AI tools for sellers and managers

The problem

Reps spent time on work a model can draft well: follow-ups, account research and CRM fields from call notes. Managers had no consistent way to coach calls or judge deal health. The tools also had to run on the company’s own AI-workflow product, so each one doubled as a customer proof point.

What I built

• AE follow-up writer. Any AE or SDR pulls CRM data and call notes and gets a drafted follow-up. I picked the model from a three-model bake-off.
• Account brief. Researches an account, verifies its claims before writing, and adds CRM relationship context. It is on version 25. A light version runs inside the rep queue app.
• MEDDPICC to CRM. Writes MEDDPICC fields from call notes. A retry queue holds calls that arrive before the deal exists.
• Call coaching for AEs and SDRs. Scores calls against rubrics on a schedule.
• Deal scores. Weekly scores per open deal, fed into the manager workspace.
• AI council. Four models draft, cross-examine each other and adjudicate. I used it to fact-check a published comparison article.
• Message writers. 15 writers produce outbound sequences per play from a 196-row knowledge base.

Design decisions

• Every tool shows its sources or verifies its claims. Sellers stop using tools they can’t trust.
• I ran small, capped first runs with a cost estimate before scaling anything.
• Writer fixes were verified at zero diffs before each publish.
• Team knowledge went into shared Claude skills: ICP across 9 segments, 17 white papers and 29 case studies.

Result

Coaching scored its first 30 calls with 477 checks, none refused, for about $1. The deal-score backfill covered 476 of 480 deal-weeks for about $13 to $16. The AI council’s run time fell from 20 to 29 minutes down to 7.6 minutes.