How To Get More From Your Existing IVR You don t need an expensive IVR project to improve self-service, reduce mis-routes, and improve IVR satisfaction. Analyzing customer behavior inside your IVR will reveal fixable obstacles and frustrations that will make a big difference. The need to both reduce costs and increase customer satisfaction has never been more urgent. Call centers cannot afford to waste agent time on avoidable calls so they need the IVR to resolve selfservable calls and route the rest correctly. And everyone, including the IVR team, must consider their impact on customer experience. Bad experiences in the IVR can show up in social networks, damaging the brand and negatively impacting customer acquisition and retention. At the same time, large IVR projects are expensive and risky. Changing your entire IVR flow, or converting from touchtone to speech, can be a huge disruption. And such large scale changes are frequently unwarranted. After all, the business goal is not to replace the IVR or adopt speech. The goal is to reduce agenthandled call volume by increasing self-service success and by routing more calls to the correct queue. And to increase customer satisfaction with the IVR or at least to prevent it from showing up in satisfaction surveys as a major source of frustration. The real question is not whether to replace or upgrade your IVR it s what specific changes will measurably improve IVR performance and satisfaction? Exactly where do callers hit obstacles, and why? Would you get more value from adding functions, or from improving the functions you already have? What are the top 3-5 changes that would have the highest measurable impact on agent-handled volume? Or, on customer satisfaction with the IVR? Once you know exactly what changes you need to make to your IVR, and exactly how much impact they ll have on agent-handled volume and satisfaction, then you can map those changes to the implementation requirements. You may need very small and surgical changes to a couple menus, or perhaps an entirely new module. You may need a couple ports of speech recognition, or none at all. You may need to bring touchtone back into your application where customer behavior and the task makes it the more effective strategy. Most IVR systems have untapped potential, which is unlocked by analyzing actual customer behavior. Optimization of your IVR driven by customer behavior analysis can reduce agent handled call volume by 10-20% and remove negative experiences from 30-50% of calls. Four Ways To Get More From Your IVR In this paper, we ll use the term IVR (Interactive Voice Response) to refer generally to call automation in other words, using menus, prompts and automated responses to deliver your callers to the right place
or to provide self-service. Such call automation may be provided by menus from your switch or automatic call distributor (ACD). Menus and prompts can also be presented in the network by your tollfree provider. You may also have a Voice Response Unit (VRU) or IVR system in your data center or hosted by a partner. In fact, many companies have a combination of these forms of call automation. Until recently, too much attention has been focused on input modality in other words, how the caller is asked to provide information. The choices are touchtone or one of two forms of speech recognition (directed dialog and natural language). Fortunately, debates over the merits of each are slowly being replaced with data about which provides a better caller experience, increased self-service, and a more cost effective solution. Based on over 30 customers with touchtone and speech recognition, we ve found that speech systems do not consistently deliver higher customer satisfaction or reduced agent call volume. In fact, many companies can realize very significant agent time savings by optimizing their existing touchtone applications. We ve also found that the highest performing systems use both speech recognition and touchtone since each is uniquely suited for specific tasks and environments. Regardless of your IVR technology platform, the goals of call automation are the same. Every call automation strategy has 2 measureable and quantifiable objectives: IVR Routing IVR Self-Service Get the caller to the right place Eliminate the need for an agent to handle the call (or at least automate part of the call before reaching an agent) If your automated systems are able to consistently and reliably deliver the caller to the right place, then you win on two counts fewer calls and higher satisfaction. The reason is that every time a caller reaches the wrong agent, they must be transferred. One caller is now driving two agent-handled calls. The first agent s interaction was wasted and avoidable. That s excess cost in your center and a caller that is frustrated by having to spend more time just trying to reach the right agent. Similarly, if your system achieves a high self-service success rate, then you win again on both counts fewer agent-handled calls and higher satisfaction. Callers are happy because they got their answer without waiting in a queue. Many of your callers may actually prefer dealing with a well designed automated system. Even if callers proceed to an agent after self-service, agent time is reduced by the tasks already handled by your automated system. If you ve implemented IVR self-service, your menu systems are even more important. Poorly designed menu systems can wrongly send a caller to an agent when they should have been sent to your selfservice application. So, how can you optimize your call automation systems switch/network menus and IVR applications to maximize correct IVR routing and self-service success?
