Using Behavioral Segmentation to Detect Churn Before It’s Too Late

May 21, 2019 Brady Walker

Detecting fading and lost customers is one incredibly valuable form of behavioral segmentation. In the retail world, this is no easy task, but there are a lot of tools at your disposal to get started.

If you’ve been following our series on segmentation best practices, then you already know 

For this week's segmentation investigation, we'll describe how different segmentation methodologies can work (or not work) for your churn prevention program. 

Let's first think about what "churn" is, because in truth, it looks different from customer to customer. 

Someone might drop off your radar for six months and pop back up, right? Some customer who normally buys on a weekly basis might go on vacation for three months. These kinds of pauses in buying cycles happen all the time. But how do you know when and how to react?

Especially when you’re thinking of that top 10% of customers, retailers dearly want to know when those customers are cooling and fading away because losing them cuts a deep hole in the business. Just to throw some scary numbers out there, consider this:

  • It’s 5x more expensive to acquire a new customer than it is to retain an existing one.
  • The top 10% of customers commonly represent 50% of a retailer’s revenue.

As we’ve mentioned in earlier posts, the beautiful thing about your best customers is not just how much they’ve spent with your brand, but how much wonderful behavioral data they’ve provided — their relatively high purchase frequency gives you this deep history of information; it tells you when they buy, their typical basket size, the things they like, the channels that draw them in, and more.

You want to hold onto these people, but how do retailers do it?


Days Since Last Purchase

The most basic starting point is a rules-based approach like “days since last purchase,” which — depending on your brand — may be 60 or 90 days, possibly up to 6 months.

This is very easy to run. You take the last transaction date of every customer plus today’s date — is it greater than x-number of days? You set it up as an email trigger, and everyone who crosses that dangerous threshold gets a 25%-off discount code in their inbox.

But there are big problems with this kind of simplistic rules-based approach.

The first one is that different customers behave differently, and setting up a single rules-based trigger to treat them all the same ignores those differences to a detrimental effect. What we mean is: some customers buy every week, some every month, some every three months, some once per year.

If you had a weekly buyer, would you really want to wait 90 days before you think about how to keep them from fading away? No, they’re long gone.

And that once-per-year group? You’re just throwing discounts and messaging at them when they’re not ready to buy, so either they simply don’t convert or they get annoyed and unsubscribe and they’ll forget their annual purchase with you nine months from now.

And then there’s the group that buys approximately every three months without needing a discount, but your trigger just so happens to give them 25% off every time they would’ve bought at full price anyway.


One step further in sophistication than “days since last purchase” is RFM, which stands for Recency, Frequency, Monetary Amount. Pretty self-explanatory that these are the variables included in the model:

  • How recent was a customer's last order?
  • How frequently do they order?
  • On average, what's the average order amount?

Those three variables together put you in a better place in terms of predicting who is likely to return. So this goes beyond the 90-day rule and gets closer to the individual level.

You might create buckets for each of these attributes, and each segment comprises a unique combination of buckets. For example, a segment might be “customers who last purchased within the last 90 days, who buy every two weeks, and who spend under $100 each time.

While this method takes into account the shortcomings of a “Days Since Last Purchase” approach, it doesn't factor in where a customer is in their lifecycle. Lifecycle transitions can have a large effect on the likelihood of response.

In addition, RFM doesn't try to predict behavior; it segments your customer base but doesn't make any assessment of how valuable those segments are or will be in the future.


Advanced Probabilistic Models

Then there are some types of advanced probabilistic models include pattern recognition and machine learning algorithms. There are many approaches, but the general idea is that these approaches look at individual customer stories and whether their types of ordering behavior are indicative of customers who drop off the map.

With these types of models, you will be able to reach the ideal where you're alerted about a customer who orders weekly not ordering for two weeks and a customer who orders monthly not ordering in the last 45 days. The algorithms learn the patterns in the data and alert you to anomalies.

While these types of models may seem ideal, they are complicated and often require a lot of data science help and computational horsepower to get up and running.

The trade-off between complexity and predictive power you choose to make really depends on the resources at your disposal.

In the world of retail, it’s not common that a customer lets you know they’re never shopping with your brand again, especially not the really valuable customers. But with some predictive modeling in place that can trace a pattern and build segments based on various criteria, you can figure out the customers who are likely to churn and even which subsegment of that group show signs of being potentially very high-value customers if they stick around. As a retail marketer, this is probably some of the most valuable information you could hope to have.

Once you've figured out a customer is lost, you're going to want to try to win them back. Given the amount of information you probably have on them, it's almost always worthwhile reaching out to them and bringing them back into the fold.

If you want to learn more about segmentation, check out our previous articles on the topic:

And if you want to learn more about the most powerful form of segmentation — segmenting by Customer Lifetime Value (CLV) — then click the link to read our latest book, The Chance of a Lifetime.


Previous Article
How the Bouqs Co Uses Event Propensity Modeling to Drive Customer Loyalty
How the Bouqs Co Uses Event Propensity Modeling to Drive Customer Loyalty

Next Article
The Challenges Of Adopting a CLV-First Strategy Across The Organization
The Challenges Of Adopting a CLV-First Strategy Across The Organization

When you optimize around product or channel metrics without regard to the customers who are behind the metr...