The Iron Rule of RFM

Few rules of thumb are so deeply embedded in our thoughts that we’re surprised to recall that they’re really just rules of thumb, not scientifically proven facts.

And for many, that’s the case for this month’s rule of thumb: the rule of recency, frequency, monetary value (shortened to RFM). As catalogers we use RFM constantly, almost without thinking about it, not because psychologists have proved to us that it should work, but because as marketers we know that it simply does work, day in and day out, and has been working since the earliest days of cataloging.

As with most everyday things though, a closer look turns up a variety of unexpected twists, which we’ll explore in this column devoted to RFM.

Let’s begin by defining the rule:

The “RFM” Rule of Thumb

You’ll get higher response rates when you mail to customers who have (a) bought more recently, (b) bought more often and (c) spent more lifetime dollars.

Conversely, you’ll get lower response rates from mailing to customers who have bought less recently, less often and spent fewer total dollars.

Is RFM obsolete?

If you’re thinking that RFM is too old-fashioned for our post-modern world of neural nets, synergy models and so on, think again. Buzzwords like CHAID, regression, data mining and the like all refer to analytical techniques, but to function, these analytical techniques must be applied to real-world data. And that’s what RFM is—three vital chunks of real-world data that we maintain for each customer, and that we use to predict how they’ll respond to our offers.

In fact, RFM is a central part of the power of most current statistical tools used by catalogers. Take RFM data away, and the power of modern statistics for catalogers would be drastically reduced.

How is RFM measured and stored?

The RFM rule is relative—that is, it refers to “higher” response rates, “more recent” buyers, “higher” lifetime sales and so on.

But computer databases can store only specific, absolute numbers, not abstractions like “higher” or “more.” So to save RFM data about our customers, most modern mail-order software packages store three specific data variables for RFM in each customer’s master record:

1. Date of most recent contact (to measure recency).

2. Quantity of lifetime purchases (to measure frequency).

3. Total lifetime dollar sales (to measure monetary value.) Actually some catalogers prefer to use only total sales during the last 12 months for this measure, but they are in the minority.

For example, if we looked up John Doe in our master customer file, we might find (among other data) these three pieces of data (separated here by commas and enclosed by double quotes):

The first is recency—John last bought from us on June 6, 1999.

The second is frequency—John has bought from us five times.

The third is monetary value—John has spent $230.35 with us since he first became a customer.

How can I use RFM to increase my catalog response rates and sales?

In a nutshell, you will boost your overall response rate and total sales if you mail more often to customers with “higher” RFM values, and less often to

customers with “lower” RFM values.

How much more, specifically, can you gain from this strategy? Absolute numbers will vary from cataloger to cataloger—but in general, the older your house file is, and the less rationally you’ve been mailing it—for example, if your list is 4 years old and you’re still mailing equally to every name on it—the bigger the benefits that RFM-based mailing will bring. Every cataloger I’ve ever seen who switched to RFM-based mailing has concluded that the improvements were “very significant.”

Why is RFM such a good predictor of future customer buying behaviors?

This is an interesting question—and the closer you look, the more interesting it becomes. Intuitively, it’s fairly easy to rationalize why M (monetary value) is a good predictor of a customer’s future buying behavior: If Mary Smith has a high lifetime value, chances are she’s happy with what you offer, has a continuing need or desire for it and will probably buy again.

Of course we can’t know that for sure—Jane’s last big purchase may have finally satisfied all her needs, and you may never hear from her again. But big spenders like Jane tend to spend again—hence the power of M for predicting future sales.

And also intuitively, it’s fairly easy to rationalize why F (frequency) is a good predictor of a customer’s future buying behavior. If Mary Smith has ordered from you several times, chances are she is getting what she wants and will probably buy again.

Of course, once again we can’t know that for sure—Mary’s need may have terminated with her last purchase, and she may never buy again. But generally, repeat buyers keep repeating.

Which brings us to recency—and intuitively, this is the real puzzler. The rule of RFM says that if Mary Smith has bought recently, she’s highly likely to buy again, right away.

And for me, that’s not intuitively obvious at all. If Mary buys a new outfit, why should that make her likely to buy another outfit right away? Won’t she be feeling a bit poor from her recent purchase, won’t she perhaps wait to “save up” for the next purchase? And even if Mary isn’t feeling poor, won’t she be diverted from buying again for a while as she enjoys her recent purchase?

My intuition answers “yes” to both questions.

But clearly, my intuition is wrong—because in fact, recency isn’t just a good predictor of a high likelihood to re-buy, many catalogers consider it to be the strongest predictor of the three RFM variables.

Why? Perhaps buying something puts people into a “buying mood,” so they feel like buying again and again. Or maybe a first purchase creates a need for additional, supporting purchases—a new sweater to go with the new skirt, then new shoes to finish the outfit, then a new necklace to match. Whatever the reason (and no one really knows for certain why recency is such a powerful predictor of future sales), recency is a very strong predictor of immediate re-buying.

Which is why when you rent lists, you will almost always do a recency select (12-month buyers, six-month buyers, hotline buyers), but you will only sometimes do a dollar value select, and least of all will you do a frequency select.

What practical issues must I deal with to use RFM data to boost response?

The rule of RFM says that customers with “stronger” RFM values will produce “higher” response rates and sales. Which is intriguing, as far as it goes.

But to create real-world mailing plans, we usually need to know much more—specifically, we need to know exactly how to rank our customers, and exactly how much stronger our results will be from our higher RFM-value customers.

And the only way to get that kind of information is to start coding your mailings.

