Skip to content
Rush Commerce
AI & Automation3 min read

Synthesia's AI roleplay training is enterprise-only. For now.

Synthesia launched Roleplay Sessions — AI avatars that run practice conversations and score them. Enterprise-only until inference gets cheaper. Build it yourself instead.

The thing that actually trains a new hire isn't the video. It's the twentieth time they handle an angry customer and someone tells them what they got wrong. Synthesia — the AI video company last valued at $4 billion — shipped Roleplay Sessions on July 22, moving from generating training videos into running the practice loop itself. It's enterprise-only, which is exactly the problem, because the businesses with the worst training gap are the ones that can't buy it.

What actually happened

Per TechCrunch, Roleplay Sessions puts employees in a live conversation with an AI avatar that plays the other side — a prospect, a complaining customer, an employee in a performance review. The avatar responds, objects, and scores the session against a rubric the company defines, with analytics across attempts.

Companies build the scenarios from training material they already have, or pay Synthesia consultants to build them. CEO Victor Riparbelli's pitch is unglamorous and correct: "we learn the best by actually practicing something rather than just reading it."

Two honest caveats. No pricing was disclosed, and it's enterprise-only today — Synthesia says small businesses and educational customers come "in the next few months" as inference costs fall. Early customers are described only by category: a top-three European company by market cap, a Fortune 100 firm, a large recruitment agency. No named logos, no outcome data.

Why AI roleplay training matters for your business

A ten-person shop can't run a training department. What it has instead is a good employee who explains the job to every new hire, badly, between other work — and turnover that quietly resets the whole thing twice a year.

The interesting part of this launch isn't Synthesia's product. It's that the primitives are now cheap and public: a full-duplex voice model that doesn't wait for you to finish talking, a rubric written in plain English, and transcript scoring. We covered GPT-Live's full-duplex voice models and Claude Cowork's record-a-skill flow — those are the same building blocks, available now.

Scope it small and it's real work you can finish. Pick one conversation that costs you money when it goes badly — the inbound price objection, the refund request, the estimate walkthrough. Write the rubric your best person already applies in their head. Build one scenario, run your team through it, score the transcripts, and check whether the score moves. If it doesn't, you learned that in a week rather than after an annual contract.

The enterprise vendors will get to your segment eventually. You have the problem now.

Key takeaways

  • Synthesia launched Roleplay Sessions July 22: AI avatars run practice conversations and score them against a company-defined rubric
  • Enterprise-only today, with SMB availability promised in "the next few months" as inference costs drop; no pricing disclosed
  • The practice loop — not the training video — is the part that actually changes behavior
  • The primitives are already public: a full-duplex voice model, a written rubric, and transcript scoring. Start with one high-cost conversation

Training that lives in one person's head? We turn the conversations your business actually loses money on into scored practice runs your team can repeat. See what we build or tell us which conversation hurts most.

Sources: TechCrunch.

  • #ai-training
  • #synthesia
  • #voice-ai
  • #onboarding
  • #automation
TR

Tommy Rush — Founder, Rush Commerce

Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More

Get The Rush Report weekly — one email, zero fluff.