1. Reduce IVR abandons. 2. Maximize identification success. 3. Maximize IVR navigation accuracy. 4. Maximize self-service success. Four Ways To Get More From Your IVR The first step is to reduce IVR abandons. These are callers that hang-up the phone in the IVR before getting anything useful done. There are two kinds of abandons those caused by a distraction that requires the caller s attention (e.g. someone at the door, or another important call), and those caused by frustration with your automated systems. The latter are the ones to worry about. They ll call back and immediately opt-out to reach an agent. Essentially, problems in your automated systems have trained these callers to always opt-out and get to an agent as quickly as possible even if it means they ll have to be transferred to get to the right agent. The second step is to maximize identification success rate. The identification of the caller may be the most important part of the automated process. When a caller is identified in the IVR, their segment and status can be used to route them to the right place, they can complete self-service tasks, and screen pops can help the agent service the caller more effectively. The third step is to maximize IVR navigation accuracy. Menus and prompts are designed with a specific intended path for each type of call. If callers follow the intended path, they ll be delivered to the right place be that a self-service application or a specific skill group. If they don t, the call center will have excess transfer rates and low self-service utilization. The final step is to maximize self-service success. Self-service success is best measured as a percentage of callers who attempt self-service. How many callers who start the self-service transaction are able to complete it successfully? In the rest of this paper, we ll walk thru a methodology to examine your existing performance in these 4 areas. The methodology also describes how to pinpoint specific changes in your existing systems to improve IVR routing and self-service. A Data-Driven & Caller-Centric Approach Understanding exactly what your callers are doing so that you can design and implement precisely targeted improvements requires a new approach. Historically, IVR optimization has been a trial-anderror exercise. Call automation systems typically don t provide much reporting, so teams seek consultant s opinions and benchmark data to come up with a list of possible improvement actions. The team then tries one or more of the possible fixes and watches for signs of whether things have gotten better or worse. With few resources to invest and urgent needs to reduce agent call volume and increase satisfaction companies don t have time for this kind of trial-and-error. Scarce resources can t be spent on changes that may not deliver improvement or that may even have a negative impact. The reason why traditional methods fail to reliably deliver improved performance is because benchmarks and best practices are a poor substitute for a true understanding of your actual callers. Benchmarks and best practices are created by aggregating data and experiences from a large number
of companies. Therefore, they represent an average from a diverse set of companies - each with very different kinds of customers, very different competitive positions, and very different business strategies. Such averages are helpful to spot big issues and learn from other s mistakes. But, they can t help you optimize your call center for your unique callers and how they interact with your company. Your callers and their behavior are a function of your unique products & channels, customer communications, and competitive market position. Every Call Center Is Different Data from 6 centers in 4 companies all in the same industry For example, your self-service rate may look good against benchmarks. But, your inbound volume may contain more self-servable calls than the average company in the benchmark. Data from 9 BBN customers in one industry revealed that the average number of self-servable calls was 50% of the total inbound volume. However, 3 of the companies received a significantly higher number of self-servable calls. In fact, 65% of the inbound call volume was self-servable for one company. If this company made decisions based on the 50% benchmark instead of the 65% of self-servable calls they actually received, they would miss the opportunity to reduce agent-handled volume by another 15%!
