Determining scientifically which offer is the most effective and economical for retaining a specific customer is much harder than it initially seems.

For example, if you have 1 000 customers churning each month and offer five different cancellation reasons (Too expensive, Not using, etc..), you'd end up with roughly 200 customers per reason.

If you further segment these customers—for instance, distinguishing between someone canceling after two days versus two years—you're dividing those 200 into even smaller groups. Now, each customer group (cohort) might only have 50 customers per month.

When you run an experiment to compare which offer works best for each cohort, you're now down to just 25 customers per offer, meaning it takes months to gather enough data to achieve statistical significance. So even with 1 000 churning customers a month, you likely don't have enough data to reliably segment and determine the most effective offer.

Refloat solves this by aggregating cancellation data across all companies using Refloat. We segment the companies customers into cohorts, such as the lowest 20% of spenders or the top 20% for each company. Cancellation flows are standardized, so a customer in the lowest 20% cohort at Company A, on a monthly plan selecting Too expensive will experience the same cancellation flow as a customer with identical characteristics in Company B.

This approach ensures we have a large number of customers in each cohort, allowing us to quickly measure which cancellation flow work best to retain them, and to reach statistical significance more often.

Futhermore, Refloat's data aggregation enables even more fine-tuned offers due to the high volume of customers and data. Beyond targeting a specific cohort, such as top 20% spenders choosing Not using enough, we can craft cancellation flows for multiple segments—like customers whom have churned before, are in the top 20% spenders, bottom 25% in activity, and selecting Not using enough as their cancellation reason. The more specific the segmentation, the better the results.

Contrast this with the earlier example, where even with 1 000 churning customers a month, you wouldn't have enough data to determine the most effective offer.