Published on: March 2010 Author: Sumant Sahoo 2009 Hexaware Technologies. All rights reserved.
Table of Contents 1. Introduction 2. Problem Statement / Concerns 3. Solutions / Approaches to address the issues 4. Summary 5. Case Study 03 03 04 04 05 2009 Hexaware Technologies. All rights reserved. 2
1. Introduction The demand for Business Intelligence is increasing at a rapid pace across all industries, more so during touch economic conditions as seen in 2009. Many senior executives / CXO s are looking to optimize the existing business process that can lead to top and bottom-line benefits. This paper presents the different levels of analytical solutions that are required for organizations with specific focus on Predictive Analytics. It is Hexaware s view that Predictive Analytics is not leveraged optimally for various reasons and it is time that enterprises realize the power of predictive analytics to help them move from answering the question of How are we doing to What does our future look like. 2. Problem Statement / Concerns Though many companies have adopted Business Intelligence solutions and it enables slicing and dicing of their data and provides detailed view of what s going on, they are challenged on getting insights into the future ("What should we do") or ( What will happen next ). The lack of predictive analytical capabilities hinders organizations to take the right decision at the right time. Figure 1: Levels of Business Intelligence Figure 1 depicts how intelligence can help organizations become more competitive in the market place. While most of the companies are able to answer the first four steps in the above diagram with the help of BI solutions, they are not able to move to the higher levels, which is the domain of Predictive Analytics. The difference in benefits for projects that incorporated predictive analytics versus those that did not shows that predictive projects yielded higher median ROI of 145% whereas non-predictive projects yielded ROI of 89% (Source: IDC). The major benefits of business analytics projects that employed predictive analytics centered on business process enhancement, especially improving the quality of operational decisions. Increasingly, organizations in virtually every industry around the world are realizing the benefits of using data to align their current actions with their future objectives. By incorporating predictive analytics into their daily operations, these organizations can gain control over the decisions they make every day. Most executives, while extremely interested in implementing predictive analytics, are perplexed by the perceived complexity of such applications. Most of the problems in implementation of predictive analytics boil down to the following items: How to get started: What is the approach, whom to hire, how to organize the project, or how to architect the environment. How to develop the model (issues related to preparing huge data, training models, statistical application etc) Identifying and applying the right predictive model Ongoing maintenance of models and validations. 2009 Hexaware Technologies. All rights reserved. 3
3. Solutions / Approaches to address the issues Before addressing these issues, let s define Predictive Analytics. Predictive analytics comprises of a variety of techniques from statistics and data mining that analyze current and historical facts to make predictions about future events. In business, the predictive model exploits hidden patterns found in historical and transactional data and predicts the probable future outcomes with a certain degree of accuracy. These Predictive models capture relationships among many factors in the transactional data associated with a particular set of conditions, guiding decision making for candidate transactions. Basically these models ensure that the actions taken today will directly achieve the organization s goals tomorrow. That s the way a Predictive analytics model works, and that s what gives it competitive advantage in the marketplace. Predictive analytics can help companies optimize existing processes, better understand customer behavior, identify unexpected opportunities and anticipate problems before they happen. Vladimir Stojanovski, a blogger and an engagement manager/solutions architect, offers a nice metaphor to describe the relationship between predictive analytics and business intelligence (BI): If BI is a look in the rearview mirror, he writes on his blog, predictive analytics is the view out the windshield. While BI is reactive and looks backward to gauge performance, predictive analytics seeks to use data in real time and helps to take decisions which affect future performance. As industry wide focus is now on high end predictive analytics, Hexaware BI Center of Excellence (COE) team has developed many analytical packs across domains, independent of BI tools. These analytical packs have rich domain content with BI slice and dice dashboards and many predictive analytics scenarios. In this paper the content & structure of analytical packs and a case study with predictive analysis scenarios will be explained. Analytical Packs are developed for a specific functional purpose. This purpose can be completely domain focused (Insurance Analytics, Fund Transfer Analytics) or can be applicable across multiple industries (Human Resource, CRM analytics etc). The Analytical Packs provide the flexibility of a custom built solution and also the benefit of faster turn-around time as provided by packaged BI apps. Also, the analytical packs can be used by organizations to understand their analytical needs better before embarking on bigger BI initiatives. The following are the pre-built components of an Analytical pack: List of Subject Areas that make up a functional domain (Example - HR Analytics will cover the subject areas of Staffing, Retention, Workforce, Organization Effectiveness, Compensation & Benefits and Environment etc) Set of Business Questions for each subject area Data Model for the functional domain / specific subject areas Semantic Layer for ad-hoc analysis Canned Reports Pre-defined Metrics / KPI's Executive Level Dashboards (based on roles) Predictive Analytics Scenarios and Mining models Connectors to source systems (if feasible) Based on our past project implementation experience, we have developed this analytical packs in a structured approach. The subject areas in a function/domain and the metrics/kpi are identified from past experience and domain experts. Then the data model is developed for all subject areas and a data warehouse created which pulls all relevant information from multiple source systems and then integrates and standardizes this information for data to be reported and analyzed. Once this warehouse is built, the predictive model is built on the warehouse for getting data from one source instead of getting data from various source systems. In addition, the data warehouse loads, cleans, integrates, and formats data which are the major time consuming tasks when building the predictive models. We have a pool of people (Data Mining PhD, Statistician, Six sigma Black belts and Data Analysts) to identify the right techniques for specific customer situations and the data requirements to develop the models. Some of the models which we have applied in our analytical packs are described in the case study. 4. Summary Following few steps/recommendations can be followed to get started with Predictive analytics. Develop Business Strategy Identify which are the key business processes and what KPI s needs to be focused on. This will provide the right direction to implement Predictive Analytics. Availability of data - Create a data warehouse and get the data in the right format which can be presented to the data modeler team to develop models on those priority KPI s. Identify Right Resource Develop data mining and modeling skills internally to develop those predictive models or identify a suitable partner who can initially setup the data modeling process and help nurturing those predictive skill sets. Develop Analytic Models Once the right resource and data sets are available, develop predictive models and validate the models for better accuracy. These models can be verified periodically for better forecasts. 2009 Hexaware Technologies. All rights reserved. 4
