Role Play Examples: 10 Workplace Simulations for Training

Scale training programs and cut development time using AI

Start Free

A new estimator sits at a customer's kitchen table. The customer says another company quoted $800 less and waits. In the next ten seconds he will either knock money off, list everything his quote includes, or ask a question. Asking is the only one of the three that tells him what he is up against, and nobody learns to ask from a slide.

That is what a role play is for. I'm the COO at Coursebox, I manage most of our Perth team, and I hire with role-play interviews. Coursebox makes branching scenario software that builds AI role plays, so I'll say plainly where that fits and where it does not: it gives people repetitions between live sessions, and it does not replace the observer or the debrief. In this guide I give ten workplace role play examples you can lift, show how to run one in 45 minutes with an observer checklist and a debrief, and set out what the research says about the debrief.

What is a role play?

A role play is a practice conversation in which one person plays a part, such as a customer, a colleague or a patient, and the learner handles it as they would at work, with an observer and a debrief afterwards.

Medicine gives one well-documented early version. Simulated patients, people trained to play a patient so doctors could practise, were first described by Howard Barrows and Stephen Abrahamson in 1964. Workplace role play uses the same move: a safe stand-in for the person on the other side, so the learner can get it wrong without losing a customer or hurting a patient.

Three formats get confused, and the difference decides what you can practise:

A branching scenario is a role play on a screen, and our guides to scenario-based learning and branching scenario examples in eLearning cover how to design one.

10 workplace role play examples

Each example is a brief you can run as written. They all use the same six parts: the situation everyone reads, the roles (including what the other person is not saying), the opening line, what good looks like, three behaviours for the observer to watch, and one debrief question. Change the names and the product to suit your team, and keep the hidden concern, because that is what makes the learner ask instead of perform.

1. Sales: a price objection at an in-home quote

The example is hypothetical; the company, the people and the prices are illustrative. A family-owned removals company in Geelong, Victoria, sends estimators to people's homes to quote for moves. Marco moved up from crew leader to estimator three months ago and knows the work better than anyone. When a customer says a cheaper quote came in, he either defends the price with a list of features or knocks money off straight away. Dana, the operations manager, wants him to find out what the other quote includes before he talks price.

• The situation: Marco has walked through a four-bedroom house in Highton with the owner, Joan, who is downsizing to a two-bedroom unit in Torquay. He has quoted $3,400 for a Friday move: a crew of three for about 8 hours, packing the kitchen, one wardrobe taken apart and rebuilt, and the upright piano with a fourth crew member and a stair trolley. Joan has the quote on the kitchen table.
• The roles: Marco plays himself. He may drop the kitchen packing ($420) if Joan packs it herself, and may not discount the hourly rate without Dana's approval. The second estimator plays Joan, 68. The piano was her late husband's, the unit is up a narrow flight of stairs, and the other company never asked about the piano. Joan mentions any of this only if Marco asks what is most important to her or asks about heavy or special items.
• Opening line: "Honestly, the other company said two thousand six hundred. Why would I pay eight hundred more?"
• What good looks like: Marco finds out what the other quote includes and what Joan is worried about before he defends the price, offers the kitchen packing instead of a discount, and asks for a next step.
• Watch for: he asks what the other quote includes before answering on price; he asks what is most important to Joan about the move; he offers the kitchen packing choice instead of a discount.
• Debrief question: "What did Joan say that told you what she was worried about?"

2. Sales: asking before recommending

• The situation: A tradesperson walks up to the counter at a building supplies store and asks for a waterproofing product for a bathroom job. The learner is the counter salesperson.
• The roles: The learner knows the product range. The role-player is redoing a shower that leaked after another product failed, and the job is for a landlord who wants it finished by the weekend. The role-player shares those facts only when asked about the job.
• Opening line: "Give me the strongest waterproofing you've got. I'm in a hurry."
• What good looks like: The salesperson asks what the job is and what went wrong before naming a product, then recommends one with a reason tied to what the customer said.
• Watch for: at least two questions before any product is named; a recommendation linked to something the customer said; a check that the customer has everything needed to finish.
• Debrief question: "What did you know about the job by the time you named a product?"

Asking before supplying is the same skill a community pharmacy needs at its counter, and one Sydney study tested and coached it with a simulated customer at the real counter.

Case study: Sydney pharmacy staff coached by a simulated customer

Pharmacists and counter assistants at 36 community pharmacies in Sydney sell non-prescription medicines and are expected to ask the right questions before they supply one. A research team checked how often a counter request reached an appropriate outcome, and coached staff on the spot.

