The ABCs of RFM

Dec 06, 2004 8:02 PM  By

Recency, frequency, and monetary (RFM) database segmentation has been proven to work. Yet many smaller companies—those with less than $15 million in annual sales or fewer than 50,000 house file names—don’t both with RFM modeling. They simply mail to everybody in their database. And that, says George Hague , senior marketing strategist at Shawnee Mission, KS-based catalog agency J. Schmid & Assoc., is a mistake.

Smaller mailers typically cite system limitations, time restraints, and a lack of manpower or knowledge for failing to apply RFM to their house file. But creating an initial RFM model shouldn’t take much time or money, Hague insists. And once you’ve built it, subsequent “list pulls” will become easier.

An RFM model assigns a numeric value to each RFM factor. When you add up each RFM value, the best customers receive the highest scores. Customers who may need an incentive to purchase are clearly identified and can be targeted for offers. Unprofitable customers receive the lowest scores and, after testing, can be removed from the mailing list.

Begin your model on a piece of notepaper. Start with what is commonly considered the most important factor: recency. For customers who have purchased within the past three months, assign a score of 6. Don’t worry about how many customers fall in this category. Right now, just assign the values.

For the customers who have purchased within the past four to six months, assign a score of 5. Next, you have the 7- to 12-month buyers with a score of 4. The 13- to 24-month buyers receive a score of 3, the 25- to 36-month buyers a 2, and the oldest buyers a 1.

Next, move to what is usually the second most important variable: frequency. If you have a feel of your range of repeat purchases, begin your rankings. If you don’t feel comfortable with your knowledge level for this factor, look at your database to get a feel for your customers’ frequency of purchases. An office supplies company, for example, will likely have more frequent and repeat purchases than a seasonal gift catalog.

If you have a significant group of customers who purchase once a month, start with the 12x-plus buyers as the highest scoring group. For many catalogers, a 5x-plus group is a good starting point. Assign 4 points to the 5x-plus buyers, 3.5 points to the 4x group, 3 points to the 3x group, 2 points to the 2x group, and 1 point to the first-timers.

When assigning frequency scores, be sure not to start your frequency scoring at a higher point value than your recency scoring, since recency is usually a more important factor than frequency. As a good starting point, have your highest recency value outscore your highest frequency value by a factor of 1.5. In turn, have your highest frequency value outscore your highest monetary value by a similar factor. If you believe your frequency of purchase needs a greater scoring range, go back and adjust your recency scoring to fit with your frequency scale.

Next, assign your monetary scoring. As a general principle, using your customers’ average order values (AOVs) is more effective than using their cumulative monetary value.

To begin your scoring for the monetary factor, start with your average order value across all of your customers. Use that as the middle score in the ranking.

Let’s say your AOV is $85. Give the $75-$100 range a value of 2 and work from there. In this scenario, $100.01-$150 would get a value of 2.5, and more than $150 would get a value of 3. To round out the bottom of the chart, give the $50.01-$74.99 orders a value of 1.5, and score the $0-$50 orders a value of 1. Once again, don’t let the monetary ranking out score either the recency or the frequency rankings.

It may take a few tries to get your entire scoring system to fit your database. Also, it doesn’t have to be perfect. It just has to be reasonable to get started.

After you work out your scoring system, bring it to your IT person. Explain that you want this scoring system to be added to your customer database, so that your RFM segmentation is now an automatic part of your list pull, segmentation and marketing process. This should require adding four fields that draw from the existing fields in your database. You will have one field for each factor’s incremental score and one field to give the sum of the three rankings. The sum field is the RFM model score and the basis for your list segmentation. For all future list pulls, you will need to include the RFM score as a field for the pull.

Your next step is to come up with a standard segmentation of this scoring. You will use this segmentation to assign key codes for tracking response.

The simplest approach is to sort the database in descending order with the highest scores at the top. From there you divide your database into 10 equal segments. The highest scoring group of names is segment one; the second highest scoring group is segment two, and so forth. Assign a unique key code to each segment for tracking purposes.

If you mail your database with these ranking assignments, and your scoring is properly weighted, you should see a standard response curve with segment 1 outperforming segment 2. Segment 2 should outperform segment 3, and so on. You should also see a clear delineation where the segments stop being profitable.

To test the validity of your model, pull a control group selected on an nth-name basis from across the entirety of the model. Each time you mail during your testing period, pull and mail this nth-name sampling. To smooth out your response curve, you may need to tweak your factor values in your model. Remember, your goal is to be able to predict which segments will be profitable.

Repeat this technique of segmentation and mailing throughout your mailing season. In addition to defining unprofitable groups, you should also be able to identify marginal segments. These segments may not be profitable now, but you can test reactivation offers on them.