Segmentation Strategies That Work
You invest a significant amount of time and effort to maintain your customer marketing database.
Indeed, you have tightly monitored procedures to be sure all new activity is applied correctly, customer summations are accurate, and the names and addresses are kept current. In time, your database has grown in both size and scope as your business has evolved.
But are you getting the most out of what you’re putting into it? As you consider the wealth of information you now maintain on each customer, have you taken the necessary steps to leverage all available data elements to optimize your customer mailing selects?
It’s common practice for catalog marketers to stick to a single “tried-and-true” housefile selection scheme, usually based on recency, frequency and monetary (RFM) criteria. While some tweaking may occur periodically, select criteria essentially stay the same in time and across all campaigns.
On the surface, the ongoing use of a consistent selection scheme appears to be a good thing. It provides a trail of segment history that can be used to forecast future mailing performance. However, this same static structure could be putting a major dent in your response and profits.
First and foremost, segmentation should be all about discerning likely responders from those unlikely to respond. Circulation-performance results from your top to your bottom segments will be relatively flat if your current selection criteria are less than optimal, because you’re co-mingling your better and weaker performing names with each segment.
How do you upgrade your housefile-segmentation architecture? Setting up in the mail keycode tests to try new criteria won’t get you there. There are far too many possible variable combinations to test.
Instead, do an analysis of predictive attributes (APA). An APA is an analysis that’s done on one or more of your previous mailings. The actual results of the previous mailings are measured and then used to rework your housefile-selection parameters for future campaigns.
Using an APA, you’ll be conducting a complete audit of all possible customer-segmentation variables and will be able to measure their projected response and sales predictability. You may want to enlist an analyst or statistician to help you drive the project, but you don’t have to be a statistical genius to interpret the findings or act on them.
With the APA, each customer characteristic on your database will be measured individually. All key variables that influence purchasing behavior will be identified by using actual results from prior campaigns. Those that are found to be predictive of response will serve as the building blocks for the actual segmentation schemes to be used on future mailings.
Following are the steps to take to conduct the APA project:
Choose a completed promotion to use. It’s best to start with one of your deeper customer mailings, because you’ll have more data with which to work. All (or nearly all) purchases resulting from the mailing should already have occurred and been applied to your database.
Recreate customer characteristics. You may not be able to use the customer information residing in your database for this. Instead, recreate a snapshot of your database to mimic how all data looked at the time when the names were selected for the mailing you’re using for the analysis.
Identify all possible variables. This is where it can pay big dividends to be creative. Are there other characteristics beyond those that currently reside on your database that may be predictive of response? A few possibilities to consider include the following:
• Create variables to summarize recent activity (such as purchase dollars within the last 24 months only) in addition to typical life-to-date summations and averages.
• Overlay external demographic variables such as age, income, family structure, home ownership, climate and region.
• Append cooperative database model scores.
• Don’t overlook other possible types of information, such as response channels, prior back-orders, promotion history, retail trade areas, address deliverability scores and gift purchasing.
Generate predictive attribute reports. By matching the file of customer characteristics to mailings and corresponding orders, you can create report tables to show sales per mailing performance for each variable from the campaign. The reports will look like the one shown in the chart on page 27.
Review your reports carefully; They’ll offer both a profile of your customers and a view of the characteristics that relate to response. During the review process, sort out those that are most highly predictive to be considered as finalists for your new segmentation architecture.
The short list of remaining variables must be looked at in combination with one another. Start with some simple cross-tabulation reports. It often works best to force recency and frequency, and continue to build from there. With every cross-tab combination, generate performance reports as if these criteria actually were used for the mailing.
• Set realistic expectations. It’s said that RFM is an 80-percent solution. Out of the APA project, you’ll uncover some refinements to enhance your current RFM selection criteria. But don’t expect RFM to be replaced; rather, it should remain the backbone of your selection parameters.
• Experiment with different rollup ranges (e.g., $0 to $15). Nothing states that $25, $50 and $75 must be the purchase dollar cutoff points you use, or that all four-time or greater multibuyers should be treated as a single customer group. Identify the ranges most appropriate for your catalog.
• Concentrate on the customer groups in which the greatest gains can be recognized. It may be best at this time to focus on the middle tiers within your housefile rather than on the loyal buyers at the top or the inactives at the bottom.
• Recognize that the APA reports only show counts of who actually had been mailed. Up to this point, the analysis project has been limited to names mailed in the campaign that you used for analysis. Were the other potentially good responders bypassed when you ran the actual mailing select that would now be chosen under the newly devised selection scheme?
• Don’t assume that your new select criteria will be optimal for all mailing campaigns. Validate your new selection criteria against other past mailings. Repeat the APA process to modify selection criteria for in-season vs. out-of-season, early vs. late holiday, etc.
• Consider multivariate segmentation modeling. If combinations of five or more distinct variables surfaced out of the cross-
tabulations, develop housefile segmentation models. Additional
performance gains should be recognized because a model will determine the appropriate relative weight for each variable when scoring the names.
Keith Pietsch is director of analytics for Donnelley Marketing Catalog Vision. You can reach him by e-mail: Keith.firstname.lastname@example.org. Or call (952) 541-6548.