The result
Before this existed, the PM team prioritized based on memory and whoever was loudest in the last call. Now they prioritize against a queryable source of truth across thousands of conversations and the full active enterprise client base.
The roadmap stopped being a function of recall bias and started being a function of evidence.
How it works
A customer-intelligence system that ingests recorded calls and meeting transcripts, runs AI analysis per-client, and surfaces churn risk, sentiment, feature requests, and complaints into a dashboard the PM team uses every week.
Operational scale:
- 74,000+ recorded calls
- 2.5 million transcript sentences
- Per-client analysis runs nightly across the active enterprise client base
I designed the analysis pipeline, picked the model strategy, built the prototype, and shipped it into production with the engineering team.
Why it matters
The PM organization could now answer questions that used to take a week of stakeholder interviews in a single dashboard query. “What are our enterprise customers saying about onboarding this quarter?” used to be a project. It became a filter.
What I learned building it
The hard problem wasn’t the AI. It was getting noisy, incomplete signal - call transcripts have hold music, off-topic tangents, half-finished thoughts - to produce stable enough analysis that PMs trusted it.
Trust was the deliverable. The AI was the means.