How Customer Behavior Helps You Define The RFM Segments

How Customer Behavior Helps You Define The RFM Segments

In the previous article we briefly introduced the RFM analysis and we detailed the 3 dimensions we can use to segment our customer base: Recency, Frequency and Monetary. The key idea is that not all customers are created equal so it’s vital to know where to draw the lines and who to target in your marketing strategies and campaigns.

Let’s continue the “Know your customer” part of this series by understanding when a customer is no longer a new customer, how many times they should buy before we should propose a subscription model to them and what is the threshold beyond which we should consider promoting them to VIP status?

The short answer is that there is no universal answer. As we are very passionate about data analytics, we think the answer should come from a data driven customer segmentation technique which relies on purchase history, and product or services usage data and not on human intuition or industry standards. Thus, the groups of customers need to be derived by analyzing their past behavior.


When deciding which of your customers are new customers you need to firstly have a look at the distribution of your customers by last purchase date or product usage date. When looking at this you will most probably notice that this data is very skewed, with lots of new customers having made a recent acquisition, which is a sign of a healthy company growth or vice versa, few customers who made a recent purchase and lots of old customers which haven’t made any recent purchase.

In the first case you will want to keep a close eye on those new customers and turn them into loyal customers. As such you might next wonder how much time usually passed before your loyal customers made their second purchase and proactively reach out to your new customers during that period. This is where you will draw the line: for example, any new customer with an acquisition no older than 14 days should be proposed a cross-sell in no more than 7 days.

In the case of old, inactive customers, you need to urgently put in place an activation campaign with maybe a stimulus/one time discount for customers which will come back to you. We will deep dive on the proposed journeys for each of the identified segments.

Moreover, as we have experienced with our customers, only looking at acquisitions might give false positives because customers who have already signed up for a yearly subscription can incorrectly get classified as inactive. As such, you will need to analyse both the purchase and the product usage data to correctly  understand who is no longer active.


You can reasonably expect that most of your customers are one time buyers and unfortunately this comes with a fixed cost that might not pay off. Remember that in online businesses, the most important equation is that the net profit comes by subtracting a customer’s lifetime value from its acquisition cost. When a customer does not become loyal, you wasted time, energy and resources to get him to make a first purchase and then lost him.

In the case of customers which make relatively frequent purchases (let’s say they make a purchase every month) you might start wondering if they make repeatable, predictable purchases so you can propose a subscription type crafted for their needs or proactively reach out when the time comes for their next purchase.


When trying to define what a valuable customer looks like for your business, you need to look at the distribution of the average order value. The reason why this approach will work better than any rule of thumb or industry standard is that your existing customer base is the result of the market segment you chose to advertise to and serve.

By looking at the number of customers by Average Order Value (AOV) you could either see a skewed distribution with a lot of customers having a low AOV or you might notice a normal distribution, with few customers placing small orders, few customers placing very large orders and the bulk of your customers placing a medium value.

Stop Relying on Error-Prone Human Segmentation and Implement Dynamic Data Driven Segmentation

The key takeaway is that you should design and implement a data driven segmentation process by looking at data distributions and then setting some hard rules (ex: customers with average order value over 200$ should be considered valuable customers) or dynamic rules (say, top 10% customers by number of purchases should be considered frequent buyers). Subsequently, these rules can be translated in programmed, recurrent queries and scripts that will automatically adjust to changing customer behavior and push each customer on the optimum marketing campaign.

For example, new studies show that people started changing their phone less often and this means that if in the past a frequent buyer would change their phone every 2 years, this definition needs to change to accommodate the new customer behavior of waiting longer than 3 years before considering upgrading to a new model.

By using a dynamic segmentation process driven by customer purchase data, you will make sure that your segmentation methodology will accommodate any change in the structure of the market you serve.

How to Perform RFM Segmentation

  1. Calculate the Recency, Frequency and Monetary value for each customer.
  2. Divide the customers list into tiered groups for each of these 3 dimensions. It’s recommended to split the customers into 3-4 groups for each dimension, resulting in 27-64 combinations. To keep it simple, we will only use 2 groups for each dimension, resulting in 8 segments.
RecencyActive / Inactive
FrequencyFrequent Buyer / Occasional or One Time Buyer
MonetaryBig Spender / Prudent Spender
  1. Intersect all the above mentioned tiers to isolate the customer segments. To align everybody in your organization, It is very helpful to assign names to the segments of interest. For example:
RecencyFrequencyMonetaryCustomer SegmentDetails
ActiveFrequent BuyerBig SpenderHigh-Spending Active Loyal Customersthis group consists of those customers that are in R1, F1, M1. It means that they made a recent acquisition, they shop oftenly and they spend a high amount of money.
ActiveFrequent BuyerPrudent SpenderLow-Spending Active Loyal CustomersThey transacted recently, they shop oftenly, but spend low amounts
ActiveOccasional BuyerBig SpenderHigh-Spending New CustomersThese are customers that transacted once or twice but did it quite recently and they spent a lot.
ActiveOccasional BuyerPrudent SpenderRegular New CustomersThey had one or 2 recent, low amounts transactions
InactiveFrequent BuyerBig SpenderChurned Best CustomersThey used to make frequent, high value transactions but it’s been a while since their last purchase.
InactiveFrequent BuyerPrudent SpenderChurned Regular CustomersSimilar to the previous segment but they used to make low value acquisitions
InactiveOccasional BuyerBig SpenderOne-time Good CustomersThis is a very elusive segment of customers, they made a single large purchase in the past and you have never seen them since
InactiveOccasional BuyerPrudent SpenderOne-time Regular CustomersThis, unfortunately, might represent one of the largest segments and is made up of one time, low spending customers that tried the service/product once and never came back.

Not All Customers Are Created Equal

While the number of groups you chose might differ, rest assured that most of the resulting segments will be well represented in your customer base and by this time you will clearly see how your customers are diverse, complex groups of individuals that have different needs and need to be approached with personalized marketing messages.

Enter Personalized Customer Treatment

As you probably anticipate, these customer segments will not react the same to your marketing campaigns, won’t be equally interested to try out new products and services and most importantly they won’t be equally profitable for the business! This means that the next step is to learn how to customize any customer interaction to improve the overall customer experience.