AI role-play is the best-known voice-powered L&D use case. It’s used mostly in sales and customer service training, and a little in leadership and coaching.
It’s likely just the beginning. Voice AI is flexible enough that almost any spoken learning interaction can be designed, delivered and administered with it, and the models are improving quickly.
Two common first steps
The first is to multiply the human-led role-play you already run. If you’ve written scenes and rubrics, the AI can be taught to play them the way your facilitators or actors do, so learners can practise between sessions as often as they like, and you can train many more learners. It doesn’t replace the facilitator; it gives learners the repetitions a workshop can’t.
The second is to add AI role-play to an existing workshop. A workshop on managing conflict, for example, can give learners difficult-conversation role-plays built on the same situations and the same rubric the workshop teaches.
What a role-play can practise
Most scenarios put learners in a difficult situation. A scenario can also make them use your industry’s vocabulary and concepts, or practise specific moves: asking coaching questions, staying curious, talking about value instead of price.
Who the learner faces
Some systems let learners describe the character they want to face, or set the difficulty with a slider.
A design choice that matters more is whether the character is fixed or varies. When the relationship history is the point, a client of eighteen years for instance, the character should be specified. When the situation is the point, leave the character open and the AI varies who shows up, so a learner practising the same scenario for the seventh time doesn’t meet the person they met the first time.
Beyond role-play
Rep practice lets people drill a short response, say thirty seconds, until it’s right. The AI gives feedback after each attempt, so the learning happens in quick iterations.
Reflection modules walk learners through content and ideas, then ask them to think about themselves and their own experience. A module on delegation might introduce the common reasons delegation fails, then ask the learner to describe, out loud, the last time they saw it happen at work, what happened, and why. This is modelled on Henry Mintzberg’s approach to management learning: “reflection on natural experience in the light of conceptual ideas.”
Learning paths
These can be combined in many ways. A sequence that works well is reflection, then rep practice, then the role-play. To help a customer service rep handle an upset customer, for example:
- Reflection. The learner first recalls a time they faced an angry customer and what made it difficult.
- Rep practice. With that in mind, they respond to three different complaints, with feedback after each.
- Scenario practice. Then the full role-play, where they have to use the moves they’ve practised to resolve the situation. Most will run it several times, and after every run they get a written debrief against the scenario’s rubric, with comments on specific things they said.
What L&D sees
Managers and L&D see results, not conversations. For each person, on each module, they see whether the person has started, is practising, or has met the bar (one run where they did everything the scenario asks), and which objectives they keep missing. Reflection stays with the learner and is never reported. That privacy is what makes people willing to practise badly before they practise well.
The scoring is about what learners say, not how they sound. The system I built doesn’t infer, label or score anyone’s emotions from their voice, and doing so at work or in education is prohibited in the EU (AI Act, Article 5).
Role-play is where most organizations start. The learning comes from what you put around it.