Change CPP On-Peak Consumption 1,333-172 -12.88% Off-Peak Consumption 5,621-402 -7.15% Overall Consumption 6,954-574 -8.25% PTR On-Peak Consumption 1,361-58 -4.30% Off-Peak Consumption 5,462-326 -5.97% Overall Consumption 6,823-385 -5.64% TOU NSS On-Peak Consumption 1,353-72 -5.30% Off-Peak Consumption 5,437 83 1.53% Overall Consumption 6,790 11 0.17% TOU SS On-Peak Consumption 1,810-36 -1.98% Off-Peak Consumption 6,674 134 2.01% Overall Consumption 8,484 98 1.16% The estimates in Table B3 show the following trends for Phase I annual kwh savings in 2012: We again see reductions in the off-peak period for PTR and CPP, probably due to increased awareness of energy use Impacts are largest for CPP participants TOU participants use more energy during off-peak periods relative to the baseline, enough to more than offset their reduction during the on-peak period Overall consumption for TOU NSS participants is slightly higher than the baseline B-4 EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results Table B4. Annual kwh Savings by Rate, 2012, Phase II Baseline kwh kwh Change Percent Change CPP On-Peak Consumption 1,433-94 -6.58% Off-Peak Consumption 6,032-27 -0.44% Overall Consumption 7,465-121 -1.62% PTR On-Peak Consumption 1,330-28 -2.10% Off-Peak Consumption 5,337-165 -3.09% Overall Consumption 6,667-193 -2.89% TOU NSS On-Peak Consumption 1,537-62 -4.05% Off-Peak Consumption 6,159-101 -1.64% Overall Consumption 7,697-163 -2.12% TOU SS On-Peak Consumption 1,805-68 -3.75% Off-Peak Consumption 6,639 39 0.58% Overall Consumption 8,444-29 -0.34% The estimates in Table B4 show the following trends for Phase II annual kwh savings in 2012: We again see reductions in the off-peak period for PTR and CPP, and also here for TOU NSS, probably due to increased awareness of energy use On- and off-peak savings are comparable for PTR participants on a percentage basis Impacts are largest for CPP participants TOU SS participants use more energy during off-peak periods relative to the baseline, but not enough to offset the reduction during the on-peak period On-peak savings are slightly larger for TOU participants than for PTR participants In comparison with Phase I, the Phase II kwh savings impacts are smaller for CPP and PTR participants. Also, in Phase II the TOU NSS participants showed savings during the off-peak, while in Phase I the participants energy consumption increased in the off-peak. Load Profiles The subsections below contain the sample load profiles from the 2012 analysis. We divide the profiles into four groups: 1) Phase I Non-Summer; 2) Phase I Summer; 3) Phase II Non- Summer; and 4) Phase II Summer. In each group, we include separate graphs for each rate type and for non-event and event days. Because TOU participants do not receive special event day pricing or notification, we do not include event day graphs for them. Each graph shows two lines, one being the adjusted average control group load shape, and the other being the EnerNOC Utility Solutions B-5

