Which RFM binning method is best for me?

[#1]

RFM stands for Recency, Frequency and Monetary. It is a technique used to identify existing customers who are most likely to respond to a new campaign or product offer. This technique is commonly used in direct marketing.

How does the RFM analysis work?

1. Customers are assigned a recency score based on date of most recent purchase or time interval since most recent purchase. This score is based on a simple ranking of recency values into a small number of categories. For example, if you use five categories, the customers with the most recent purchase dates receive a recency ranking of 5, and those with purchase dates furthest in the past receive a recency ranking of 1.
2. In a similar fashion, customers are then assigned a frequency ranking, with higher values representing a higher frequency of purchases. For example, in a five category ranking scheme, customers who purchase most often receive a frequency ranking of 5, and those who purchase least often receive a frequency ranking of 1.
3. Finally, customers are ranked by monetary value, with the highest monetary values receiving the highest ranking. Continuing the five-category example, customers who have spent the most receive a monetary ranking of 5, and those who have spent the least receive a monetary ranking of 1.

The result is four scores for each customer: Recency, Frequency, Monetary, and RFM score. The RFM score is simply the three individual scores concatenated into a single value using this formula: RFM score = 100 * Recency + 10 * Frequency + Monetary. Therefore, the number of categories is usually less than 10. The "best" customers (those most likely to respond to an offer) are those with the highest combined RFM scores. For example, in a five-category ranking, there is a total of 125 possible combined RFM scores, and the highest combined RFM score is 555.

Sounds simple right? Maybe, but how are these individual scores calculated? What really is the difference between nested and independent binning and does it matter which one I choose?

Independent Binning

Independent binning is a relatively easy concept to grasp. Simple ranks are assigned for each score: recency, frequency, and monetary. The three ranks are assigned independently. The algorithm chunks up the observations for the recency values into k bins, attempting to place (100/k)% of the observations into each bin. Once recency scores are assigned, the algorithm starts the process all over for frequency, and then again for monetary. The values of frequency and monetary do not depend on the value for recency. They’re scored completely on their own. This makes scores very easy to interpret! A monetary score of 6 means the same thing for a customer who has a recency score of 5 and a customer who has a recency score of 8. The two customers spent roughly the same amount of money.

While scores from independent binning can be easy to interpret, there is one big thing to consider when choosing this type of binning. If your data is very skewed for a specific input variable, say your input frequency variable, you may not get all the bins you want. Even if you request 9 bins, a specific value for the frequency variable in your dataset comprises over (100/k)%=(100/9)%=11% of the data, then you will end up with fewer than 9 bins. Binning algorithms will not split observations with the same value into different bins. This makes sense if you think about it from a business stand point. Why would customer A who bought from your store 10 times be placed into bin 5 while customer B who also bought from your store 10 times be placed into bin 4? They should be treated equally, so the algorithm will do just that.

The takeaway from Independent binning is that the interpretation of each of the three RFM components is unambiguous, however, for some datasets, there may be a less even distribution of combined RFM scores with some RFM score segments having no customers.

Nested Binning

Nested binning is computed quite differently than independent. The algorithm starts off the same; date of last transaction is binned into k bins with approximately (100/k)% of the observations in each bin, and those bins are assigned a recency score with the oldest transaction date receiving a score of 1. The difference comes into play when frequency scores are calculated. Whereas the independent binning method completely ignores what happened in recency calculations and does a brand-new calculation for number of transactions, nested binning creates different frequency bins for each recency score.

The nested algorithm takes all the observations that have recency 9. For those observations, it then calculates frequency. Then the algorithm takes all the observations that have recency 8. For those observations, frequency is calculated, and so on until all observations have been reached. Each recency score will calculate frequency differently.

The same kind of process happens when calculating monetary scores, however, it gets a little messier. Monetary scores are calculated within each combination of recency and frequency. Each combination (for example recency 6, frequency 6) will calculate its own set of monetary scores. Because of this process of “nesting” how scores are calculated, the interpretation of these scores is more time consuming and requires a more careful eye. To understand what a monetary score of 4 means, we must interpret it in terms of the frequency and recency scores too. That monetary score cannot stand on its own like it can for the independent binning method.

The picture below depicts this method. Notice how monetary is nested under frequency, which is nested under recency. Interpreting one of these scores requires knowledge about what scores lie above it.

While that was a lot of information describing nested binning, let’s not forget that it exists for a reason. We wouldn’t have different methods to choose from it there wasn’t a benefit to it. Nested binning provides a more even distribution of the combined RFM scores by forcing the bins to be more evenly distributed. You won’t run in to the issue of fewer bins like you may encounter with independent binning. This method also takes into consideration that purchasing behavior, number of items purchased (frequency) and amount of transactions (monetary) can change over time. For this very reason, recency is the first score listed in the RFM analysis.

The takeaway from nested binning is that the method tends to provide a more even distribution of combined RFM scores, and takes into consideration that purchasing behavior (i.e., frequency and monetary) may change over time (recency) due to changes in pricing schemes or seasonal promotions. This method has the disadvantage of taking more words and time to interpret frequency and monetary rank scores.

 

So which one do I choose?

The best answer I can give you is: it depends! Deciding which binning method to choose in your analysis depends on the business problem and justification. If you want a quick and simple interpretation, you may lean more towards independent binning. On the other hand, if you want to factor in purchasing behavoir changing over time to plan pricing and seasonal promotions, maybe nested binning suits you better. It's up to you!