Reward the team As getting the right analytical modelers are difficult and high cost is involved, reward the team not only with paychecks but also providing challenging environments to demonstrate their capabilities. In Rob Neyer s (ESPN) words, In business, as in baseball, the question isn t Whether or not you ll jump into analytics. The question is when. Do you want to ride the analytics horse to profitability Or follow it with a shovel? 5. Case Study A company wanted to introduce a new product to the market. Though the product was new, it was in a product family which existed in the market. Based on the past few years performance, the following analysis was carried out before the product was released to the market. The analysis not only helped to create target customers but also helped in increasing revenue/profit based on target marketing. Objective: Introducing a new product to the market. Before introducing the product to the market, the organization wanted to find out about the kind of campaigns that would give maximum returns and the projected revenue from the new product. Following steps proved to be helpful for the organization. Step 1: Ascertaining Overall Purchase Value Trend of the product family Step 2: Evaluating the campaign (Festival Discount, Bundling, Trade shows, Promotions etc) which would be the most effective for that product family Step 3: Zero down on the channel (Direct Contact, Newspaper, Internet, Hoarding etc) which would be the most efficient (Cost and Responses) and allocating the cost to different channels based on purchase value. Step 4: Based on the trend of the last few years responses / revenues, data were extrapolated for the current campaign to find out the revenue from this campaign (% Revenue). Step 5: Identifying Segments / Making clusters based on customer s demographic factors, product types and channels to prepare the target list of customers. Step 6: Finding out any other product that can cross sell with this new product. Output of the steps described above is shown below. Pictures marked as 1 and 2 are the output of clustering analysis. Based on the customer s demographic factors and past purchase behavior, clusters were developed and a target campaign was made for getting higher responses/revenues. Similarly picture 3 presents the output of revenue projection. The analysis was carried out by using Time series modeling. Based on last 3 years revenue, future revenue was projected considering the trend and seasonality factor. Again picture 4 is the output of Decision tree modeling which basically presents the probability of people who will join the organization after an offer has been made to them. Factors considered are demographic factors, experience, skill set and the time taken to provide the offer. The dark boxes shows high probability of joining the organization and light colored boxes represents very less probability of joining the organization. This helped the organization to take decisions/actions based on the outputs. 2009 Hexaware Technologies. All rights reserved. 5
In general, the following steps are used to develop predictive models. Identifying the output / Key metrics that needs to be analyzed Identifying Predictor(s), which is the central building block of predictive models. The number of predictor(s) in the model should be between 2 to 15. More number of predictor(s) can make the model more complex and maintenance of model will also be difficult. Once these predictor(s) are identified, data needs to be collected accordingly. i.e. Customer s demographic data, past purchase transactions etc. Identification of the right predictive model to forecast the future trend. Though there are several techniques available, below mentioned models are used frequently. Regression Techniques (Linear, Non-linear, Logistic, Multivariate etc) Time Series Forecasting (Moving Average, Smoothing, Winter s method etc) Classification / Decision Tree Association Rules Credit Scoring Clustering Analysis Optimization Techniques Final step is to verify the accuracy of the model and tweak the model for better accuracy. Parameters can be looked at for accuracy is MAD, MAPE, Lift etc. As part of our analytical pack, we have developed many predictive models across industries and across domains based on applicability. Benefits Target Modeling This type of model basically targets smaller set of people who are likely to respond to an offer. So this helps in increasing response rates and automatically reduces the cost per contact dramatically. Churn Modeling This model helped in identifying the customers who are likely to leave. So it s better to focus on retaining other customers than spending on those customers who will leave anyways. Forecasting Model These models predict the likely future outcomes. The Forecasting model applied to project revenue based on past campaign revenue help take proactive action to meet the target. 2009 Hexaware Technologies. All rights reserved. 6
Address 1095 Cranbury South River Road, Suite 10, Jamesburg, NJ 08831. Main: 609-409-6950 Fax: 609-409-6910 Safe Harbor Certain statements on this whitepaper concerning our future growth prospects are forward-looking statements, which involve a number of risks, and uncertainties that could cause actual results to differ materially from those in such forward-looking statements. The risks and uncertainties relating to these statements include, but are not limited to, risks and uncertainties regarding fluctuations in earnings, our ability to manage growth, intense competition in IT services including those factors which may affect our cost advantage, wage increases in India, our ability to attract and retain highly skilled professionals, time and cost overruns on fixed-price, fixed-time frame contracts, client concentration, restrictions on immigration, our ability to manage our international operations, reduced demand for technology in our key focus areas, disruptions in telecommunication networks, our ability to successfully complete and integrate potential acquisitions, liability for damages on our service contracts, the success of the companies in which Hexaware has made strategic investments, withdrawal of governmental fiscal incentives, political instability, legal restrictions on raising capital or acquiring companies outside India, and unauthorized use of our intellectual property and general economic conditions affecting our industry. 2009 Hexaware Technologies. All rights reserved.