average participant group load shape. (The abbreviation CTL on the graphs refers to the control group.) The adjusted savings estimate is the difference between these two lines. For the non-event days, we created the non-summer weekday graphs by averaging data across the non-summer period. Similarly, we created the non-event summer weekday graphs by averaging data across the summer period. The event-day graphs represent the average data for all event days in the given period (non-summer or summer). Phase I Non-Summer Analysis Figure B1 through Figure B5 illustrate how Phase I participants responded during the nonsummer months. The event day graphs represent averages across four non-summer event days: 10/4/2011, 11/1/2011, 12/1/2011, and 5/18/2012. As the graphs show, the control group and participant group for each rate appear well matched. This is demonstrated by the similarity in load shape between the control group curves and the pricing plan curves. The CPP participant curve is consistently below that of the control group, particularly during the on-peak hours. From this we infer that on average CPP customers are responding to on-peak pricing signals and reducing load. Additionally this energy savings seems to extend into the offpeak hours where a slight decrease in energy can be seen, even though there is no price disadvantage to using energy off-peak. This may be a result of customers increased awareness of energy use in their homes. PTR customers also appear to save energy over the control group on non-event days, but not as dramatically during the on-peak period. This intuitively makes sense as the PTR customers do not receive the price signal during peak times, but are probably more energy conscious over the general population as they have been exposed to the pricing plan. The TOU customers with Savers Switch (SS) appear to be higher energy users than those without SS, even in winter, though the general shape of the load is the same. The TOU customers without SS appear to be responding to the TOU price by increasing their usage relative to the control group during the off-peak time, and decreasing it during the peak times. While the effect is not as strong with the TOU customers with SS, the two lines are nearly the same during the on-peak period, but the participants use more during the daytime off-peak times. The increased usage during the off-peak time may be in response to the lower prices that TOU customers see during those off-peak times. Events are called only for customers on the CPP and PTR rate. Customers on the CPP rate show a much more noticeable reduction during the on-peak period during events than on nonevent days. This reflects the response to the much higher prices for the CPP during those periods. PTR customers show a reduction though it is a smaller reduction than for the CPP customers It also appears to last for the whole day, including the off-peak hours as well, though the reduction during the peak period is somewhat greater. B-6 EnerNOC Utility Solutions

kw kw SmartGridCity Pricing Pilot Impact Evaluation Results Figure B1. Phase I, 2012, Non-Summer, Non-Event Weekday, CPP 1.2 Phase I: 2012 Non-Summer Non-Event Weekday - CPP 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment Figure B2. Phase I, 2012, Non-Summer, Non-Event Weekday, PTR 1.2 Phase I: 2012 Non-Summer Non-Event Weekday - PTR 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment EnerNOC Utility Solutions B-7

kw kw Figure B3. Phase I, 2012, Non-Summer, Weekday, TOU 1.4 Phase I: 2012 Non-Summer Weekday - TOU 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending SS - Control SS - Treatment non-ss - Control non-ss - Treatment Figure B4. Phase I, 2012, Non-Summer, Event Day, CPP 1.4 Phase I: 2012 Non-Summer Event Day - CPP 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment B-8 EnerNOC Utility Solutions

kw SmartGridCity Pricing Pilot Impact Evaluation Results Figure B5. Phase I, 2012, Non-Summer, Event Day, PTR 1.2 Phase I: 2012 Non-Summer Event Day - PTR 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment Phase I Summer Analysis Figure B6 through Figure B10 illustrate how Phase I participants responded during the summer months. The event day graphs represent averages of the eleven 2012 summer event days. Similar to what was seen during the non-summer period, CPP customers appear to be reducing load at all times with greater reductions during on-peak periods. The summer graphs indicate a clearer price response than the non-summer graphs. This is likely due to the higher price differential as well as the presence of CAC for some customers. When we look at the event day, we see the CPP customers reduce load dramatically during the on-peak period. These savings are higher than at any other time and for any other rate group (except for the same group for Phase II). In contrast to CPP customers, the PTR group shows a very small reduction in all hours on the non-event weekdays. This is in line with expectations as there is no price signal driving savings for non-event weekdays. During event weekdays when the opportunity to earn a rebate applies, a much larger energy reduction is seen specifically during the on-peak hours. TOU customers, both with and without SS, also show a reduction during the peak period. We assume this reduction is driven by the higher on-peak price. Because the TOU customers do not receive any special pricing or notification of event days, event day graphs for the TOU customers are not included. EnerNOC Utility Solutions B-9

kw kw Figure B6. Phase I, 2012, Summer, Non-Event Weekday, CPP 1.2 Phase I: 2012 Summer Non-Event Weekday - CPP 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment Figure B7. Phase I, 2012, Summer, Non-Event Weekday, PTR 1.2 Phase I: 2012 Summer Non-Event Weekday - PTR 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment B-10 EnerNOC Utility Solutions

