Churn prediction
A daily-updated risk score per subscriber, with the contributing signals shown plainly: billing failures, support touches, engagement drop-off (have they opened the portal in the last 30 days), product-fit signals (have they swapped items multiple times in a row). The score is the input to your retention playbook — trigger a save-offer when risk crosses 0.6, send a check-in email when it crosses 0.4.
Save-offer drafting
When a subscriber clicks cancel and gives a reason, an LLM drafts the save-offer copy based on their order history, their reason, and your offer playbook. A subscriber canceling for “too expensive” with 14 successful orders gets a different draft than a subscriber canceling for “not using it” in their second cycle. You approve the playbook; we generate the per-customer copy.
Dunning copy optimization
Subject lines and body copy on dunning emails are A/B tested at the cohort level. The winner becomes the new default, automatically. Recovery rate, not open rate, is the objective. Across our cohort, this added ~4 percentage points to recovered revenue beyond the decline-aware engine alone.
What we don’t do
- No chatbots. Subscribers want answers, not a chat window.
- No autopilot wizards. The merchant approves the playbook. Always.
- No black box. Every model decision shows its contributing signals.
- No training on your data. Models are trained on aggregated, anonymized industry data, not on your individual subscriber records.
Who this is for
Churn prediction ships on Growth. Save-offer drafting and dunning optimization ship on Scale.