By the time your customer health score turns red, the customer has already mentally churned. Here are the 8 behavioural and billing signals AI can read 30–60 days before cancellation — and how to act on them.
The 8 signals
1. Session frequency drop. Sessions per week dropping below the 30-day median is the strongest early signal. Weight: highest.
2. Feature adoption flatline. If the customer’s feature adoption has plateaued at 40% of available features after 90 days, they’re unlikely to expand or renew. Weight: high.
3. Support ticket sentiment. Tone analysis on support tickets surfaces frustration before it becomes a cancellation request. Weight: medium.
4. Billing dispute or downgrade. Any billing event other than scheduled renewal is a leading indicator. Weight: very high.
5. Power user departure. When the most-active user in the account stops logging in, the whole account is at risk. Weight: high.
6. Login from new domain. The customer’s IT or procurement team is evaluating alternatives. Weight: medium-high.
7. NPS or CSAT drop. Any score below 7 within 90 days of renewal. Weight: medium.
8. Stalled onboarding. Accounts that don’t hit 60% activation within 30 days have 3× the churn risk. Weight: very high for new accounts.
How to wire them up
Pull from product analytics (Mixpanel, Amplitude, Heap), CRM activity (HubSpot, Salesforce), and billing (Stripe, Recurly, Chargebee). Build a unified score updated in real time. Surface in your CRM as a custom property.
For Northgate Capital, we built this in 8 weeks. Logo churn dropped 31% in the next quarter — not because the CS team changed what they did, but because they intervened earlier on the right accounts.
The intervention playbook
Score above 80 → green, no action.
Score 60–80 → yellow, automated re-engagement sequence.
Score below 60 → red, route to CSM with talking points and recommended retention offer.
Score below 40 → executive escalation, recovery call within 48 hours.
Where to start
If you’re at £1M+ ARR with 30+ paying accounts, the math justifies the build. Below that, focus on RevOps foundations first.