Analytic Opportunities Using Transactional Data

MIT Sloan in collaboration with software firm SAS recently published their Spring 2013 Research Report: From Value to Vision: Reimaging the Possible with Data Analytics. The premise is that some companies are great at analytics while most are mediocre. It is a theme other top researchers such Brynjolfsson, Laursen & Thorlund, and Davenport have addressed. The MIT Report offers evidence of superior returns for these analytic leaders along with anecdotes and attributes. My view is the essential difference between the analytic stars and the rest is that the stars think beyond using technology as a transactional tool and strive to convert transactional data into actionable information that can provide a competitive advantage in the marketplace.  They recognize the finite returns to cost saving and efficiency initiatives. In this short piece I will identify several specific opportunities that all firms have access to by leveraging their transactional databases. The three categories selected are particularly useful to Marketing decision-makers. Profitability Analytics - Measures Contribution income.  Contribution income is the amount of revenue in excess of variable or direct costs.  This is income derived from the sales of products and services and is available to cover fixed costs such as R&D funding, capital investment to maintain or expand the business, and surplus dollars providing a return to stockholders.   Once fixed costs and direct costs are covered, the remaining revenue represents operating profit.  For example, if $.60 of each revenue dollar is consumed by variable costs related to the sale, then $.40 is available to pay fixed costs and make other investments and provide stockholder’s with a return.  Contribution income is a short-term planning tool and is a better measure of profitability than net income since it is based on direct costs only which, unlike fixed costs, can be managed in the short term.  Many firms provide operating mangers a P & L statement that reports profit after fully loaded cost – including those fixed and joint costs that cannot be managed at the product or business unit level. This misleading information can result in a contrived number distorting unit cost information, leading to bad decisions. Economic value added (EVA).  Commonly referred to as economic profit, EVA[1] is a performance measure that compares net operating profit to total cost of capital.  The idea behind EVA is that businesses are only truly profitable when they create wealth for their shareholders, a measure that goes beyond net income.  EVA is based on the principle that businesses should create returns at a rate above their cost of capital (that is, the opportunity cost or rate of return that could have been earned by putting the same money into a different investment with equal risk).  The overarching importance of EVA is that it is an indicator of how profitable company projects are and serves as a gauge to measure management performance.  Understanding the significance of EVA encourages managers to think about a company’s assets as well as its expenses in their decisions. There are three components that drive EVA: net operating profit after tax (NOPAT), invested capital, and the weighted average cost of capital (WACC). Example 1:                          NOPAT = $1,000,000 Capital Invested = $5,000,000 WACC = 6% EVA = $1,000,000 - ($5,000,000 * .06) = $700,000   Example 2:                          NOPAT = $2,000,000 Capital Invested = $25,000,000 WACC = 10% EVA = $2,000,000 - ($25,000,000 * .10) = - $500,000 The positive EVA in Example 1 indicates that the company has more than covered its cost of capital for the project.  On the other hand, the negative EVA in Example 2 indicates that the project did not make enough profit to cover the cost of doing business.  Looking only at the net income may lead to the spurious conclusion that Example 2 is creating more value when, in fact, that firm‘s value is actually eroding. Predictive Analytics – Forecasting Applications Regression analysis.  Regression analysis is a very powerful predictive tool for marketers because it predicts customer behavior by determining associations, both in terms of magnitude and statistical significance, among dependent and independent variables.  Applications range from basic regression models such as simple linear regression to log transformed regression for bivariate models, and multiple linear regression and logistic regression for multivariate models.  In marketing, for example, a key dependent variable we typically wish to forecast is sales revenue, whereas, independent variables may be marketing strategies and tactics to achieve customer sales such as pricing, advertising and other promotional activities. If managers determine that regression results indicate that some promotional activities significantly increase sales, more resources can be allocated to that activity.  Or, the relative strength of a relationship between variables, as indicated by R2 (the coefficient of determination), may point to some combination of a lowered price point and percentage increase of advertising. Armed with this data, marketing managers can evaluate current marketing plans and create new promotional strategies. Monte Carlo Simulation. A forecasting method that is useful for solving complex financial and marketing models when a closed form solution is not possible. The essence of Monte Carlo is to identify critical variables in a model and their probability distribution. Through a succession of computer runs, mean outcomes and standard deviations allow the forecaster to compute a confidence interval of returns. It is especially useful for estimating break even and target return goals and other risk management issues. Logistic Regression.  