kw kw SmartGridCity Pricing Pilot Impact Evaluation Results Figure B8. Phase I, 2012, Summer, Weekday, TOU 2.0 1.8 1.6 1.4 1.2 1.0 Phase I: 2012 Summer Weekday - TOU 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending SS - Control SS - Treatment non-ss - Control non-ss - Treatment Figure B9. Phase I, 2012, Summer, Event Day, CPP 1.4 Phase I: 2012 Summer Event Day - CPP 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment EnerNOC Utility Solutions B-11

kw Figure B10. Phase I, 2012, Summer, Event Day, PTR 1.4 Phase I: 2012 Summer Event Day - PTR 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment One thing to note here is that the savings are fairly consistent throughout the event. This is typical of a pricing program that is not primarily driven by an enabling technology such as a programmable communicating thermostat (PCT). PCTs tend to have a large savings in the first hour of an event when the temperature is reset, and the savings decay across the remaining hours of the event as the temperature inside the homes slowly increases. When customer behavior is driving the load reduction, the actions taken tend to result in more consistent savings across the event, which is what we see here. Phase II Non-Summer Analysis Figure B11 through Figure B15 provide an illustration of how Phase II participants responded during the non-summer months. The event day graphs represent average load profiles for Phase II participants during the four non-summer event days (10/4/2011, 11/1/2011, 12/1/2011, and 5/18/2012). The non-summer, non-event day graphs show that the Phase II control groups and participant groups are closely matched. But, the load profiles are so closely matched that very little energy savings are apparent on non-event days. This is a distinct difference from what was observed on the Phase I graphs where slight savings were apparent. This difference likely reflects how participant s attitudes vary between the Phase I and Phase II participant. Phase I participants who voluntarily joined the rate may have greater energy awareness and be pre-disposed to conserve energy, as compared to their Phase II counterparts who did not actively seek to change rates. B-12 EnerNOC Utility Solutions

kw kw SmartGridCity Pricing Pilot Impact Evaluation Results Figure B11. Phase II, 2012, Non-Summer, Non-Event Weekday, CPP 1.4 Phase II: 2012 Non-Summer Non-Event Weekday - CPP 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment Figure B12. Phase II, 2012, Non-Summer, Non-Event Weekday, PTR 1.2 Phase II: 2012 Non-Summer Non-Event Weekday - PTR 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment EnerNOC Utility Solutions B-13

kw kw Figure B13. Phase II, 2012, Non-Summer, Weekday, TOU 1.4 Phase II: 2012 Non-Summer Weekday - TOU 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending SS - Control SS - Treatment non-ss - Control non-ss - Treatment Figure B14. Phase II, 2012, Non-Summer, Event Day, CPP 1.4 Phase II: 2012 Non-Summer Event Day - CPP 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment B-14 EnerNOC Utility Solutions

kw SmartGridCity Pricing Pilot Impact Evaluation Results Figure B15. Phase II, 2012, Non-Summer, Event Day, PTR 1.2 Phase II: 2012 Non-Summer Event Day - PTR 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment Energy savings are apparent on event days, with the CPP savings much larger than the PTR savings. Consistent with what was mentioned above, the Phase II participants underperformed the savings of the Phase I participants. Phase II Summer Analysis Figure B16 through Figure B20 illustrate how Phase II participants responded during the summer months. The event day graphs represent averages of the eleven 2012 summer event days. The Phase II results are fairly similar across all rates and day types to those we saw for Phase I. EnerNOC Utility Solutions B-15

kw kw Figure B16. Phase II, 2012, Summer, Non-Event Weekday, CPP 1.4 Phase II: 2012 Summer Non-Event Weekday - CPP 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment Figure B17. Phase II, 2012, Summer, Non-Event Weekday, PTR 1.2 Phase II: 2012 Summer Non-Event Weekday - PTR 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment B-16 EnerNOC Utility Solutions

kw kw SmartGridCity Pricing Pilot Impact Evaluation Results Figure B18. Phase II, 2012, Summer, Weekday, TOU 1.8 1.6 1.4 1.2 1.0 Phase II: 2012 Summer Weekday - TOU 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending SS - Control SS - Treatment non-ss - Control non-ss - Treatment Figure B19. Phase II, 2012, Summer, Event Day, CPP 1.4 Phase II: 2012 Summer Event Day - CPP 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment EnerNOC Utility Solutions B-17

kw Figure B20. Phase II, 2012, Summer, Event Day, PTR 1.4 Phase II: 2012 Summer Event Day - PTR 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment B-18 EnerNOC Utility Solutions