To get maximum value from your existing menus and IVR systems, you must adopt a data-driven and caller-centric approach. Such an approach has 3 major components: 1. Visibility of how your callers currently interact inside your automated systems to spot potential issues. 2. The ability to follow callers to an agent after they leave your automated systems in order to learn what the caller was trying to accomplish. 3. Having learned the caller s intent and how the agent successfully resolved the call, the ability to rewind the call, listen to the IVR interaction, and re-design it so the caller would have been successful. Next, we ll describe how to apply this data-driven and caller-centric approach to reduce abandons, maximize identification success, maximize routing accuracy, and maximize self-service success. Reduce IVR Abandons IVR abandons are those callers who hang-up the phone in your menus or self-service applications without getting anything useful accomplished. Your reporting may count these calls as having been contained within your IVR, but they are not resolved calls since the caller did not complete any selfservice. IVR abandons can be divided into 2 categories those where the caller abandons without having encountered any difficulty, and those where the caller abandons due to frustration with your automated system. The former is typically due to some distraction that required the caller s immediate attention. The caller may realize they ve dialed the wrong number, or may need to answer another call or a knock at the door. Such abandons are unavoidable and can account for 5-10% of total inbound call volume. Abandons due to frustration with your automated system are a serious and avoidable problem that should be your top priority. Problems with your menus or self-service applications are training these callers to avoid ever using your IVR. These callers usually call back after hanging-up and immediately opt-out to an agent. They have decided that getting to any human and being transferred to get to the right agent is better than trying to navigate your IVR. By not addressing issues in your IVR, you are creating unnecessary agent-handled call volume. And, you are creating a group of callers who are unlikely to ever use any new or improved call automation that you may develop in the future. a) Use log/event data to find IVR abandons Eliminating IVR Abandons b) Categorize IVR abandons by location where they abandon, and time spent before they abandon c) Listen to IVR abandons to discover what happened and why d) Use root cause understanding to design and implement precisely targeted changes Log and event data can be used to determine where callers abandon and how much time they invest before abandoning. Calls that abandon early in the call are typically not a cause for concern. The IVR abandons to focus on are those where the caller has invested a significant amount of time in your automated system before hanging-up. Such callers are likely to be frustrated by multiple re-tries to
provide information, by requests for information they don t have readily available, by self-service applications that have too many steps, or by the system s refusal to take them where they want to go (despite what they believe are correct responses to your prompts). There are several ways to find problematic IVR abandons. One of the simplest ways is to look for calls with excessively long time in the IVR, no self-service task completion, and the caller hanging-up before being delivered to a queue. Event or log analysis can also identify unproductive patterns of IVR activity, such as multiple re-tries followed by a caller terminating the call. If you have the ability to record caller experiences in the IVR, you can search for those that express their frustration before hanging-up. Once you locate a group of calls with problematic IVR abandons, the next step is to ascertain why those callers decided that continuing in your automated system was pointless. In our experience, there are many potential failure modes, and the subtleties of assessing intent, expectation, patience and other attributes of human behavior are complex. Perhaps it s not all that surprising then that it still takes human judgment and analysis to pinpoint the true reasons for a caller s interaction behavior with an automated system. Fortunately, there are tools to accelerate and simplify the process of capturing whole call recordings, categorizing IVR abandons, selective listening to critical parts of a call, and documenting the interaction. Analyzing IVR abandons in this manner reveals which of the following types of problems are present: Example Problems That Drive IVR Abandons Caller fatigue due to length of time in the IVR Frustration because IVR takes the caller down the wrong path Too many unsuccessful retries to provide information Caller doesn t understand what the IVR is asking for This approach also reveals root cause. For example, there may be a mis-alignment between the words in your prompts and the words the caller would use to express their needs. Other causes may include: time-outs too long or too short, barge-in problems, turn-taking issues, speech recognition problems, unhelpful error handling, excessively verbose prompts, or confusing messages. When you discover the exact failure mechanism that leads to an IVR abandon, you can design and implement a precision fix. By reducing IVR abandons, you ll eliminate repeat calls. You ll also reduce the number of callers that your agents have to deal with who were unnecessarily angered and frustrated by your IVR before they reached the agent. Maximize Identification Success The identification or authentication of callers may be the most important part of your automated call handling process. Once a caller is successfully identified, you have many opportunities to better serve the caller and use agent time more effectively. You can use the caller s identification to look-up their status and route them appropriately. For example, you may immediately route high-value customers to a special relationship skill group. Or, you may route callers with outstanding balances immediately to your collections partner. A successfully
