Find the at-risk renewals before they're at risk.

What this actually looks like.

How you'd actually build it.

Each step is its own agent-or-pipeline component. Don't try to do this in one giant prompt — the failure modes compound.

1. Score every account weekly

   Composite score across: usage trend (DAU, feature depth), support load and sentiment, exec sponsorship continuity, contract horizon, payment timeliness, NPS. Weighted by your historical churn signals.
2. Classify with reason codes

   Red / yellow / green tiers. Each red account has a primary reason code. CSM knows what to address before they pick up the phone.
3. Auto-draft the outreach

   For each red account: draft an exec-level email referencing the specific signal, schedule a QBR if none on the books, alert the CSM in Slack.
4. Track save attribution

   Every action is tracked to the renewal outcome. Six months in, you know which interventions actually move save rate.

What it runs on.

No vendor lock-in. Every piece is replaceable. The orchestration outlasts the model — when Claude 5 or GPT-6 ships, swap the model, keep the pipeline.

What this actually returns.

Calibrated to mid-market scale (50–1000 staff, $1M–$200M revenue). Your numbers will vary; the framework gives you the order of magnitude to plan against.

Build it yourself, or have us ship it.