C. Appendix C 2013 Difference of Differences Results Impact Results Table C1 through Table C4 summarize the difference of differences results for 2013. Table C1 and Table C2 present the on-peak kw reduction results for Phase I and Phase II, respectively, while Table C3 and Table C4 present the kwh savings results for Phase I and Phase II. Table C1. kw Reduction by Rate, 2013, Phase I Baseline On-Peak Average kw Average On-Peak kw Reduction Percent Reduction CPP Average Non-Summer Non-Event Weekday 0.94 0.09 9.74% Average Summer Non-Event Weekday 0.90 0.14 15.54% Average Summer Event Day 0.93 0.20 21.80% Average Non-Summer Event Day 1.08 0.18 16.45% PTR Average Non-Summer Non-Event Weekday 0.89 0.04 4.51% Average Summer Non-Event Weekday 1.06 0.06 5.67% Average Summer Event Day 1.13 0.09 7.68% Average Non-Summer Event Day 0.94 0.05 5.18% TOU NSS Average Non-Summer Weekday 0.93 0.04 4.16% Average Summer Weekday 0.99 0.05 5.08% TOU SS Average Non-Summer Weekday 1.04 0.01 0.61% Average Summer Weekday 1.69 0.12 6.83% The estimates in Table C1 show the following trends for Phase I on-peak kw reduction in 2013: Impacts are higher in the summer Impacts are higher on event days, both in summer and non-summer Impacts are higher for CPP participants than for PTR participants For summer non-event weekdays, TOU NSS impacts are comparable to those for PTR participants, but lower than for CPP participants For non-summer non-event weekdays, TOU impacts are lower than for CPP and PTR participants Non-summer impacts are lower for SS participants than for NSS participants EnerNOC Utility Solutions C-1

Table C2. kw Reduction by Rate, 2013, Phase II Baseline On-Peak Average kw Average On-Peak kw Reduction Percent Reduction CPP Average Non-Summer Non-Event Weekday 1.01 0.03 2.64% Average Summer Non-Event Weekday 1.00 0.08 7.57% Average Summer Event Day 1.02 0.13 12.82% Average Non-Summer Event Day 1.12 0.09 7.78% PTR Average Non-Summer Non-Event Weekday 0.89 0.02 2.32% Average Summer Non-Event Weekday 1.03 0.04 3.60% Average Summer Event Day 1.11 0.09 8.30% Average Non-Summer Event Day 0.93 0.02 1.78% TOU NSS Average Non-Summer Weekday 1.01 0.03 3.03% Average Summer Weekday 1.23 0.04 2.93% TOU SS Average Non-Summer Weekday 1.03 0.01 1.30% Average Summer Weekday 1.63 0.08 5.13% The estimates in Table C2 show the following trends for Phase II on-peak kw reduction in 2013: Impacts are higher in the summer, except for TOU NSS participants Impacts are higher on event days, both in summer and non-summer Impacts are higher for CPP participants than for PTR participants For summer non-event weekdays, TOU NSS impacts are lower than for PTR and CPP participants For non-summer non-event weekdays, TOU NSS impacts are higher than for CPP and PTR participants Non-summer impacts for SS participants are lower than for NSS participants, while summer impacts for SS participants are higher than for NSS participants Except for the TOU participants, the 2013 Phase I and Phase II impacts follow the same trends. However, estimates suggest the Phase I impacts are larger than the Phase II impacts for CPP and PTR participants. When compared with 2011 and 2012 results, the 2013 trends are generally consistent, with the magnitudes of the kw reduction estimates closer to the 2012 values than the 2011 values, especially for the CPP and PTR participants. C-2 EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results Table C3. Annual kwh Savings by Rate, 2013, Phase I Baseline kwh kwh Change Percent Change CPP On-Peak Consumption 1,394-169 -12.14% Off-Peak Consumption 5,908-526 -8.90% Overall Consumption 7,302-695 -9.52% PTR On-Peak Consumption 1,435-73 -5.09% Off-Peak Consumption 5,652-340 -6.02% Overall Consumption 7,087-413 -5.83% TOU NSS On-Peak Consumption 1,426-65 -4.58% Off-Peak Consumption 5,656 78 1.38% Overall Consumption 7,082 13 0.18% TOU SS On-Peak Consumption 1,891-66 -3.51% Off-Peak Consumption 6,854 24 0.34% Overall Consumption 8,744-43 -0.49% The estimates in Table C3 show the following trends for Phase I annual kwh savings in 2013: We again see reductions in the off-peak period for PTR and CPP, which is probably due to increased awareness of energy use Impacts are largest for CPP participants TOU participants use more energy during off-peak periods relative to the baseline; however, it is not enough to offset the on-peak reduction for the SS customers, but is very close to the same as the on-peak reduction for the NSS customers, resulting in almost no overall change for the NSS group Overall consumption for TOU NSS participants is slightly higher than the baseline On-peak impacts for PTR and TOU NSS participants are comparable EnerNOC Utility Solutions C-3

