Personalisation & ML · ML
Flag churning members fourteen days early
14d
Churn is not an event on the day a member leaves — it is a decay curve that starts weeks earlier. This engagement builds the decay features, sets the flag at a fourteen-day horizon, and hands retention a list short enough to act on.
The process
Assemble the behavioural history
Order, session and engagement history pulled into the warehouse, one row per member per week, so recency and frequency become measurable series rather than impressions.
Define churn honestly
The churn label fixed to your actual repurchase cycle, not a borrowed ninety-day default — agreed before any model is trained.
Model the decay
Gradient-boosted features on decaying recency, basket drift and support contact; the output is a ranked risk score with its reasons attached.
Ship the flag into the retention stack
Scores written back daily to your CRM or PostHog cohort, so a flagged member enters a journey instead of sitting in a dashboard.
Hold out and measure
A hold-out group receives nothing; win-back is read as the gap between treated and untreated, not as raw campaign response.
What we need from you
- Twelve months of transaction and session history
- Access to your CRM or marketing automation tool
- An agreed definition of an active member
What you get
- Weekly member-level risk score with reason codes
- Churn definition documented against your repurchase cycle
- Retention journey wired to the flag, with a hold-out
Timeline
Four to six weeks to the first scored list.