A form of multiple regression that will predict the probability of an outcome (the dependent variable) as a function of multiple independent variables. An advantage of this model is the independent variables can be a combination of data types. For instance, a firm may be interested in determining the probability of a customer purchasing their product. It has been determined that their decision is influenced by gender (a binary variable), income (a ratio variable), occupational rank (an interval variable), and attitudinal response to a survey question (ordinal variable). This model can predict the probability of a sale for each customer, and the primary factor (s) driving that sale. It is a great tool for market segmentation and direct marketing efforts. Customer Relationship Management Models and Analytics The importance of managers understanding that not all customers bring the same value to a company cannot be underestimated.   Failure to segment customers, investing in financially valuable customers and the divesting of unprofitable customers, and not managing costs to improve margins and profitability measures can lead to costly CRM initiatives that underperform.  On the other hand, a well-designed CRM plan that merges a differentiated customer base with the appropriate customized marketing tactics will positively influence economic profit. (RFM) -  Recency, frequency, and monetary a CRM model.    Using a customer database to aggregate data and then combine how recently, how frequently they purchased, and how much customers spend is found to be a good indicator of customer loyalty and a predictor of retention and future profitability. The beauty of this model is its power and utility since it uses transactional data that every organization has for their customers.  Each customer in a transactional order database is sorted and assigned a score on a scale from 1 to 5 (1 = lowest and 5 = highest) on each RFM variable, depending on how well that customer has performed relative to all the other customers.  Next, the RFM variables are combined into a single rating score, such as 555, 321, or 112.  If the scoring is based on quintiles, then the highest RFM score a customer can get is a 5 in each of the categories.  Obviously, a 555 customer is an organization’s very best and most active customer, and one to reward and recognize.  In and of itself, RFM is an interesting way to classify individual customers, however, the real power of this model comes from analyzing and tracking individual customers as they migrate from one prescribed time to the next such as from Q1 to Q2.  Analysis of a customer’s RFM migration score lends a great deal of insight and information to the marketing analyst and can lead to rewarding customers for positive growth or to ward off customer dilution.  For example, Customer A had an RFM score of 544 (a good customer) in Q1 and an RFM score of 433 in Q2, a migration score of -111, or a total of -3.  Although not necessarily cause for great alarm, this migration score does indicate that there have been some downward fluctuations in business with this particular customer and some type of stop gap marketing tactic could be in order.  On the other hand, if Customer B had an RFM score of 333 in Q1 and migrated upward to 345 in Q2, Customer B’s score has a migration score of 012, a total of +3.  Customer B should be rewarded for its added business to the company. RFM when combined with multiple regression can be an effective predictor of Customer Lifetime Value. Customer lifetime value (LTV), a CRM analytic.  The LTV analytic predicts long-term profitability and factors in all the typical challenges facing most organizations such as how to retain customers, how to increase customer spending, how to reduce the cost of servicing customers, how much to spend on marketing to various customers based on that customer’s monetary value over time.   LTV is a forecasted metric that predicts how much economic value customers will bring to a company in some prescribed time interval – typically 5 years.  While its closest financial cousin is Net Present Value, LTV is nested in RFM data, and, although often underutilized, it is a key customer relationship management predictive analytic. Once customers have been segmented based on their LTV, results-oriented relationship sales and marketing initiatives can be designed.   For example, for those customers who have lower LTV scores, appreciation of current business through discounts or bundling could encourage greater spending and inspire loyalty.  On the other hand, for more profitable customers with higher LTV scores, the firm may allocate greater resources to develop long-term, rewards-based marketing initiatives to retain their future business. About the author: Craig Miller holds a BS in logistics and operations and an MBA is managerial accounting and finance from The Carlson School of Management in Minneapolis; and a Doctorate emphasizing computer simulation and multivariate modeling from Metropolitan State University in Minneapolis. He has been teaching, consulting, and developing decision support software for 25 years.

[1] Stern Stewart & Co. is credited with the development and trademark of the EVA concept.
Categories: BI and Big Data, Analytics
Tags: ;