Table C4. Annual kwh Savings by Rate, 2013, Phase II Baseline kwh kwh Change Percent Change CPP On-Peak Consumption 1,515-73 -4.82% Off-Peak Consumption 6,412-6 -0.09% Overall Consumption 7,927-79 -1.00% PTR On-Peak Consumption 1,413-43 -3.06% Off-Peak Consumption 5,569-227 -4.07% Overall Consumption 6,982-270 -3.87% TOU NSS On-Peak Consumption 1,628-50 -3.07% Off-Peak Consumption 6,418-150 -2.34% Overall Consumption 8,047-200 -2.48% TOU SS On-Peak Consumption 1,847-57 -3.07% Off-Peak Consumption 6,699 80 1.19% Overall Consumption 8,546 23 0.27% The estimates in Table C4 show the following trends for Phase II annual kwh savings in 2013: We again see reductions in the off-peak period for PTR and TOU NSS, probably due to increased awareness of energy use TOU SS participants use more energy during the off-peak periods relative to the baseline, and overall consumption is negligibly higher than baseline On-peak impacts for all rates are comparable, with only slightly higher savings for CPP participants The percent change in consumption for PTR participants is similar during on- and offpeak In comparison with Phase I, the Phase II kwh savings impacts are smaller for CPP and PTR participants. Also, in Phase II the TOU NSS participants showed savings during the off-peak, while in Phase I the participants energy consumption increased in the off-peak. Load Profiles The following subsections present sample load profiles for 2013 to illustrate how participants responded relative to the control group. We divide the profiles into four groups: 1) Phase I Non- Summer; 2) Phase I Summer; 3) Phase II Non-Summer; and 4) Phase II Summer. In each group, we include separate graphs for each rate type and for non-event and event days. Because TOU participants do not receive special event day pricing or notification, we do not C-4 EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results include event day graphs for them. Each of the graphs shows two lines, one being the adjusted average control group load shape, and the other being the average participant group load shape. (The abbreviation CTL on the graphs refers to the control group.) The adjusted savings estimate is the difference between these two lines. The load profiles represent seasonal averages for the day types, where event day graphs average results for all event days in each season. In general, the graphs indicate the control group and participant group for each rate are well matched, as demonstrated by the similarity in load shape between the control group curves and the pricing plan curves. Phase I Non-Summer Figure C1 through Figure C5 show the average load profiles for Phase I participants during the non-summer period. We make several observations from these results: CPP: The CPP participant curve is consistently below that of the control group, particularly during on-peak hours and non-event days. The savings fluctuate more during event days. Overall, the average Phase I CPP customers are responding to on-peak pricing signals and reducing load. Additionally these energy savings extend into some of the off-peak hours where a decrease in energy can be seen, even though there is no price disadvantage to using energy off-peak. This may be a result of customers increased awareness of energy use in their homes. PTR: The PTR participants also appear to save energy over the control group on both event and non-event days, but not as dramatically during the on-peak period. This intuitively makes sense as the PTR customers do not receive the price signal during peak times, but are probably more energy conscious over the general population as they have been exposed to the pricing plan. TOU: The TOU SS customers are higher energy users than those without SS, even in non-summer, though the general shape of the load is the same. The summer usage is higher due to the fact that the TOU SS customers all have CAC, whereas the TOU NSS, like the CPP and PTR, include a mix of customers with and without CAC. The TOU NSS customers appear to be responding to the TOU price by increasing their usage relative to the control group during the off-peak time, and decreasing it during the peak times. While the effect is not as strong with the TOU customers with SS, the two lines are nearly the same during the on-peak period, but the participants use more during the daytime off-peak times. The increased usage during the off-peak time may be in response to the lower prices that TOU customers see during those off-peak times. EnerNOC Utility Solutions C-5

