/ Predictive Analytics

Increase winback campaign performance with predictive churn

Acquiring new customers can be as much as seven times more expensive than retaining the ones you already have.

What are win-back campaigns?

Winback campaigns can reactivate customers who are losing engagement with your brand. Many win-back campaigns offer an incentive for an at-risk customer to shop again with the brand. Many of them include product recommendations below the main content to increase engagement. Take the one below for example:

Rule-based vs. predictive churn models

Most brands use a rule like months from last purchase to identify at-risk customers. For instance, many brands consider a customer to be at-risk if he/she has not bought for 12 months.

The problem with a rule like that is it only accounts for the recency of purchases. It does not take into consideration other important factors like the time between purchases, the frequency of purchases and recency of engagement with the brand.

By contrast, Metisa's predictive churn model takes into consideration factors including recency, the time between purchases, the frequency of purchases and recency of engagement with the brand. As such, it is more accurate at identifying customers who are truly disengaged and at-risk.

Why does your churn model matter?

It matters because win-back campaigns usually come with a discount or incentive to reactivate customers.

If you offer discounts to customers who are active and loyal, you risk diluting the value of your brand. Correctly identifying your at-risk customers can save a lot of lost sales from unnecessary discounts and damaged brand equity.

Just how much more accurate is Metisa's predictive churn?

Many customers ask us how much more accurate Metisa's churn model is against a typical rule-based definition of churn. We ran a backtest to compare both models:

  • Model 1: Metisa churn rate > 0.5
  • Model 2: Customer has not bought in 12 months

For this test, we used a training dataset of 14,000 customers and test dataset of 20,000 customers. We trained both churn models with 24 months of historical purchases and tested it against the next 18 months of purchases. In other words, for any customer that was identified as at-risk, we checked whether he/she actually made a purchase during our testing period.

The results: Metisa's churn model was 80% more accurate

Metisa's churn model identified churn customers 57% of the time, compared to 31% for the rule that a customer has not bought in 12 months. This means that Metisa's churn model was 80% more accurate than a rule-based churn for this test. In fact, we could see statistically significant improvements in accuracy across the brands we work with.

In other words, Metisa's predictive churn rates are much better at identifying customers who are truly at-risk and can save a lot of lost sales from unnecessary discounts and damaged brand equity.

If you are interested to find out more about churn modeling, you may also want to check our related article on how to A/B test different churn models.

Justin Yek

Justin Yek

Partner & cofounder at Altitude Labs, creator of Metisa, former investment banker, public speaker, hobbyist musician

Read More