Data base marketing, touted as a sure-fire method of fattening banking's anemic share of its customers' wallets in the early to mid-'90s, is now viewed in some quarters as a failed initiative. A number of banks have spent lavishly to build a data-mining capability without, however, achieving improved cross-sales performance or a much-needed reduction in costs. True, these institutions have learned a good deal about customer characteristics. But they have been unable to translate this knowledge into higher profits. The knowledge is descriptive, not yet prescriptive.
But there is still hope of converting customer information into effective sales and cost-reduction programs. Those banks that would make a success of data base marketing must institutionalize a process that goes far beyond simple data-gathering. The process can be characterized as a continuous learning cycle. It encompasses the following eight steps: (1) capturing relevant data; (2) creating valid measures of customer value; (3) brainstorming value-enhancing strategies; (4) designing, via fractional factorial analysis, appropriate value propositions; (5) fielding or implementing these value propositions; (6) collecting data on the results of this field activity; (7) modeling the changes in expected customer value attributable to each value proposition; and (8) marrying every new customer to his or her optimal value proposition.
Step 1-the data-building phase-is necessary but by no means sufficient. It is the sine qua non of successful marketing, but still just the beginning of a long journey. The availability of good data will ideally provide timely notice of certain customer events-say, the receipt of an annual bonus. However, unless the institution can craft the right value proposition for the right bonus recipient, it will not succeed as an event- driven marketer. Acquiring this capability depends in turn on executing the other seven steps in the continuous learning cycle-and doing so day in and day out as a matter of corporate routine.
Step 2 in the journey is the creation of valid measures of customer lifetime value. A bank is nothing more than a collection of customer cash flows, and the institution must be able to estimate, at any given time, the magnitude and duration of future flows and the resulting worth of each customer relationship to the shareholder.
Substituting customer contribution for lifetime value is convenient but not really adequate. Contribution is an accounting concept, while value is a financial term that serves as the raw material for stock price determination. The contribution of a typical mortgagor to the bank is negative in the first few years of the life of the mortgage (because of the high cost of loan origination), but the expected lifetime value of that same mortgagor, the more important measure, is usually strongly positive.
Many institutions have trouble measuring expected value. One problem, which also bedevils the accurate assessment of contribution, is the inability to assign realistic costs. In part, this is because of the dearth of valid transaction data. In equal part, however, it is traceable to methodological confusions about whether to allocate fully-loaded or marginal costs. Another difficulty, one peculiar to valuation analysis, stems from lack of clarity as to the rate to be used in discounting future cash flows. A surprisingly large number of financial institutions employ multiple discount rates instead of selecting that single rate which reflects what the shareholder can earn on other investments of comparable risk. (It is useful, indeed essential, to regard each customer as a kind of capital project in which the bank is investing its equity.)
Furnished with good data and valuation assessments, a bank can transition to Step 3, the brainstorming phase of the learning cycle. Here the marketer seeks to increase customer value by manipulating, either singly or in combination, the levers of revenue, expense, customer risk, and customer longevity. If the strategy is to increase value by improving the cross-sell ratio, the brainstormer first hypothesizes the components of a viable value proposition-that is, the combination of stimuli (e.g., product attributes, delivery mechanisms, and sales emphases) that will predispose a particular customer type to buy. This is essentially an exercise in a priori analysis, buttressed by interviews (with customers and salesmen), focus groups, survey information, and in some cases, the use of conjoint, a technique for determining the relative importance customers ascribe to discrete product features.
Having arrived at the set of factors that govern customer purchase, the marketer is like a chef who has assembled his ingredients and now has the job of combining them in the right measures needed to create a savory sauce. Suppose the marketer is working on a credit-card promotion. Suppose further that credit-card sales are determined principally by the interest rate, the annual fee, the credit line, and the gift program. Which combination of the above will tempt the customer while satisfying the bank's need to earn an adequate profit?
If there are three possible rate combinations (low, medium, and high), three possible fee combinations (same), two credit lines, and two gift alternatives, the total mix of possibilities sums to 36. So to determine which combination of the four factors is optimal for particular customers, the marketer would ideally have to test market 36 value propositions. That would be costly and time-consuming. And because it would take so long, the conditions that might predispose customers to accept one of the offers could easily change while the test was in progress, vitiating the usefulness of the entire exercise.
That being the case, the most sophisticated marketers are now using a statistical tool that economizes on the number of joint factor combinations without sacrificing a measurable amount of informational content. The technique, known as a fractional factorial design, requires a skilled statistician to select those few factor combinations-often only 8 to 10 out of a possible 36 or even 8 to 10 out of a possible 256-which will serve as surrogates for the sum total of possible combinations (Step 4). Thereby, the marketer can reduce the number of marketing packages (factor combinations) to a manageable proportion without adversely affecting customer response rates.
Having designed the most parsimonious set of value propositions, the marketer has to distribute each offer to a random sample of the bank's customer base (Step 5). Unfortunately this pivotal step is often impeded by staff-line friction. The people who design the value propositions are, of course, staff, while those who implement them are line. At best, line people will agree to be guided by statistics and data, but they resent being driven by them.
Yet the process we have just been describing is unquestionably a statistics-and-data-driven exercise that is meant to supersede a traditional and, for the most part, relatively unsuccessful selling approach. The resulting conflict can torpedo any marketing campaign unless the bank's senior management is adept in the art of change management-a skill in notoriously short supply.
Attitudinal problems aside, there is also a host of more mundane obstacles to surmount before any campaign can succeed. These range from control problems (the failure to hold constant as many variables as possible so that different response rates can be reasonably ascribed to the different marketing packages) to the overall inadequacy of the order-taking and fulfillment procedures. Many institutions simply underestimate what it takes to run a professional marketing campaign.
Assuming successful implementation, we move to Step 6, which requires the institution to collect and analyze the responses to the various marketing packages. This not only includes verifying that the offer was received and the sale made, but also whether the customer paid in a timely fashion and the bank earned what it expected to earn.
In Step 7, the marketer plugs the above response data into a multiple- regression equation that relates the customer's probability of purchase and the resulting increase in bank net present value to two types of independent variables: first, the nature of the marketing treatment (which package was received) and second, the bank's file of historical data on the customer (age, longevity, balance levels, transaction patterns, etc.) The marketer can then use the regression results-the computed coefficients of the independent variables-to anticipate which value propositions would be optimal for any given new customer, one who has not yet participated in the marketing program (Step 8).
Otherwise put, once the bank has found a compelling relationship between people with certain data base characteristics and a particular marketing treatment, it can assign other people with similar characteristics the same treatment with some degree of confidence that what worked for the first population will also work for the second. It is in this sense that the above process allows the institution to inexpensively individualize its customer treatments, giving each individual what he wants without actually interacting with him-a procedure celebrated in the oxymoron, "mass customization."
A final phase of the cycle (call it Step 8a) is designed to facilitate continuous change. In brief, because the responses of people to given offers will change over time, the bank does not assign 100% of its customers what it currently believes are the optimal marketing treatments. It randomly selects a small group (anywhere from 1% to 10%) and offers these the next generation of treatments being devised by its intrepid brainstormers and refined by fractional factorial analysis. Thus the cycle begins anew.
Does the continuous learning cycle work? Remarkably well, in our experience. One company boosted hitherto stagnant sales of a key product by 17% after only one marketing cycle. Another that was cross-selling insurance products to a credit-card population upgraded its performance from $4 to $10 per active cardholder within a year. A third used the methodology not to increase sales but to improve credit-card collections by 10%. Integrating the cycle into the fabric of institutional life is unquestionably tough, but, by all indications, the effort will be amply rewarded.