kw kw Figure C1. Phase I, 2013, Non-Summer, Non-Event Weekday, CPP 1.4 Phase I: 2013 Non-Summer Non-Event Weekday - CPP 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment Figure C2. Phase I, 2013, Non-Summer, Non-Event Weekday, PTR 1.2 Phase I: 2013 Non-Summer Non-Event Weekday - PTR 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment C-6 EnerNOC Utility Solutions

kw kw SmartGridCity Pricing Pilot Impact Evaluation Results Figure C3. Phase I, 2013, Non-Summer, Weekday, TOU 1.4 Phase I: 2013 Non-Summer Weekday - TOU 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending SS - Control SS - Treatment non-ss - Control non-ss - Treatment Figure C4. Phase I, 2013, Non-Summer, Event Day, CPP 1.4 Phase I: 2013 Non-Summer Event Day - CPP 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment EnerNOC Utility Solutions C-7

kw Figure C5. Phase I, 2013, Non-Summer, Event Day, PTR 1.4 Phase I: 2013 Non-Summer Event Day - PTR 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment Phase I Summer Figure C6 through Figure C10 show the average load profiles for Phase I participants during the summer period. We make several observations from these results: CPP: The CPP participants appear to be reducing load at all times with greater reductions during on-peak periods on both non-event days and event days. The summer graphs indicate a clearer price response than the non-summer graphs. This is likely due to the higher price differential as well as the presence of CAC for some customers, which makes more load available to reduce. When we look at the event day, we see the CPP customers reduce load dramatically during the on-peak period. These savings are higher than at any other time and for any other rate group (except for the same group for Phase II). PTR: In comparison with CPP, the PTR participants show smaller reductions in all hours on the non-event weekdays. This is in line with expectations as there is no price signal driving savings for non-event weekdays. During event weekdays when the opportunity to earn a rebate applies, a much larger energy reduction is seen specifically during the onpeak hours. TOU: The TOU participants, both with and without SS, also show a reduction during the peak period. We assume this reduction is driven by the higher on-peak price. C-8 EnerNOC Utility Solutions

kw SmartGridCity Pricing Pilot Impact Evaluation Results Event Savings: The savings are fairly consistent throughout the event for CPP and PTR participants. This is typical of a pricing program that is not primarily driven by an enabling technology such as a programmable communicating thermostat (PCT). PCTs tend to have a large savings in the first hour of an event when the temperature is reset, and the savings decay across the remaining hours of the event as the temperature inside the homes slowly increases. When customer behavior is driving the load reduction, the actions taken tend to result in more consistent savings across the event, which is what we see here. Figure C6. Phase I, 2013, Summer, Non-Event Weekday, CPP 1.2 Phase I: 2013 Summer Non-Event Weekday - CPP 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment EnerNOC Utility Solutions C-9

kw kw Figure C7. Phase I, 2013, Summer, Non-Event Weekday, PTR 1.4 Phase I: 2013 Summer Non-Event Weekday - PTR 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment Figure C8. Phase I, 2013, Summer, Weekday, TOU 2.0 1.8 1.6 1.4 1.2 1.0 Phase I: 2013 Summer Weekday - TOU 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending SS - Control SS - Treatment non-ss - Control non-ss - Treatment C-10 EnerNOC Utility Solutions

