ML for Retention
What is reactivation scoring?
Reactivation scoring estimates the probability that an already-inactive customer will come back if you contact them - a different question from whether an active customer is about to leave.
Why it is a separate model
Churn models are trained on active customers and answer 'will this person leave'. Once someone has already gone silent, that question is answered. The useful question becomes 'is contacting this person worth the cost and the goodwill'. Different population, different features, different label - so it is a second model, not a threshold on the first.
What predicts a return
How the relationship ended matters more than how long ago. A customer who tapered off gradually behaves very differently from one who stopped abruptly after a support incident or a failed payment. Historical response to previous reactivation attempts is usually the single strongest feature - and it is the one most teams never store.
The cost side of the equation
Every reactivation contact costs money and burns a small amount of goodwill. Score-driven outreach means the list is ranked by expected return, not by recency. A team that can make 300 calls a week gets a materially different result from the top-300 by model score than from the top-300 by days-since-last-session.
The consent and compliance gate
Reactivation lists are exactly where compliance mistakes happen. Marketing consent state, self-exclusion flags, do-not-contact requests, and regional communication rules have to be checked at send time, not at list-build time. In regulated verticals this gate sits between the model output and the outreach system as a hard block, not as a filter someone can override.
We build automation systems for small teams. Free audit call to map your specific workflows - no pitch, just a plan.