Jim answers questions from fellow Drillers
Hi again folks, Jim Novo here.
Sometimes the traditional RFM model does not work very well for a specific business model. For example, small business databases can be too small to fill out all 125 RFM segments properly, resuslting in distortions of predictive capability. Optimizing the traditional RFM approach for unique business model criteria is a very useful skill, and it’s actually not difficult if you understand the levers of the business model. To Drill or not to Drill, that is the question …
Q: I’ve used your site a lot and found it to be very informative.
A: Thanks for the kind words!
Q: I have a question about the use of RFM analysis for a low margin, eCommerce business. I read that for a relatively small customer list (<50k) using just the “RF” of the RFM analysis would be preferred since the “M” tends to hide shifts in behavior.
A: Well, the M tends to smooth shifts regardless of the size of your list. In addition, if you have a small list, 125 segments is too many to be really useful, so RF at 25 segments in more intuitive. The real issue with M or Monetary Value is up and coming, accelerating customers. If you use total spend (M), it will “punish” them with a lower rank. But the fact is they have more future potential because Recency is low and Frequency is ramping. Inversely, M tends to reward customers who have spent a lot in the past with a higher rank, though they may actually be declining or defected customers. Predicting the future is more profitable than reporting on the past, so given a choice, I would drop “M.” This is especially true on the web, where communication costs are low and changes in behavior can be very rapid.
Q: My question to you is, since I’m talking about a low margin business, wouldn’t “M” actually be more valuable than “F” for the analysis? For example, if 40% of my customers are driving 70% of my sales and 100% of my profits, that says that 60% of my customer base is losing me money. I don’t want them to be given a higher value rating because they’re placing MORE unprofitable orders than someone placing fewer but profitable orders. You see what I’m saying???
A: Absolutely, and you have just proven to me you really understand the concept. It’s a tool. The more you can customize it to your situation, the better. There is actually some discussion of this situation in the book, the idea of “M” as a “check digit” on profitability rather than using F, if the business is low margin or certain very popular items are “loss leaders.” It’s not common, but this model does exist, for sure.
Q: Does that then support my belief that an “RM” analysis would be more appropriate?
A: Well, I’m not sure I understand your situation completely, but if I’m getting it I would be more likely to use Recency-Gross Margin because if I’m hearing you correctly, you sell some (perhaps many) items at a negative profit. However, some of those customers may go on to buy profitable items, and I would want to consider that. So I wouldn’t use sales, it could be deceiving; I would use cumulative Gross Margin.
In the end, there are 2 components to this model: Recency, which predicts likelihood to buy or visit again (future value), and the “Money” variable, which indicates how profitable the customer is to you now (current value). You can plot the two variables on a two-dimensional space and literally “map” the current and future value of your customer base, and then use this knowledge to make marketing or service decisions.
You should design the money variable to be the one that makes the most sense for your business, according to your model and available data. If total page views are your measure of a value of a customer (ad supported site), you use Frequency of visit for current value. If you are selling products with an evenly distributed price scale and roughly the same profit margin, you can use M. Recency, or sometimes Latency, are used to measure the future value of the visitor or customer.
Frequency is actually a “tweener” variable, it has implications for both current and future value. But the largest predictive power of Frequency is really in the distinction between one-time and multi-buyers.
So, if you have a small list you might want to score one-time and multi-buyers each by themselves. This will buy you a lot of the power of the Frequency variable without having to mess with 3 variables and the 125 segments in the traditional RFM model. The one-time buyers you can simply score on Recency and the rest you use RF, R-GM, or whatever financial metric makes sense for the biz.
If I have failed to explain this sufficiently, please let me know!
Download the first 9 chapters of the Drilling Down book: PDF