Problem
After each call I pasted the transcript into a chatbot. No memory per client, no cross-client view. Pain points lived in my head, and churn signals got missed until renewal.
Users and jobs to be done
- CSM after a call: send an accurate follow-up in minutes, not half an hour.
- CSM before a call: see every open action item for that client.
- CS lead: spot which accounts are drifting toward churn.
Decision log
| Decision | Why |
|---|---|
| Two AI passes per meeting, not one | Client-facing output and internal risk notes need different tone and structure. |
| Action items persist across meetings | A task list that resets every call is useless to a CSM. |
| Pain points stored as structured data | Topic, severity and quote make cross-client rollups possible. |
| Client data kept private | Call transcripts contain client information, so keeping it private comes first. |
Results
- 45 min saved per client call
- 5+ churn risks flagged early
- 50+ pain points shared with product
What I'd do next
Date-range filters and CSV export for the pain-points table, then a shared version so the whole CS team feeds one view to the product team.