At its core, the RFM model compares your customers' purchases made to each other against three dimensions of data: Recency (R), Frequency (F) and Monetary (M).
Recency (R) is scored between 1 (worst) and 3 (best), Frequency (F) and Monetary (M) are both individually scored between 1 (worst) and 5 (best). This is useful to marketers because it lets you target and report on groups of similar customers.
Recency (R)
Recency refers to the time in days since the contact's most recent purchase.
A different method is used for Recency compared to Frequency and Monetary. This is because the time since the last purchase increases when a contact does nothing – unlike the frequency (number of orders) and monetary metrics (money spent) that only increase when a contact acts. High amounts of single purchase contacts, and contacts that remain inactive, mean the list of unique values for Recency is increasing over time.
This problem means that the method used for Frequency and Monetary would cause the range of values for the score brackets to consistently increase in size at the same rate (by one day each for every five days). Therefore, the highest-scoring bracket, five, is covering a larger and larger period of time and will eventually become unrepresentative of when a contact is actually considered active.
Instead, Recency is scored from one to three with each number representing a stage in the customer purchase lifecycle.
Recency score | Lifecycle stage | Description |
3 | Active | The time window during which a repeat purchase is most likely to occur. |
2 | Lapsing | For contacts whose recency has passed the active time window and a repeat purchase is less likely. |
1 | Inactive | The most recent purchase was long enough ago that the contact is extremely unlikely to make another purchase. |
The Recency ranges for these time windows are dynamic based on the amount of time between concurrent purchases for repeat contacts.
Frequency (F) and Monetary (M)
The Frequency and Monetary scores represent the number of orders and the amount of money spent by the customer respectively.
For each metric, a list of all unique values for all contacts is compiled, ranked, and split into five quintiles. A quintile here means that five slices of the data are made.
The contact is scored for each dimension based on which quintile they fall into. The distinction of unique values here is important, as it prevents situations where two contacts with the same value for a metric can be given different scores. Dotdigital's model irons out these micro variances which would cause inaccurate profiling.
To visualize this specific problem, below is a table showing how contacts would be scored on Frequency by ranking using all contacts compared with ranking using all unique values.
Contacts scored and ranked by Frequency
Contact ID | Number of orders | Score (all values) | Score (unique values) |
001 | 8 | 5 | 5 |
002 | 6 | 5 | 4 |
003 | 6 | 5 | 4 |
004 | 4 | 4 | 3 |
005 | 4 | 4 | 3 |
007 | 2 | 3 | 2 |
008 | 2 | 3 | 2 |
009 | 2 | 3 | 2 |
010 | 1 | 2 | 1 |
011 | 1 | 2 | 2 |
012 | 1 | 2 | 1 |
013 | 1 | 1 | 1 |
014 | 1 | 1 | 1 |
015 | 1 | 1 | 1 |