kw kw SmartGridCity Pricing Pilot Impact Evaluation Results Figure C9. Phase I, 2013, Summer, Event Day, CPP 1.2 Phase I: 2013 Summer Event Day - CPP 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment Figure C10. Phase I, 2013, Summer, Event Day, PTR 1.4 Phase I: 2013 Summer Event Day - PTR 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment EnerNOC Utility Solutions C-11

kw Phase II Non-Summer Figure C11 through Figure C15 show the average load profiles for Phase II participants during the non-summer period. We make several observations from these results: Non-Event Savings: For all rate groups, the non-summer, non-event day graphs demonstrate that the Phase II control groups and participant groups are closely matched. The match is so close that very little energy savings are apparent on nonevent days. This is a distinct difference from what was observed on the Phase I graphs where more savings were apparent. This difference likely reflects how participant s attitudes vary between the Phase I and Phase II participant. Phase I participants who voluntarily joined the rate may have greater energy awareness and be pre-disposed to conserve energy as compared to their Phase II counterparts who did not actively seek to change rates. Event Savings: Overall Phase II on-peak energy savings are small on event days for both CPP and PTR during non-summer periods. The on-peak kwh savings for CPP are larger than the PTR savings. Consistent with what was mentioned above, the Phase II participants underperformed the savings of the Phase I participants. Figure C11. Phase II, 2013, Non-Summer, Non-Event Weekday, CPP 1.4 Phase II: 2013 Non-Summer Non-Event Weekday - CPP 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment C-12 EnerNOC Utility Solutions

kw kw SmartGridCity Pricing Pilot Impact Evaluation Results Figure C12. Phase II, 2013, Non-Summer, Non-Event Weekday, PTR 1.2 Phase II: 2013 Non-Summer Non-Event Weekday - PTR 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment Figure C13. Phase II, 2013, Non-Summer, Weekday, TOU 1.4 Phase II: 2013 Non-Summer Weekday - TOU 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending SS - Control SS - Treatment non-ss - Control non-ss - Treatment EnerNOC Utility Solutions C-13

kw kw Figure C14. Phase II, 2013, Non-Summer, Event Day, CPP 1.6 Phase II: 2013 Non-Summer Event Day - CPP 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment Figure C15. Phase II, 2013, Non-Summer, Event Day, PTR 1.4 Phase II: 2013 Non-Summer Event Day - PTR 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment C-14 EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results Phase II Summer Figure C16 through Figure C20 show the average load profiles for Phase II participants during the summer period. We make several observations from these results: CPP: The Phase II CPP participants appear to be reducing load predominantly during the on-peak period during both non-event and event days. The summer graphs indicate a much clearer price response than the non-summer graphs. As mentioned for the Phase I participants, this is likely due to the higher price differential as well as the presence of CAC for some customers. When we look at the event day, we see the CPP customers reduce load dramatically during the on-peak period. These savings are higher than at any other time and for any other rate group (except for the same group for Phase I). PTR: In comparison with CPP, the PTR participants show smaller reductions in all hours on the non-event weekdays. This is in line with expectations as there is no price signal driving savings for non-event weekdays. During event weekdays when the opportunity to earn a rebate applies, a much larger energy reduction is seen specifically during the onpeak hours. TOU: The TOU participants also show a reduction during the peak period. We assume this reduction is driven by the higher on-peak price. Savings for TOU SS participants are more noticeable than for NSS participants. Event Savings: As we noticed for Phase I, the savings are fairly consistent throughout the event for CPP and PTR participants. This is typical of a pricing program that is not primarily driven by an enabling technology such as a PCT, but instead is more driven by customer behavior. EnerNOC Utility Solutions C-15

kw kw Figure C16. Phase II, 2013, Summer, Non-Event Weekday, CPP 1.4 Phase II: 2013 Summer Non-Event Weekday - CPP 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment Figure C17. Phase II, 2013, Summer, Non-Event Weekday, PTR 1.4 Phase II: 2013 Summer Non-Event Weekday - PTR 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment C-16 EnerNOC Utility Solutions

