MK

martinkrizan.com / aarrr / churn-prediction-winback

AARRRRetention

Churn prediction + win-back campaigns

Flagging at-risk accounts from usage drop-off and running targeted rescue campaigns against them.

Heuristics or a model that estimate churn probability from usage data, wired to a rescue campaign that fires automatically when an account crosses the risk threshold.

Why this ring

Needs enough historical volume to work; most teams start with simple usage-drop heuristics long before a model earns its keep.

SaaS fit

Only makes sense past a certain number of active accounts — too small a base gives the model no signal.

How to apply it

When to apply: You have enough churn history (dozens of cases or more) and usage events to detect an activity drop.

First steps: Start with a simple heuristic (e.g. a key action drops X% over two weeks), wire a win-back campaign to it, and only reach for a predictive model once you have enough data.

Pitfalls: Building a full ML model before validating that the simple heuristic plus rescue campaign even works is a wasted investment.

Metrics to watch: Heuristic/model precision (how many flagged accounts actually churned), win-back campaign save rate.

Resources

Who does it well

  • ChurnZeroCS platform built around usage-based churn risk scoring and automated playbooks.
  • BaremetricsWidely used by SaaS teams specifically to track churn and run recovery/win-back flows on failed payments.