RFM Migration Analysis
A New Approach to a Proven Technique

by Jim Sellers and Arthur Middleton Hughes

The value of RFM (Recency, Frequency, Monetary) analysis as a method to identify high-response customers in marketing promotions, and to improve overall response rates is well known and is widely applied today. RFM has been around for more than forty years. Less widely understood, however, is the value of applying RFM scoring to a customer database and measuring how customers migrate from cell to cell over time. This article examines one approach to an RFM Migration analysis which was applied recently in a segmentation exercise at Federal Express. The analysis explained herein actually took place during the past year. The numbers used have been disguised to protect the confidential results of this program.

Why RFM works.

Customers who have purchased from you recently are more likely to respond to your next promotion than those whose last purchase was further in the past. This is a universal principle which has been found to be true in almost all industries: insurance, banks, cataloging, retail, travel, etc. It is also true that frequent buyers are more likely to respond than less frequent buyers. Big spenders often respond better than low spenders. These are the three simple principles lying behind RFM analysis. What skilled marketers have done is to take these three ideas and quantify them. They code all customers into RFM cells and examine the response rates of the customers in each cell when exposed to the same promotion. It is true, of course, that only a percentage of customers will make an additional purchase based on a new promotion. But, of those that do respond, the responses usually come from customers in higher ranking RFM cells. Federal Express has good data on customer purchase history. They used this data to code all customers by RFM. Then, they used this coding in a very interesting way.

Objectives

The goals of the analysis were to:

  • Identify patterns of migration from RFM cell to cell over time.

  • Determine the extent to which customer migration patterns fell into discernible clusters

  • Identify investment and marketing strategies appropriate to each migration cluster.

  • Assess the effectiveness of RFM migration vs. other segmentation and targeting strategies available for marketing promotions.

Migration means, of course, that some customers improve their performance over time. They move to a higher ranking RFM cell. Other customers regress to lower ranking RFM cells. Profitable marketing comes from anticipating the migration of groups of customers so that the marketing and service dollars are spent on higher value customers who will, in return, improve their spending and retention habits. Marketing dollars are thus not wasted on lower value customers who are less likely to migrate up.

The customers for this exercise at Federal Express were selected based on whether they had purchased any one of a particular "family" of express services in a two year period. Purchase transactions were summarized at a half year level for each of the four services. The analysis file included the summary transaction data along with other demographic or behavioral data such as SIC code, company size, age of account, discount status, and whether customers used a FedEx automation device for their shipping.

The first step was to determine which products to use in the migration analysis. Migration takes time. The goal was to select customers who used the service over both of the two years. The following chart shows the result of the preliminary analysis:

Activity Timeframe

Product Level Group % Accounts
One Year Both Years
Product A 58.2% 41.8%
Product B 59.7% 40.3%
Product C 85.2% 14.3%
Product D 94.2% 5.8%
Total Users 54.9% 45.1%

For the two years studied, Products C and D had only a small percentage of users in both years. For this reason, these products were eliminated from the migration analysis. The analysis, therefore, was based on three groupings: Product A, Product B and Total Usage. The purchase patterns of the customers were studied over the four half year periods.

The RFM score for each half year period was defined as follows:

  • Recency = 1 if the most recent purchase was in the first quarter of the half year period and 2 if the most recent purchase was in the second quarter.

  • Frequency = Total shipments during each half year period.

  • Monetary = Total monetary value during each period.

The RFM score was then determined by multiplying each of the above scores for each individual customer. This definition tends to dilute the impact of Recency on the RFM score, since 1 or 2 are much smaller numbers than the typical number of shipments or monetary value. Why was this done? Because Federal Express was primarily interested in the monetary value (dollars and shipments) of the customers for studying migration, rather than Recency of purchase.

Customers were then categorized into deciles based on their Period 1 RFM score. Deciles rather than quintiles were used to identify the relatively fine movements that were expected to appear during the balance of the analysis. Period 1 deciles (10 = best decile, 1 = worst decile) were used as the benchmark for monitoring the subsequent migration.

wpe5F.jpg (24646 bytes)

 

The third step was to complete the actual migration analysis. Migration patterns were identified using the statistical technique called Cluster Analysis. Seven clusters with common behavior patterns were identified. They were:

Stable customers – top 10% Top10
Lapsed 6 month customers – Medium Value LapMV
Seasonal Shippers – Low Value SeasLV
Growing Shippers – High Value GrowHV
Stable customers – Medium Value StayMV
Lapsed 6 month customers – Low Value LapLV
Reactivated Low Value Customers ReacLV

The object of the analysis was to determine how each of these groups migrated from RFM decile to decile during the two years. Once this was known, the goal was to use this knowledge to develop an appropriate marketing program for each group.