identified caller improves routing accuracy and reduces transfers. Identification is also required for self-service. In our experience, identification problems are the 2nd largest obstacle to increasing self-service rates. High identification success rates are especially important for proactive messaging. Proactive messaging is a strategy to achieve self-service by providing information without forcing the caller to navigate your IVR. For example, if a caller has a pending claim, you may deliver claim status information immediately after the caller authenticates. Successful identification is also necessary to achieve maximum return-on-investment from CTI and screen pop technology. If the caller fails to identify in the IVR, then the agent will not receive a screen pop and will spend extra time identifying the caller and retrieving their information. Maximizing Identification Success a) Use log/event data to find identification failures b) Categorize identification failures by location and failure mode c) Follow failures after they exit the IVR to observe how the agent successfully identifies the caller d) Use root cause understanding and agent success strategies to design and implement precisely targeted changes Log and event data can be used to quickly isolate calls with identification failures. Depending on the tools you have available, this data can also be used to create an initial categorization of failures by the location of the failure in your identification process. Identifying the exact failure mechanism, and how to fix it, is considerably more challenging. For example, your reports may help you identify calls that exceeded the maximum number of re-tries for a particular piece of information. But, the reports may not tell you: (a) how many re-tries each caller attempted, (b) whether the errors were caused by the caller providing the wrong information, a backend look-up problem, or a speech recognition issue, or (c) what strategy the subsequent agent used to successfully identify the caller. Example Problems That Drive ID Failure Caller confusion about which number to provide (account number, member number, group number, claim number, social security number, card number, personal identification number ) IVR asks for information the caller does not remember or have readily available IVR asks for information that is difficult to capture reliably (alpha-numeric codes, email addresses ) Back-end issues (errors in member database, host connection problems ) Information entry problems (inter-digit timeouts, speech recognition issues ) In our experience, the fastest way to pinpoint the exact failure mechanism and how to fix it is to listen to callers fail in the IVR and then succeed with the agent. Listening to callers fail in the IVR quickly reveals which of the above issues is the culprit.
And, listening to those same callers as they are successfully identified by an agent reveals strategies you should consider adopting in your IVR. It s not uncommon to discover that agents require fewer items to authenticate the caller than your IVR. Or, that agents have learned to ask for different forms of identification those that are easier for the caller to provide or quicker for the agent to validate. Maximize IVR Navigation Accuracy The fundamental question to be answered in this analysis of your existing IVR is whether your callers navigate your menus properly given the reason for their call. Every menu system (or natural language router) is designed to guide the caller to the right place for their need. IVR and voice user interface (VUI) designers have an intended path and destination for each type of call. If callers behave as expected, they ll be delivered to the right skill group or to the right self-service application. However, if you have high transfer rates or low self-service utilization, the problem may be that your callers are not behaving as your IVR/VUI designers hoped they would so, your IVR is failing to deliver them to the right place. Maximizing IVR Navigation Accuracy a) Find calls that went to an agent, but should have been routed to a self-service application b) Find calls that went to the wrong agent and had to be transferred c) Examine these caller s paths thru your menus to discover where they made the wrong choice and why d) Use root cause understanding of your caller s errant selections to design and implement precisely targeted changes The key to improving IVR navigation accuracy is to examine agent-handled calls and compare the caller s true reason for call with the selections they made and the information they provided in the IVR. The true reason for call can be obtained from the beginning of the agent-caller dialog. The caller will state their reason for calling, and the agent will frequently paraphrase it back, or re-state it in terms of the transaction that the caller needs to complete. And, menu systems and IVRs can provide event and log data that describes the caller s selections and responses in your automated system for comparison. The difficult problem for many call centers is acquiring data that is both dependable and sufficiently detailed. Agent or quality coding of call types may not be consistently performed and the categories are frequently too broad. Similarly, IVR data may be difficult to extract and interpret. Even if such data is available, aligning it so that callers can be traced thru the IVR and on to an agent may not be possible. One way to overcome these obstacles is to utilize whole call recordings. An audio recording of the entire call from dialing to hang-up can be analyzed to extract both the true reason for call and the caller s activity in your menu systems. This gives you one solution to both identify navigation failures and drill-down to understand root cause.
