Coaching built on the call itself.

A sales manager can’t listen to every call. Coaching ends up based on whichever call they happened to hear, and the advice is hard to tie back to what the rep actually said.

I built the whole thing solo: preparing 150 call transcripts across 15 reps, the AI analysis pipeline, the manager workflow from rep to saved plan, and a 60-test suite.

1. 01Every suggestion links to the exact transcript passage, and the app checks that each quote exists in the source before showing it.
2. 02Calls with missing or conflicting speaker labels stay readable but never feed coaching, so a rep is never coached on something the prospect said. That left 61 eligible sales calls.
3. 03The Coach next list ranks repeated concerns, not reps. It points the manager at a pattern worth a conversation instead of building a leaderboard.
4. 04Analysis is versioned. When the instructions change, affected results are marked out of date instead of mixing old and new advice.
5. 05Prepared results ship with the app, so anyone can run the full workflow with no API key and no paid requests.

A manager opens a rep, sees what they do well and what to work on across discovery, connecting a benefit to the prospect’s need, objection handling and next steps, checks the evidence in the transcript, and saves a practice to the rep’s plan. The team view surfaced a clear pattern: helping prospects see the need came up as an objection in 23 eligible calls.

This was a demo built for Owner; it was not adopted or used by a team. Built on a 150-call sample with no CRM link, it demonstrates the prototype workflow, not a revenue result. The validation plan below is how I would measure that.

Two experienced sales managers review 60 calls blind, then compare against the tool: at least 70% of suggested practices accepted and under 5% unsupported before a pilot. Then a four-week pilot against a two-week baseline, tracking manager time saved, rep behavior on later calls, and qualified demos held.