Daring to look beyond RFM

One of the standard ways to segment customers is RFM that stands for Recency, Frequency and Monetary. The main idea of RFM is to sort customers based on the following attributes -

• How recently the customer transacted with your Business i.e., Recency
• How often they make the purchase i.e., Frequency
• How much they spend i.e., Monetary or Average bill value per transaction

One day I and my lead, Sai, were sitting and discussing more on RFM.

Me: Sai, when we have such a simple way (RFM) to segment customers, then why do we need any other way?

Sai: No doubt, the RFM method is straightforward, but it has its drawbacks. It merely ignores two critical dimensions, Latency, and Profitability.

Sai: Latency is the calculation of the average time gap between two subsequent transactions. Buying patterns of frequent customers are quite different from infrequent customers. But the RFM method can not distinguish between the two customers. That’s why we need latency.

Me: Sorry, Sai, I did not understand. Can you please elaborate more about latency?

Sai: Imagine we have two hypothetical customers — Mr. X and Mr. Y. Mr. X is a frequent customer, and he transacted in January, March, June and August, whereas Mr. Y takes a longer time and transacted in January and August.

Recency score = Last month of the calculation period - Last transacted month
Frequency score = Avg transaction in a month

A simple RFM calculation might suggest that Mr. X is more loyal than Mr. Y. But it will fail to take into account that Mr. X’s latency is 2–3 months and he has not transacted in the last four months.

Me: Pardon me for interrupting, but Mr. Y also did not buy anything in the last four months.

Sai: You are right, but he is within his historical range of 7 months. Mr. Y’s latency is 7 months and Mr. X’s latency is 2–3 months. For clarity, let’s calculate the drop-off rate of Mr. X and Mr. Y.

The model of buying behavior we are describing here is a particular case of the event-history model. The simplest way to calculate is tⁿ.

t = last transacted month ÷ last month of the calculated period.
n = total transactions in a month

Therefore for Mr. X, the probability is (8÷12)⁴= 0.198, nearly 20%, and for Mr. Y it’s (8÷12)² = 0.44, almost 45%. Mr. Y is twice more likely to remain active than Mr. X. This method is much more efficient in predicting customer drop-off than RFM evaluation.

Me: Wow! I didn’t realize the importance of latency until this conversation. Latency brings an entirely new dimension to the problem. It gives you a new perception of thinking.

Me: We are already looking at the revenue. So, how does profitability affect the RFM calculation?

Sai: In the current method of RFM, marketing decisions are made based on revenue. By marketing, I mean investing in customer relationships. But the decision to invest in customer relationship needs to be made based on profitability and not on revenue. Let’s take our previous example of Mr.X and Mr. Y. For the sake of simplicity, we will focus only on two dimensions of RFM i.e., Frequency and Monetary.

Frequency score = Avg transaction in a month
Monetary = Avg transaction value (ATV)

A simple RFM evaluation might suggest that Mr. X is better than Mr. Y. But if I bring one more dimension to the problem, then it will make more sense. Let us look at the table again after adding a new column for profitability

Profitability = Total revenue earned - Cost of goods sold.

There can be multiple reasons for negative values in profitability. Some customers purchase only when they get significant discounts. Some buy low-margin products. At times, the cost of servicing customers who buy only small quantities of low-margin products may exceed the revenue they bring in.

In our case, it makes more sense for the company not to give any offer to Mr. X. If he still stays active and transacts, that’s obviously good news, but it’s not worth it for the company to chase him.

Me: Brilliant, Sai! I had no clue that profitability plays such a significant role. But how can a company utilize all these pieces of information about latency and profitability?

Sai: These scores can help companies in planning customer retention strategies, minimizing marketing spend, improving RoI, etc. Profitability can assist in answering whether the company should give offer/s to a customer or not. If yes, then latency can help in deciding the time for sending the offers, notifications, emails, etc. For example: If recency equal to latency and the customer has not transacted yet, then the company can send a reminder to the customer. Offers can be triggered if the difference between recency and latency is more than x. We can make data-driven decisions based on the above five dimensions. Some other metrics like offer redemption percentage, Cash to discount ratio, etc. can also be used.

If you are looking for a product that can do such analyses, check out our website — First WalkIn Technologies.

Assumption: We are analyzing data for a Year from January to December, and all calculations are done at the end of December.

Note: In practice, of course, our calculations are more sophisticated than the foregoing example and can take into account any number of variables, including demographics, monetary value, products purchased, etc.

The content was written, keeping in mind only big retailers. Generally, for startups, revenue and cashflow take precedence over profitability.