Specifically, you need to devise a mailing code system that encodes each customer’s individual RFM values into their mailing code (see Figure 1 on p. 62). That mailing code should be applied to each catalog you mail, often in a yellow or pink box (which is where we as catalogers have trained our customers to look for it.) Then, as orders arrive, your operators should ask each customer for this code and enter it into each order, along with the customer’s name and address. Then, at season’s end, you should produce a sales report with a separate row for each code, so you can see individual response rates and sales for each mailing code (see Figure 2 on p. 62).

This is easier than it sounds—you can do the whole thing on a simple spreadsheet, and when it’s done, you’ll be able to see exactly what combinations of RFM are strongest for your business and your customers, you’ll know exactly how much stronger some segments are in real dollar terms and you’ll be able to figure out who to mail more to, and who to mail less to.

What’s a good coding scheme for including RFM data in mailing codes?

Your catalog software is already saving and maintaining RFM values in absolute terms for each of your customers—that is, it knows the exact day that each customer last bought, the exact quantity of lifetime orders and the exact dollar value of lifetime sales.

But we can’t use those exact numbers directly for coding, because we’d need literally thousands of different codes to distinguish between customers whose lifetime value was $200, $201, $202, etc.

To cut the number of mailing codes to manageable size, we need to create RFM GROUPS.

Figure 1 on page 62 shows a simple mailing code system that divides customers into a reasonable number of RFM groups.

For RECENCY, rather than coding by each specific date of most recent contact, this system codes by YEAR Why not by half year or by quarter? Many catalogers use finer recency gradations, but if you’re just getting started, coding by year will give you a useful number of recency groups without overwhelming you with codes. After you gain experience with how your customers are responding, you can try going to finer recency groups.

For FREQUENCY, rather than coding by individual lifetime order count (which again would require too many codes), we usually create just two groups: one-time buyers and multi-buyers.

Why not assign different codes to three-time buyers, four-time buyers, five-time buyers and so on? Larger catalogers do this, but the major difference in response comes between one-time buyers (who may only be testing you out, and may yet be disappointed by what they receive), and multi-buyers (who have come back at least once, proving that they really are satisfied.) So smaller catalogers often opt for the greater simplicity of just two frequency groups.

For MONETARY VALUE, rather than coding by individual lifetime dollar sales (which again would require too many different codes), we create instead a small number of lifetime dollar sales groups, (e.g., $0-$50 buyers, $51-$100 buyers, etc.) What specific dollar thresholds should you use? If you’ve been in cataloging for a while, you probably have a good sense of where the breaks should be.

For example, if you’re a gift cataloger with an average catalog price point of $30, you might try segments like this: $0-$30, $31-100, $100+. The goal is to select thresholds that combine similar buyers into one group, and divide dissimilar buyers into different groups.

So, by bringing all the above together, we have about five recency groups (current year, one year ago, two years ago, three years ago, four or more years ago), two frequency groups (single-buyer, multi-buyer), and three monetary groups ($0-$30, $31-$100, $101 and up), which means we will need just 5x2x3=30 different mailing codes, which means you’ll need just 30 rows in your end-of-year “Sales by Mailing Code” report.

What does a sales report for RFM analysis look like?

Figure 2 below shows part of a typical sales report that breaks out sales and response rates by mailing code.

By studying the columns for response rate, sales per catalog, and profit per catalog, you can see that this cataloger is leaving money on the table by undermailing his top performing RFM groups, and by overmailing his worst-performing RFM groups. This cataloger could significantly boost his results, even without printing any more catalogs, just by mailing less to the RFM groups identified in this report as weakest, and using the freed-up catalogs to mail more often to the RFM groups that this report shows to be more responsive.

What trouble spots should I avoid when using RFM to create a mailing strategy?

There’s really only one: You must take special care when purging duplicate names from your house file.

The reason is that modern cataloging software saves variables for RFM directly in the customer master file, along with each customer’s name and address. So if John Smith appears twice in your file both as John Smith and as J. Smith, you can’t just erase the record for J. Smith, because that would also erase whatever RFM data has been stored in J. Smith’s record. Before removing any dupe, you must first intelligently transfer the dupe’s RFM data into the surviving non-dupe record.

So for example, if John Smith’s date of last contact was 6/9/1999, frequency 5, monetary value $200, and J. Smith’s date of last contact was 8/9/1999, frequency 2, monetary value $50, the new combined John Smith’s record should be adjusted as follows: date of last contact should become 8/9/1999 (the most recent date in the two records), frequency should become 7 (the sum of the frequencies in the two records), and monetary value should become $250 (again, the sum of monetary values in the two records.)

If you aren’t transferring data in this way before purging duplicate names, you’re making your better customers look worse (by throwing away part of their RFM data). This will cause you to mail your best customers less than you should, reducing your sales and raising your costs.

When can RFM can be ignored?

If you’re just getting started in cataloging, you can’t use RFM to guide your mailings for a while, because you won’t have any RFM data to guide you (all your customers will be current-year single buyers with relatively low lifetime sales).

But that doesn’t mean you should ignore coding. If you start coding your mailings immediately, even when your customer file is small and new, you’ll be all ready to take advantage of RFM as soon as your house file becomes large and old enough to benefit from it, which should be two to three years after you start.

Today’s complex statistical techniques can produce remarkable improvements in response rates and sales—at a price that is almost always rather high.

But RFM is almost always effective and usually quite inexpensive too, since RFM analysis uses only data that your cataloging software is probably already saving for you, and you don’t need to be a statistical guru to make it work.

Cheap, easy to understand, reliable, foolproof—that’s the magic of the RFM rule of thumb.

Susan McIntyre is president of McIntyre Direct, a catalog consulting company based in Portland, OR. Author of the regular column “Cataloging Rules of Thumb,” she can be reached at (503) 735-9515.