Example Problems That Drive Navigation Failure Prompt wording (or speech grammer) doesn t match words the caller uses to describe their needs Option to exit to an agent is featured too prominently Prompts are too verbose or have too many choices Short time-outs, poor error handling, or logic problems kick callers to default agent skill Identifying which navigation problems exist in your IVR starts with finding calls where the true reasonfor-call doesn t match the caller s path thru the IVR. Then, having learned the caller s true need from the agent dialog, re-trace the caller s path thru your IVR. This will reveal where your callers actual behavior deviates from the assumptions that were made by your IVR/VUI design team. And, this root cause understanding will illustrate precisely targeted changes that will reduce transfers and increase self-service utilization. It s not uncommon to discover that small changes in prompt wording will make a big difference in IVR performance. Maximize Self-Service Success If your IVR keeps callers engaged, identifies them successfully, and delivers the right callers to your selfservice applications, the last step is to make sure callers are successful completing their self-service transactions. The goal in this step is to identify changes that will improve your self-service conversion rate. In other words, a high percentage of callers who start a self-service process are able to complete it successfully. This is different than the typical metric of self-service as a percentage of inbound call volume which can t be meaningfully compared across companies due to differences in customer interaction preferences and each company s strategy and overall communications effectiveness. Maximizing Self-Service Success a) Use log/event data to find self-service failures b) Categorize self-service failures by location and failure mode c) Follow failures after they exit the IVR to observe how the agent successfully serves the caller d) Use root cause understanding and agent success strategies to design and implement precisely targeted changes The process to maximize self-service success is very similar to the one described previously for maximizing identification success. The key difference is the type of transaction. Instead of identifying the caller, self-service transactions seek to complete a payment, check on a claim status, or make a purchase just to name a few examples. Ideally, you can assemble the tools and data necessary to quickly locate calls with self-service failures, drill-down to listen to the actual interaction and failure mechanism in the IVR, and then follow the caller to an agent to discover how the agent successfully completes the transaction.
Example Problems That Drive Self-Service Failure Caller fatigue from the number of steps and length of the process IVR asks for information the caller does not understand, remember, or have readily available IVR asks for information that is difficult to capture reliably (alpha-numeric data, email addresses ) Back-end issues (errors in databases, host connection problems ) Information entry problems (inter-digit timeouts, speech recognition issues ) The two best sources of targeted IVR improvements to maximize self-service performance will be the depth of understanding you achieve by: (a) listening to callers fail in your IVR, and (b) observing how agents rescue a failed self-service call and complete the transaction. Proven Success The data-driven and caller-centric approach described in this white paper has been successfully applied across a variety of industries in call centers with one million to over 50 million calls per year. Using this methodology, companies consistently discover a short list of specific actions that will reduce total agent time by 10-20%, while simultaneously improving customer satisfaction. The greatest value is realized by integrating the methodology into a program of continuous monitoring and improvement. A New Breed Of Analytics BBN developed AVOKE Caller Experience Analytics to simply and quickly deliver data and insights about a center's actual customer experiences to operational management and strategic decision making. BBN's solution consists of the AVOKE Call Browser and the AVOKE Caller Experience Methodology. The methodology provides a repeatable framework for synthesizing quantified business insights and action plans from your customers actual interactions. And, the AVOKE Call Browser is a revolutionary new software-as-a-service that captures and analyzes whole call experiences. Key attributes of the AVOKE Call Browser system include: True End-to-End View The AVOKE Call Browser captures the entire call from the moment the caller dials the contact number until the caller hangs-up, including all transfers, all IVR automation, all agents, and all outsource partner resources. Data, Audio and Transcription The AVOKE Call Browser provides IVR and speech analytics together so you can understand and optimize the customer s complete experience and the entire interaction process. The AVOKE Call Browser combines call data, IVR navigation, dialing-to-hangup audio recordings, and an automated full text transcription in a single real-time environment. Search and data views enable discovery of patterns (what happened), with drill-down to specific locations in the recording to understand caller intent and behavior (why it happened). No Software, No IT Integration Using patented BBN technology, the AVOKE Call Browser is integrated in the telephone network, not in your data center. There is no hardware or software to install at any of your locations.
About AVOKE Analytics AVOKE Analytics is a cloud-based whole call recording and analytics solution. The solution enables companies to optimize service performance from the customer s perspective. The AVOKE system records calls in the telecom network, eliminating IT requirements and obstacles. It follows your customers, uninterrupted, through their entire journey as they traverse the IVR, all transfers, internal call centers and partner sites. The solution consists of the AVOKE Call Browser system, the AVOKE Customer Effort Index benchmark service, and AVOKE Professional Services. For more information on AVOKE Analytics, contact: Raytheon BBN Technologies 10 Moulton Street Cambridge, MA 02138 avoke@bbn.com 617-873-1600 www.avoke.com