kw kw SmartGridCity Pricing Pilot Impact Evaluation Results Figure C18. Phase II, 2013, Summer, Weekday, TOU 2.0 1.8 1.6 1.4 1.2 1.0 Phase II: 2013 Summer Weekday - TOU 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending SS - Control SS - Treatment non-ss - Control non-ss - Treatment Figure C19. Phase II, 2013, Summer, Event Day, CPP 1.4 Phase II: 2013 Summer Event Day - CPP 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment EnerNOC Utility Solutions C-17

kw Figure C20. Phase II, 2013, Summer, Event Day, PTR 1.4 Phase II: 2013 Summer Event Day - PTR 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour Ending Control Treatment Summary of 2013 Findings Based on the difference of differences ex post analysis described above, we see an indication of energy and demand savings achieved for each of the rate options in 2013. We also saw this for the 2011 and 2012 analysis years. The following bullet points summarize the direction and magnitude of savings observed during the 2013 difference of differences analysis. It is important to note that we did not assess the statistical significance of any of the differences of differences results. CPP: Participants on the CPP rate demonstrate the greatest savings to date. During summer event days these participants had an average demand reduction of about 22% (0.20 kw) for Phase I and about 13% (0.13 kw) for Phase II. In addition, the on-peak price signal on non-event days throughout the year encouraged reduction in on-peak demand: 10% during non-summer months and 16% in the summer for Phase I, and 3% and 8% for non-summer and summer, respectively, for Phase II. This behavior change carried over to some extent into the off-peak hours for Phase I, as we can see energy reductions during both time periods. The average overall energy reduction for the analysis year was 10% for Phase I participants. PTR: During summer pricing events, PTR participants had an average load reduction of less than half that of the CPP participants: 8% for Phase I and 5% for Phase II. Overall annual energy savings were about 6% for Phase I and 4% for Phase II. This energy savings during times without a price signal is most likely due to the participant s high energy awareness. C-18 EnerNOC Utility Solutions

SmartGridCity Pricing Pilot Impact Evaluation Results TOU: Phase I and Phase II TOU participants showed very small peak demand reductions in the winter, with those who were SS customers saving less than NSS participants. There were somewhat higher savings during the summer, with the SS participants reducing more, likely because they all had CAC systems. On an annual basis, the Phase II SS group and both the Phase I TOU groups actually used about the same or slightly more energy than their control group, differing by less than 1%. The design of the TOU rate may be the explanation of this, since off-peak energy is less expensive. Thus, energy use during peak periods declines, but the off-peak usage increase is greater than that on-peak reduction. EnerNOC Utility Solutions C-19

About EnerNOC Utility Solutions EnerNOC Utilities Solutions is a comprehensive suite of demandside management (DSM) program implementation and consulting services, technology platforms, and applications designed to address the evolving needs of utilities and grid operators worldwide. Hundreds of utilities have leveraged our technology, our people, and our proven processes to make their energy efficiency (EE) and demand response (DR) initiatives a success. Utilities trust EnerNOC to work with them at every stage of the DSM program lifecycle assessing market potential, designing effective programs, supporting program implementation, and measuring program results. The EnerNOC Utility Solutions consulting team has decades of combined experience in the utility DSM industry. We provide expertise, insight and analysis to support a broad range of utility DSM activities, including: potential assessments; end-use forecasts; integrated resource planning; EE, DR, and smart grid pilot and program design and administration; load research; technology assessments and demonstrations; EE project reviews; EE and DR program evaluation; and regulatory support. Our consulting engagements are managed and delivered by a seasoned, interdisciplinary team comprised of professional electrical, mechanical, chemical, civil, industrial, and environmental engineers as well as economists, business planners, project managers, market researchers, load research professionals, and statisticians. Utilities view EnerNOC s experts as trusted advisors, and we work together collaboratively to make any DSM initiative a success. EnerNOC, Inc. 500 Ygnacio Valley Road, Suite 450 Walnut Creek, CA 94596 Tel. 925.482.2000 Fax. 925.284.3147 www.enernoc.com