The Basics of RFM Segmentation - RFM Values
This section discusses the definitions of recency, frequency, and monetary values that are used to create RFM Cells.
RFM Values versus RFM Cells
Before we can create RFM cells we have to have RFM values - the actual date, number of orders, and sales amount for each customer, such as 138 days since last order, 7 total orders, and $1,465.27 in sales. Once gathered these values are used to assign an RFM Cell.
Recency is simply the date of the most recent order. This is very straightforward.
Recency is the most powerful predictor of who is likely to order. The customers who ordered this year are more likely to order than the those who ordered last year. The customers who ordered today are more likely to order than the customers who ordered a week ago. This fact is so important that it might be called the "central dogma" of Direct Marketing.
Recency groups are sub-segmented by frequency, the next most powerful predictor of response. Given the same recency, customers who have ordered two times are more likely to respond than one time buyers.
If you have lots of historical data going back many years I do not recommend using lifetime orders for frequency and monetary values. Lifetime orders are misleading - it gives too much weight to activity in the far past. This very old activity is not a good indicator of current responsiveness.
It is better to limit frequency to the orders placed within a fixed period of time before (and including) the most recent order. If this period is set very long (not recommended), the effect is essentially the same as using lifetime orders. We can favor very active customers if the period is short, say 12-18 months. Three years is a good default value.
So, using the three year number, for a customer who last ordered two years ago, we would count orders back to five years ago. For customers who last ordered six years ago, we would count orders back to nine years ago.
Monetary value can be used to further segment a recency/frequency group, although its effect is not as strong as the other two factors. It is the weakest predictor of performance.
Many people use total order value to define monetary. This surprises me because it is so strongly correlated to the number of orders. I define monetary value as the average order size of the orders that are used to compute frequency. Using average order size makes monetary an independent variable.
I tend to think of monetary as a tie breaker for cells with the same recency and frequency: all else being equal, customers with a greater average order size are more valuable.
Another use for monetary is to avoid mailing very low value customers. Some companies make low value offers to add new "suspects" to the file. Since they showed interest and sent in some money they are well qualified as prospects, but they are not really customers yet. These names can be segregated for special treatment using monetary value.
Next section: RFM Cells and Cell Groups