RFM Analysis: How to Segment Clients by Loyalty

What Is RFM Analysis

RFM analysis is the analysis of the client base that is based on the purchase history. It allows you to segment customers into groups: the most loyal ones and the most inactive ones. RFM acronym stands for Recency, Frequency, and Monetary:

1. Recency — recency of purchase or the period of time since the last purchase. The clients who have bought recently are more likely to buy again.
   
2. Frequency — frequency of purchases or the number of purchases for a certain time period. The probability of the sale will be higher if a person has made many purchases.
   
3. Monetary — the sum of all the purchases for a certain period. The clients who spent a big sum of money on purchases are more likely to spend it again.
   

RFM analysis helps us build communications in a way that stimulates clients to move from one group into another, retains them, and motivates them to make recurrent purchases.

How to Perform RFM Segmentation

1. Input Data for RFM Analysis

To perform RFM analysis, you'll need the data on all the purchases made by all the clients. They can be downloaded from a CRM system or an analytics platform. There may appear to be some difficulty with the data, so, as a rule, you'll have to pre-process it.

You'll need to download transactions data. Every line is a separate transaction; the number of columns can be different, but there must be:

• a unique client identifier (email, phone number, id);
  
• date of purchase;
  
• the sum of purchase.
  

Here is what the prepared data for RFM segmentation may look like:

2. Collect the Information to Create RFM Segments

You should find the following data: timespan since the last purchase (Recency), the amount of purchases (Frequency) and the sum of all purchases (Monetary).

3. Choose the Range of RFM Segmentation

There are three approaches to this:

• segmentation by the ranges of values;
  
• segmentation by the number of clients;
  
• segmentation with the fixed range.

Segmentation by the Ranges of Values

In this case, segmentation is based on the content of rates. You look at the dispersion of figures, define the number of possible segments, take the peak figure at every rate, and divide it by the number of segments.

As a result, you get segments with equal ranges on every rate:

In our case, R and F values were equally divided into 3 parts.

Recency:
1 - 214-320 days
2 - 107-213 days
3 - 0-106 days

Frequency:
1 - 13-18 purchases
2 - 7-12 purchases
3 - 0-6 purchases

The advantage of this method is that segments are easy to single out. However, if we resort to this division, customers segmentation will be lop-sided (90% of the buyers can get to one segment and only 1% into another).

Segmentation by the Number of Clients

In this mechanism, the division by every rate is done in a way that an equal amount of buyers gets into the segments.

This kind of RFM segmentation allows you to quickly single out the segments without making a huge disbalance between the groups. The disadvantage of this approach is the low quality of segmentation. In this example, the clients who made purchases one day got into different segments by the Recency rate.

Segmentation With the Fixed Range

It's our favorite method. You must define the value borders for every rate on your own while segmenting with the fixed range.

There are some nuances. When defining the range by the recency of the purchases, we need to take into account:

• seasonality;
  
• the time period between purchases;
  
• an average client's lifetime;
  
• promotions;
  
• holidays.

In Recency, we referred to the average timespan between the orders and the timespan between the first and the second order. In Monetary, we referred to the value of an average check. In Frequency, we had to do it by the average amount of orders per person.

We got the following thresholds:

Recency:
1 - last purchase was more than 6 months ago
2 - last purchase was 3-6 months ago
3 - last purchase was 0-3 months ago

Frequency:
1 - 1-2 orders
2 - 3-5 orders
3 - more than 6 orders

When we estimate F, we need to keep in mind that the average time span between the first, the second, and all the other orders will probably be different. The client hasn't got used to the brand yet after the first purchase, and they will spend more time deciding to make the second one if they decide to make it at all. Further, they convert much easier.

In our case, the average timespan between the first and the second order was 53 days, and between the other ones — 36. We got this into account to single out the newcomers' segment correctly. If we apply just the average time period between the orders, the newcomers can get into the group of more loyal clients, and all future work with them will be done in the wrong way.

Also, we should consider the seasonality and the product type for the ranges by the frequency of purchases. E.g., for a beauty store, one order in two months is good, and a client with such frequency of purchases will probably get into the segment of regular customers. And if we are talking about a grocery store, one purchase in two months is something of a random client.

To estimate the ranges by Monetary, we need to take into consideration the following facts: the type of business, the price, the average check, seasonality.

The advantage of the described method is that this segmentation is of really good quality and the clients get into their segments correctly. The disadvantage is the process is laborious and demands much time and effort.

Don't forget to segment the wholesale and retail customers. The R, F, and M thresholds will be different for them.

4. Organize the Segments on Totality of the R, F, M Rates

Now we have the ranges for every rate and we need to evaluate them, i.e. give them a quality coefficient. For example, it can be from 1 to 3, but it might be bigger depending on how detailed your work will be. 3 is just a standard number.

We decided that 1 will be the worst value and 3 will be the best one. So there are 3 ratings by recency for each of the 3 rates. The sum of the 3 rates with the same ratings gives us a segment:

• R1—F1—M(1-3) — lost;
  
• R1—F(2-3)—M(1-3) — loyal clients who lost activity;
  
• R2-F(1-2)-M(1-3) — sleeping;
  
• R2-F(3)-M(1-3) — loyal sleeping clients;
• R3—F1—M(1-3) — newcomers;
• R3—F2—M(1-3) — developing;
• R3—F3—M3 — regular (loyal).

As a result, every client is assigned to a certain segment:

Or, for the sake of convenience, we can display the same information as an interactive dashboard:

As a result, we get the RFM matrix:

How to Use the RFM Matrix

RF Analysis

It is the analysis by the recency and frequency of the purchases. As monetary depends on frequency, we can sometimes just leave it out.

Segmentation by RF shows how frequently clients make purchases. It allows us to see those clients who have recently bought something and interacted with the company.

RM Analysis

RM segmentation shows the division of the clients by Recency and Monetary. RM analysis allows us to see the clients that bring the most revenue and those whose contribution to the total profit is minor.

FM Analysis

FM analysis segments the clients depending on the frequency and sum of the purchases. FM segmentation allows us to see the clients who buy rarely but spend big sums and those who make frequent purchases but with small checks.

This is where we leave off for now. We hope the article gave you the understanding of what RFM segmentation is and how to perform it.