Everybody would agree that data mining and modeling should be a process of continuous learning, not a series of standalone projects. But even within organizations that have standing resources to support data mining and modeling, the analysis is often much less valuable because the learning process is haphazard, not planned. Too often, new marketing programs are designed and implemented with no prior thought to learning; and then, after the results are in, the analyst is sent in to see what actually can be learned. To keep ahead of the competition, you have to have more than a commitment to learning – you have to proactively design a learning strategy that maximizes the rate of marketing improvement.


Figure 1: Impact of a Learning Strategy

This column describes a series of elements that can be part of a learning strategy. Some of these will be very familiar concepts. Yet in many marketing organizations even these are not implemented with regularity or effectiveness.

Adopt the Right Metrics

As one of my coworkers is fond of saying, "If you can’t measure it, don’t do it." At the most basic level, you must be able to know if something worked or not. And you need to know this early in the process. You don’t want to overcommit to a program that isn’t working or move too slowly on a program that works well. In addition to overall measures of success, diagnostic metrics need to be used that will indicate the reasons for poor performance and indicate ways to improve results. ROI is usually the most important single metric – after all, the purpose of marketing programs is to make a profit. But other metrics such as response, conversion, turnover, sales cycle time, customer complaints, etc., help pinpoint the reasons for ROI performance.

Test and Control

By "holding out" part of a target market from a marketing program, the performance of the program can be determined with a straightforward comparison of the program results vs. results within the holdout sample. Pretty basic stuff – but there are plenty of occasions when it isn’t done. The reason for this omission is not always a lack of planning. Marketing managers with aggressive sales targets can be unwilling to hold out a group of prospects that are then less likely to buy. One way around this is with champion/challenger testing. In this scheme, most marketing is done under "champion" programs – the current best practice marketing programs. New ideas for marketing programs are implemented on a comparison sample, and the results compared to the champion. If the "challenger" wins, it is adopted as the new standard. But there is never a group of prospects that receives no marketing.

Keep the Right Data

Metrics, tests and other data mining and modeling can only contribute to marketing knowledge if the right data is captured and available for analysis. In my experience, the key data element most often missing is the ability to tie program results to individual prospects or customers. For example, catalog marketers often print an individual code on each catalog mailed. When a customer calls the 800 number to place an order, they are asked to read off the code. This serves the dual purpose of allowing the sales representative to call up personalized information on that customer and of allowing the analyst to know what orders each catalog customer placed. (Perhaps the same could be accomplished by name and address matching of mail lists and orders, but this is expensive, time- consuming and less accurate.)

Generate Incremental Data

Don’t be satisfied with the data that is collected under old business practices. Instead, take every opportunity to capture additional information about customers and prospects. This has been done for years on warranty cards, where the purchaser is asked to provide name, address and also complete a simple survey about usage and purchase patterns, other interests, etc. (Unfortunately, this information is often neglected after it is received.) Many other means are now used to coax people into volunteering information. Web sites often offer small incentives or require information in return for services. Contest entries can also be used to collect prospect names. These activities must respect reasonable limits on the use of private information, but many legitimate and valuable strategies are available.

Offer Choices

One of the capabilities of data mining and modeling is to guide the choice of which offer to make to individual customers or prospects. But even when very good targeting information exists, you will never know exactly what offer would be best for each individual. Instead of making an offer that requires a simple yes/no response, make offers with a personalized choice of offers. This will probably increase sales, since the choices will be more likely to contain something the prospect will want. In addition, when a customer makes a choice they are providing valuable information about their preferences. A simple yes/no choice provides a lot less information than making a choice among several alternatives. Catalog marketing provides a good example. When a person buys dress shirts and neckties from a general clothing catalog, it won’t be long before they receive a catalog focused on business attire.

When you’re planning, include a plan to learn. When you’re setting objectives, include learning objectives. And when you’re measuring results, measure how much you’ve learned. After all, if you can’t measure it, don’t do it.

Register or login for access to this item and much more

All Information Management content is archived after seven days.

Community members receive:
  • All recent and archived articles
  • Conference offers and updates
  • A full menu of enewsletter options
  • Web seminars, white papers, ebooks

Don't have an account? Register for Free Unlimited Access