Pitfalls in Using RFM technique in Response Modeling
We’ve seen before how RFM (Recency, Frequency, Monetary) technique evolved from mail catalog retailers and its popularity because of its common-sense premise, its rather relative easy statistics associated with it, and the easy availability of the data with most mail-catalogue retailers, and other similar businesses interested in increasing the response to their direct marketing.
But is it really the best technique to predict response to a direct marketing campaign?
Let us examine some key pitfalls for the use of this technique for this purpose:
1. First of all, predicting the responses rate for a direct marketing campaign is a problem of predictive modeling whereas RFM tells you only how the customers have behaved in the past – When was the last they bought anything? How frequently do they buy? How much of totally buying have they done so far?
It tells you nothing about whether how they will respond to this particular campaign.
2. That response is predicated on the assumption the assumption that your best buyers will continue to be the best responders. And that brings us to the second major pitfall. This assumption ignores the fact that the customer behavior might change over time or might have already changed.
I’ll give an actual example. A seafood retailer observed that its best customers in terms of monetary value and frequency were lying in the lower recency segments – that is, they had not come for some time. So it sent them many flyers and expected that the majority would come back. But the results were far from spectacular. What had happened? It seemed that the retailer had overhauled its assortment and positioning, moving from a cheap everyday fish provider to an exotic premium retailer of seafood. The attrited customers belonged to the market that had patronized the retailer for its economical prices. This, of course, could never have been gauged from the RFM analysis. Now, either the customer continues with its realigned strategy and shrug these customers off as collateral damage or it go back to serving these customers also. Whatever it does, just sending them flyers would not work. This is a call to be taken at the strategic level and RFM does not offer enough information to take that decision.
3. And that brings us to the next pitfall, which is that RFM is too basic a technique and does not describe the customer behavior well enough. The propensity of a customer to respond to marketing stimuli is predicated on a variety of factors other than recency, frequency and monetary value – for example demographic information, geographic spread, and, if available, response to similar marketing campaigns in the past.
Think of a situation like this recession where beside a RFM analysis, a value retailer might also want to consider demographic variables which indicate the customer’s financial situation of he’s offering some really great discounts in his marketing campaign.
Hence, it is always advisable that even when is building the response model on RFM technique, one add and test as many other relevant variables beside these three attributes.
4. Customers are rarely distributed evenly along the Recency, Frequency and Monetary distribution curves. For example, in the picture, we see the distribution of frequency of visits for a retailer.
As seen, while the frequency ranges from 1 to over 500 visits over the lifetime, it is pretty bottom heavy. In fact, customers with less than 4 trips make up 50% of the customers and customers with only 1 trip make up 26%.
If one had applied RFM these 26% customers would definitely have come in the bottom segment, and possibly even the 50% below 4 trips, and been ignored. Many of these customers might have such low frequencies is that they’re simply new customers (The retailer should then consider their high recency) and, if not, customers who have tried the store once or twice and still are unconvinced.
If they are the right kind of customers (as can be ascertained by demographic/ behavioral profiling), the marketing program should definitely target them and win them over. Unfortunately, RFM might never lead you here.
5. Another shortcoming of RFM technique is that while it points out the most engaged customers it does not tell you why these customers are so. Now think you’re expanding to a new geography and wanting the same engagement with the customers as your current store is. What are the attributes that customers see in your store driving that engagement? Can you replicate your success without knowing that? And what if something changes and you unknowingly slip in those attributes – how would you retain your current best customers if you don’t know why they were there in the first place?
6. Lastly, the technique totally ignores the new customers. If one keeps using this technique to make one’s mailing list, there will never be any new customers targeted.
To conclude, while RFM is a simple way to look at currently engaged customers, one should not stop at this technique and define customers more definitively in terms of demographic and behavioral classification, run surveys and determine why they favor the particular retailer and what they still find missing there, and keep taking a dipstick on changing trends.
And, wherever possible, I would recommend that response modeling should be done by building a scorecard on as many of these variables as one can collect. RFM should be used only when a proper scorecard is not feasible to build.
Interested in a career in Retail Analytics? Explore Jigsaw Academy’s Retail Analytics Course.