How a roleplay practice is set up

In this tutorial, we'll set up a roleplay practice.

Trainees talk to an AI customer, get a score, and then talk it through with a coach.

Our example is a bank that trains new advisors, but the same setup works for any team that learns by talking.

A roleplay practice is built from four pieces in one project, and we'll go through each of them in this video.

The practice customers and the coach are agents, so they live here in the Agents list.

The handovers between them are created in Interactive Content, and the evaluation is set up on the Evaluations page. We'll get to both later on.

For our bank, we created one coach, Avery Brooks, and two practice customers: Elena Marsh and Marcus Hale.

Elena plays a first-time homebuyer, and Marcus a frustrated cardholder. Both use the same fields, so let's open Elena's agent form.

Her Agent's Description is what the trainee reads before the practice starts. So we write it as a briefing: who she is, her situation, and the trainee's goal.

Her Objective and Guardrails make her act like a real customer. She asks for clear monthly numbers, and she never coaches or scores the trainee.

At the end of her Objective, we add the handover back to the coach as a reference, the tag you see here. That way, she returns the trainee to Avery once the conversation is over.

We just used a handover in Elena's Objective. So what is a handover? Let's go through it. Handovers are created in Interactive Content.

Interactive Content has several types, and Handover is one of them: it passes the conversation to another agent. We have three: one to each practice customer, and one back to the coach.

Let's open the one for Elena. The Handover type is selected, and the Handover target is the agent it hands over to.

With Show as detail on, the trainee first sees Elena's card, with the briefing we just wrote.

Now that the handovers are ready, let's set up the coach. In Project Settings, the project uses the Empower package, which unlocks the coaching workflow.

And Avery is the Featured Agent, so trainees meet her first when they open the app.

Next, let's open Avery's agent form from the Agents list. Like Elena, she's an agent, but she's the one trainees talk to first, and come back to.

Her Business Process Steps tell her what to do. The first process asks what the trainee wants to practice, and references the matching handover with the same kind of tag.

The second process is the debrief. When the trainee comes back, she gives the overall result, one strength, and one thing to try next.

Avery's debrief needs results, and those come from the evaluation. Evaluations have their own page in the sidebar, and ours is called Advisor Conversation Review.

It uses Post-Conversation Analysis, so it runs once the conversation ends.

It scores four areas from zero to five, like Needs discovery in the preview here, and for each one, it writes what worked and what to try next time.

Then we choose who gets the results. Featured agent passes them to Avery for the debrief. User Facing shows them to the trainee. And Admin Facing sends them to the Analytics Suite.

Finally, back in Elena's agent form, we attach it in her Evaluations section. It goes on every practice customer, not on the coach, because their conversations are the ones we score.

With all four pieces in place, let's see what the trainee sees in the Promethist app.

Avery asks what we'd like to practice, and gives us one thing to focus on.

Then the handover opens Elena's briefing card: the description we wrote in Studio.

We practice the conversation.

A few minutes later, we say goodbye, and Elena hands us back to the coach.

The evaluation runs, and the trainee sees the overall result: three and a half out of five, passed.

Below that, each area has its score, what worked, and what to try next time.

Back with Avery, the debrief starts with the result: we passed, with three and a half out of five.

Then one strength: we built trust by asking what feeling safe meant, and by using a clear monthly example.

And one thing to try next: ask a few more budget questions before suggesting a loan amount.

Her suggestion for the next practice: try Elena again, with that single focus.

Once trainees have practiced, the manager can review the results in the Analytics Suite. Featured Evaluations shows how many trainees were evaluated, their average best score, and the average for the first, second, and later attempts.

Under Conversations, we can open any practice, read what was said, and on the Evaluations tab, see how it was scored.

That's the whole setup: a practice customer, a coach, handovers between them, and an evaluation.

To build your own, swap in your own customers, briefings, and scoring areas.