Churn-Save Playbook
Trigger Conditions
Agent churn risk often appears in task quality before it appears in login frequency. A customer submitting tasks every day but failing repeatedly may be more at risk than a lower-frequency customer whose tasks succeed.
Monitor the following signals. Thresholds are starting points; calibrate them by customer size, task type, and historical volatility:
| Signal | Meaning | Severity |
|---|---|---|
| Task completion rate drops | The same customer and workflow perform below their own history | High |
| Human-review rate rises | Risk checks, confirmations, or corrections are increasing | High |
| Return rate declines | Users do not come back after first success, or stable users go quiet | Medium-high |
| Active workflow count drops | Users keep only the simplest use case and stop trying adjacent work | Medium |
| Primary user stops using | Champion leaves, changes role, or stops submitting tasks | Medium |
| Cost or billing objections rise | Customer starts questioning whether value matches price | Medium |
High-severity signals should be confirmed and handled quickly. Medium-severity signals can be watched over a short trend window before human intervention.
Roles And Responsibilities
| Role | Primary responsibility |
|---|---|
| CSM | Owner; customer outreach, root-cause diagnosis, internal coordination |
| Engineering | Engaged for product quality, tool failure, model routing, or reliability issues |
| Product | Engaged for workflow mismatch, poor onboarding, or wrong recommendations |
| Sales | Engaged when the customer raises contract, budget, or pricing issues |
Intervention Timeline
Day 0-3: Confirm Signal And Root Cause
CSM verifies whether the alert is real, then reviews recent task logs, error types, human-review reasons, user feedback, and billing changes.
Root causes usually fall into four groups:
- Product quality: completion decline clusters around one workflow, tool, or model version.
- Usage pattern: the user is trying unstable tasks or has wrong input/permission configuration.
- Value explanation: tasks succeed, but the customer cannot see what time, quality, or review cost was saved.
- Organizational change: owner changed, budget tightened, or the team’s workflow moved.
Day 4-7: Targeted Intervention
Choose action by root cause:
- Product quality: engineering investigates; CSM proactively explains impact and remediation.
- Usage pattern: CSM recommends fitted workflows and gives reusable example tasks.
- Value explanation: use the customer’s own task data to review time saved, quality, and review cost.
- Organizational change: sales and CSM re-check budget, procurement path, and internal champion.
Do not recommend new workflows before root cause is clear. When the customer already feels value declining, a new task suggestion can feel like extra work.
Day 8-14: Watch Recovery
Keep tracking whether the trigger signals improve. Recovery is not just login volume; check whether:
- the affected workflow is stable again;
- users submit adjacent tasks again;
- human review and support tickets decline;
- the customer accepts the new value explanation.
Day 30: Classify Outcome
Classify the account:
- Saved: quality and usage signals recovered; customer explicitly intends to continue.
- Stable but low activity: customer stays, but usage scope shrank; lower operating priority.
- Churn confirmed: customer stops paying or does not renew; complete exit interview and archive root cause.
Common Mistakes
- Saving with discounts alone: discounts can address price objections, but they do not fix quality, task mismatch, or organizational change; if unit economics are tight, they worsen margin.
- Treating quality problems as training issues: users may not give the agent another chance.
- Letting support replace CSM judgment: support closes tickets; CSM sees account health and value change.
- Arguing during exit interviews: the goal is to learn the real reason, not win the debate.
Metrics
Track quarterly:
- Early-signal coverage: whether alerts cover true churn without creating too much noise.
- Save success rate: percentage of triggered accounts whose signals recover within 30 days.
- Median save time: days from trigger to recovery.
- Churn reason distribution: product, usage, value, organization, price; feed back into product and operations.
Cross-Section Connections
- Churn signal definitions: metrics/overview
- Product quality escalation path: incident-response
- ROI impact of agent completion decline: economics/controls-and-roi
- Internal support tickets as early churn signal: support-runbook