Retain Revenue by Predicting Attrition Risk

Doug Young, the chief operating officer at Easalytics, shares how to retain revenue by predicting attrition risk.

For many undesirable things, early detection and intervention can dramatically reduce the severity of the outcome. Think cancer, fires, termites and yes, member attrition. Even if you’re focusing heavily on retention cornerstones like onboarding, engagement and member experience, chances are you still shake your head at the number of lost members per month. And if you’re also thinking about revenue retention and looking at member lifetime value of those lost members, it’s likely even more sobering.

One common method is to use “Days Since Last Check-In” as a red flag metric, such as members who haven’t checked in for 30 days. This can certainly be a red flag, but the problem is that it lets a lot of members slip through the cracks. For example, a look at one brand’s lost members for a month showed that 550 lost members had checked in within the last 29 days, while 468 hadn’t checked in within the last 30 to 59 days. Not very predictive, and the difference is about the same if you compare zero to 59 days to 60-119 days.

A more sophisticated method is to track statistically significant changes in average check-in frequency — greater than two standard deviations —which can detect more subtle declines in very active members, and is also less sensitive to occasional benign lapses in check-ins. Even that method will miss about 15-20% of lost members who were checking in at their usual frequencies when they canceled.

Turns out check-ins are just one of about 60 variables that could in combination be predictive of a member’s attrition risk — including age, gender, membership type, payment term, payment type, join month, join method, current duration, spend patterns, and club seasonal patterns to name just a few. It’s impossible for a human to see patterns among that many variables, let alone make accurate predictions from them, but it’s just the kind of thing machine learning was built for.

If a club has at least 500 members with minimum six months of member data, a machine learning attrition risk predictive model can accurately predict the probability of each member canceling within the next three months. And once set up, it can provide that assessment on a monthly basis without further effort on your part. Efforts to save members after they’re already showing extreme check-in lapses, or have already requested to cancel, are often too little too late. Imagine being able to identify and target at-risk members months before the obvious red flags start showing.

Because you can’t realistically save everyone, your best ROI on effort/resources will come from focusing more on revenue retention than on member retention. Combine the attrition risk probability scores with metrics like member lifetime value or average monthly spend, and you’ll be able to laser focus proactive retention efforts on the highest value members for the greatest impact on revenue retention.