A large Israeli content provider was losing thousands of paying subscribers every month, and no one could see it coming. We moved retention from firefighting after the cancellation to a quiet, personal nudge weeks before it.
A large content and broadcast business was losing thousands of paying subscribers a month — and no one in management could say in advance who was about to leave, or why.
So retention happened after the flood: marketing burned huge budgets on blanket campaigns and sweeping offers, but only once a customer had already phoned in to cancel. And the vast content library worked against them — with no way to recommend the right thing at the right moment, subscribers drifted into non-use and the feeling that they weren't getting their money's worth. The data to see it coming existed; nothing was reading it.
Catch the drift weeks ahead, then act on it — personally, automatically, before the phone rings.
We joined the app and online usage data — who watched what, where they stopped, how often they logged in each week — with the older billing and service-centre systems, into one live view of each subscriber.
A model that reads behaviour patterns and spots a subscriber losing interest — a gradual drop in watch time over a fortnight, say — roughly 30 days before they even think about cancelling.
Instead of a giant, unnavigable library, the system surfaces content matched to each person's taste at the moment it matters — turning drift back into use, which is the real cure for churn.
Rather than nagging with generic texts, the system sends an at-risk subscriber a tailored content suggestion, or triggers a precise price offer — the intervention most likely to keep that particular person.
Every save is built from that subscriber's own behaviour, not a segment average — so the nudge feels like the service understanding them, not a mass mailshot.
The whole loop runs before the customer decides to leave, which is the only point at which retention is cheap — and quiet, instead of a call-centre scramble.
Retention stopped being a monthly emergency and became a background process.
The monthly churn rate fell by around 25% in the first quarter alone, and average watch time and usage per subscriber rose 40% on the back of smarter, more precise content recommendations.
The service centre's retention spend dropped sharply as it moved from putting out fires — expensive last-minute saves — to calm, proactive retention that happens before anyone reaches for the cancel button.
It lasts because it's early and personal: the model watches the leading signals rather than the lagging cancellation, and every save is fitted to the individual, so the intervention lands instead of annoying. Prediction wired to an automatic, tailored action — that's the difference between knowing your churn number and actually moving it. It's the same production discipline behind our operational-assistant work.
Reef TRH