The cluster analysis clearly identified particular customer types. The Stable, Top 10% cluster were the best and most valuable customers. They had nearly three times the lifetime value of the average customer in the study, and were remarkably consistent in their behavior over time. They did not migrate up or down, but remained in the top 10%. What should the marketing strategy be for these good customers? Federal Express should work to retain them. They should invest sufficiently in services which will protect them against the possibility of defection.

Segment

Frequency

Growth Target

Reward

Group 1

Low

33% - 300%

$20 Coupon

Group 2

Average

15% - 100%

$25 Coupon

Group 3

High

10% - 38%

$30 Coupon

Group 4

Very High

Thank You Letter

The Growing, High Value cluster increased their monetary value by 1,500% over the two year period. Although this cluster was only one third as valuable as the Stable, Top 10% cluster, their migration pattern shows that they are clearly worth serious marketing attention. Few clusters could promise as good a return on marketing investment.

On the other hand, the Lapsed, Medium Value cluster experienced a 90% loss in average monetary value from Period 1 to Period 4. What went wrong with these customers? Did they go to competitors, or was their business declining for other reasons. Surveys and market research were appropriate for these customers. There is often more to be learned from failure than there is from success.

A very valuable part of the analysis resulted from identifying the Seasonal, Low Value cluster. These people only ship at certain parts of the year. Spending a lot of money to get them to ship at other periods would be a waste of marketing dollars. The marketing program should be timed to adjust to their schedules. That way, the marketing dollars would be far better used.

As a result of identifying these clusters, therefore, Federal Express was able to channel its marketing dollars where they would do the most good. Beside brainstorming possible marketing and investment strategies for each migration cluster, a comprehensive profiling exercise was conducted of each cluster. A number of clear characteristics were evident, including key differences in the frequency of sales contact rates, discount status, SIC classifications, etc. As a follow on to the findings of the migration study, a series of predictive models were carried out to identify customers in low value clusters who "looked" as if they should be in higher value clusters.

For example, modeling the customer in the "Growing, High Value" cluster against a look alike model built from the "Stable, Top 10%" cluster allowed FedEx to separate growing customers with additional upside potential from those who had reached the limit of their growth. The key differences between the two clusters, such as sales contact rates, automation status, discounts levels, etc. could then be worked into promotional programs designed to continue to grow these customers with untapped upside potential.

 

Conclusions

This kind of RFM Migration Analysis can easily be duplicated by any business engaged in database marketing. The benefits are:

  • You can identify changes in RFM behavior patterns that would be invisible with the relatively static traditional application of RFM as a response improving technique

  • RFM migration can be a valuable segmentation tool alongside your traditional segmentation approaches

  • To the extent that you are able to identify or model key differences between the RFM clusters, the output can provide a clear course of action for the marketer.

  • Investment strategies and marketing pro-forma are much easier to produce with the wealth of customer behavior and value information generated by this kind of study.

If you are considering RFM migration analysis, what are the points you should keep in mind?

  • Have a set of business objectives and be prepared to modify your methodology as new information becomes available

  • Before your original definition of the RFM score is precise enough to identify fine changes in behavior. For this analysis, deciles and better than quintiles. Don’t however, make your definitions so fine as to prevent action later. The typical 125 cell approach used in RFM response analysis scoring is too fine to yield sufficiently distinct clusters. Recency, which is usually the most powerful factor in normal RFM analysis is less important in Migration Analysis than frequency or monetary behavior.

  • Marketing analysts and marketers must work closely together during migration analysis. Why? Because the analysis must be focused on how to use the output later in a way that is meaningful to the customer. Database marketing only works if the customer benefits from it. It is useless to identify a cluster unless you then use that knowledge to adjust your marketing investments up or down and in creative ways that will be meaningful to your valuable customers.

Using RFM Migration analysis, you will be able to identify opportunities to create marketing messages that are relevant to your customers. You will be rewarded with increased business and improved customer satisfaction.

 


Arthur Middleton Hughes is Vice President of The Database Marketing Institute. Ltd. (Arthur.hughes@dbmarketing.com) which provides strategic advice on relationship marketing. Arthur is also Senior Strategist at e-Dialog.com (ahughes@e-Dialog.com) which provides precision e-mail marketing services for major corporations worldwide. Arthur is the author of Strategic Database Marketing 3rd ed. (McGraw Hill 2006). You may reach Arthur at (954) 767-4558 .


The articles on this web site are available to the general public to read, enjoy and for limited business use. If you want to reprint more than one or two of them for resale or use in a business or educational environment, send an email to Arthur Hughes at arthur.hughes@dbmarketing.com. He will give you permission by return email. The cost, depending on the number of copies you want to reprint, is very inexpensive.