Opens in a new tab

Enterprise enablement · Simulation

Sales Practice

A rehearsal room for the conversations that actually close.

Role
Forward-deployed builder
Type
Enterprise enablement
Status
CAAT pilot
Live at
sales.usaw.ai
Sales Practice library showing AI buyer personas ready for rehearsal

Part of a system

Act Every buyer here starts as a Protagons file — a character USAW extracted from real writing — carrying the standard’s behavioural layer plus USAW’s decision layer. That’s what makes them behave under pressure instead of reciting a persona description.

The friction

Roleplay for complex, high-trust pension sales is scarce and awkward. It needs a manager or peer to play the buyer, difficulty swings with whoever’s available, feedback is subjective, and reps who’ve seen the “answer key” never practise real discovery. In deals with union, finance, and HR stakeholders, one weak first call can burn a rare opening.

Where AI helps

AI gives reps a realistic buyer on demand — 24/7, no scheduling partner. What holds that buyer together for twenty minutes isn’t the prompt; it’s the layers underneath: a Protagons file carries the voice, its behavioural layer carries how they react when pushed, and USAW’s decision layer carries what they actually want from this purchase and who they still have to convince. Reps see only LinkedIn-level facts, so discovery has to be genuine. The same practice runs in the browser or over a real phone line, and every attempt is scored against the team’s own bespoke sales framework from the actual transcript — objective evidence instead of a manager’s gut.

Who it helps — CAAT Pension Plan’s pension-solutions sales team — reps selling DBplus to employers, HR and Total Rewards leaders, CFOs, founders, and union leadership.

What I built

  • Buyers built on the Protagons stack

    Each buyer is a Protagons character file carrying the voice, the standard’s Behavioural Agency Layer for how they hold up under pressure, and USAW’s decision layer for how they navigate this particular purchase. Voiced through OpenAI Realtime and Grok across ten voices: 12 curated buyers plus unlimited generated ones.

  • Phone practice over a real line

    Twilio-routed calls scheduled by PIN, recorded with consent, transcribed, and auto-emailed back.

  • Competency scorecards

    An LLM-as-judge grades the transcript against a bespoke sales framework: weighted competencies, coaching notes, and a score—configurable to whatever methodology a team sells on.

Proof

12 curated buyersBehavioural + decision layersBrowser + phoneTranscript-scoredOn the CAAT design system