How we coach hundreds of sales reps with AI-powered roleplay

The moments that decide deals are almost never the ones you can prepare a slide for - a VP pushing back on pricing, a question you didn't anticipate, a buyer who won't volunteer how they actually make decisions. Role-play has always been how we train for those moments, but at most organizations I've worked at, it's also the training that reps are least likely to finish. Completion rates on role-play modules typically sit between 40% and 60%.

The content is usually fine. The problem is peer role-play itself. It depends on a partner being available, willing to give critical feedback, and skilled enough to simulate a real buyer in a high pressure situation. In most cases, reps walk each other through the pitch politely and move on.

As we scaled, we needed a way to keep the value of role-play while removing the parts that held it back: scheduling, inconsistency, and the awkwardness of pushing back on a colleague. So we built an AI coach that does exactly that.

How we made AI sales coaches that reps love

We built a series of AI role-play coaches on ElevenAgents: we designed the conversation structure and coaching logic in the visual workflow builder, and uploaded product docs, competitor intel, and persona briefs to each agent's knowledge base - all without engineering support.

Each coach simulates a realistic buyer persona through a live voice, natural-sounding conversation. Three things made the program work: the buyer had to feel real enough to push reps, the coaching had to be useful enough to bring them back, and the data had to feed back into the broader enablement program.

The buyer makes you earn every piece of information

Each persona is designed to reveal information gradually, the way a real buyer would.

1. Shared freely: Job title, company size, general pain points - the basics any buyer would mention upfront.
2. Earned through sharp questions: Specific challenges, what they've tried before, internal politics - the stuff that only surfaces when a rep asks something worth answering.
3. Surfaced after careful discovery: Budget authority, decision timeline, past vendor failures - the information that shapes whether a deal is real.

Generic questions get generic answers. "What are your challenges?" gets a surface-level response. "You mentioned compliance training across 35 offices. How are you handling regional regulatory differences today?" unlocks the real story. This trains reps to do the work that matters most in enterprise sales: earning information through curiosity, not waiting for the buyer to volunteer it.

Reps can ask for help without penalty

At any point during the role-play, a rep can say "I'm stuck" or "Can you teach me?" The AI buyer immediately pauses the scenario, switches to coaching mode, and teaches the specific skill the rep needs: what to say, why it works, and sample language they can adapt. Then it offers to jump back into the role-play so the rep can try again. 

This was the decision that changed everything. When we added teaching mode, voluntary repeat sessions went up dramatically and reps started treating the AI coach like a practice partner, not a test.

It scores you, then coaches you

Each role-play follows a beat structure that mirrors a real customer conversation. After the final beat, the AI agent drops character and becomes a coach. It scores three criteria on a 1-3 scale: Was the pitch technically accurate? Did the rep reference the buyer's specific situation, or could that pitch have been for anyone? Was the conversation at the right altitude for the buyer's seniority level?

Most enablement programs settle for completion. Now, we can measure competency too. Every transcript gets structured data extracted automatically. Scores, pass/fail, per-beat breakdowns. This feeds leaderboards and gives us aggregate visibility into which skills the team is strong on and where they need help. For the first time, we could see skill gaps at the organizational level without reviewing individual recordings. 

What the numbers showed

We rolled it out to hundreds of sellers across North America, EMEA, and APAC. Completion rates hit above 90%, and reps rated it 5 out of 5 in feedback surveys, saying that the persona felt realistic and the coaching was specific enough to act on. We’ve seen reps coming back to their AI role play coaching agents time and time again, and we’ve seen qualitative impacts on rep preparedness.

This is the first enablement exercise I've ever voluntarily done more than once.

- AE, EMEA @ ElevenLabs

Where this is going

The version we're building next is deal-specific. Before a discovery call with the VP of Customer Experience at a mid-market insurer, the coach assembles a persona from the buyer's LinkedIn profile, company history, industry vertical, and CRM data. The rep practices against a simulated version of their actual upcoming buyer. After the call, they do a targeted coaching session focused on where they lost momentum or missed an opportunity.

The endgame is a continuous practice layer that wraps around every deal in the pipeline, compounding improvement across a rep's entire book of business.

Try it yourself

We put together a demo version of one of our coaches so you can experience this directly. You'll play an AE pitching a workforce training platform to a VP of Learning & Development.

• AI Sales Coach Demo
• AI Roleplay Guide for Reps

Everything in that demo was built on ElevenAgents - the personas, coaching mode, and scoring are all configurable for your own team's scenarios. Learn more about ElevenAgents here, or talk to your account team for help building your own AI sales coaches.