• What they did: Pharmacy students trained as "mystery shoppers" learned 10 scripted scenarios through role play with the research team. Each pharmacy then had one visit a week, nine in a row, posing as an ordinary customer; staff knew visits were coming but not who the shopper was, which scenario they would see or exactly when. Within 5 minutes of each visit the shopper came back, asked the staff member to score themselves on a blank scoresheet, and then gave verbal feedback.
• Evidence and result: In Collins and colleagues' 2017 study (521 of 540 visits analysed; some pharmacies took part in two semesters), the share of visits reaching an appropriate outcome rose from 43% in visits 1 to 3, to 56% in visits 4 to 6 and 64% in visits 7 to 9. Interactions that involved a pharmacist reached an appropriate outcome in 62% of cases, against 33% without one.
• Limitation: There was no control group, staff knew visits were coming (the authors raise the possibility that this alone changed behaviour), the pharmacies volunteered, and the scoresheet was not validated. No patient outcomes were measured.
• Lesson: Practise where the work happens, debrief within minutes, and let the person score themselves before you give feedback.

3. Leadership: hard feedback to a strong performer

• The situation: The team's best account manager has cut off junior colleagues twice in meetings this week. The learner is their manager, and a one-to-one is booked for this afternoon.
• The roles: The learner plays the manager. The role-player plays the account manager, who believes results speak for themselves and is privately stretched across too many accounts. The overload comes out only if the manager asks how things are going before giving a verdict.
• Opening line: "I hit my numbers every month. Is this really what we're spending time on?"
• What good looks like: The manager describes one specific moment and its effect, hears the pushback without dropping the point, and ends with one agreed change.
• Watch for: a specific moment named (what was said and when); a question about their view before any fix is suggested; one agreed change with a date to check in.
• Debrief question: "When did the conversation turn, and what had you said right before it?"

I write most of our Perth team's reviews myself, and in a team our size there is no HR layer between me and this conversation. That is the reason to rehearse it before it happens for real.

4. Negotiation: asking another team to move a deadline

• The situation: A trade show has moved forward a week, so the marketing project lead needs product screenshots from the product team seven days earlier than agreed.
• The roles: The learner is the project lead. The role-player leads the product team, which has two releases due and one person on leave. They share that only when asked what is driving their date.
• Opening line: "We agreed the 20th. I can't move that."
• What good looks like: The project lead asks what is behind the date, offers something in return, such as asking for fewer screenshots or writing the captions themselves, and confirms the new plan in writing.
• Watch for: a question about their constraints before pressing; at least one trade offered; who does what by when, confirmed at the end.
• Debrief question: "What did you learn about their workload, and at what point in the conversation?"

5. Hiring: a role-play interview where the interviewer plays the customer

I use this format myself when I hire: a colleague and I play a customer, and the candidate handles the conversation as they would on the job. It is the quickest way I know to simulate what the person would actually be doing, and it shows me behaviour rather than a description of it. The example here borrows Joan's objection from the removals company, but the shape works for any customer-facing role.

• The situation: A candidate for an estimator job gets Marco's brief five minutes before the interview: the $3,400 quote and what it includes, the $420 kitchen packing option, and the rule against discounting the rate.
• The roles: One interviewer plays Joan, with her hidden concern. A second interviewer observes.
• Opening line: "Honestly, the other company said two thousand six hundred. Why would I pay eight hundred more?"
• What good looks like: The candidate asks before defending the price, stays inside the authority limit in the brief, and stays calm when pushed.
• Watch for: questions before any answer on price; no discount beyond the brief; a steady tone when Joan pushes back.
• Debrief question: "If you ran that again, what would you do differently?"

The answer can prompt a discussion about feedback and another attempt. It is not, on its own, evidence of how a candidate accepts feedback or will perform at work.

A role-play interview is a work sample test, and there is good evidence on those. Sackett and colleagues' 2023 review of selection methods revised the older estimates downwards:

That review supports using work samples as one source of selection evidence, rather than assuming this particular role play has the reported average validity. Use a job-relevant brief, consistent opportunities and behavioural criteria for every candidate, with any required adjustments. A structured interview can provide additional evidence.

6. Health and safety: stepping in when a colleague skips a step

• The situation: On a packing line, an experienced operator reaches into a jammed machine without isolating it first (switching off and locking the power so it cannot restart). The learner sees it happen.
• The roles: The learner is a colleague on the same shift. The role-player has done this many times without harm and has been told by the supervisor that the line is behind. The pressure comes out only if the learner asks why.
• Opening line: "It's fine, I've done this a hundred times. We're already twenty minutes behind."

Use a safe representation for this exercise. Nobody reaches into equipment or bypasses an isolation control. The local procedure specifies who can stop, isolate and restart equipment; the learner practises communicating and escalating within that authority.

• What good looks like: The learner stops the action first and talks second, names the exact step that was skipped, and offers to help do it safely.
• Watch for: the task stopped before any discussion; the skipped step named, not a general "be careful"; practical help offered, such as isolating the machine together.
• Debrief question: "What made it hard to speak up, and what got you started?"

7. Emergency response: a timed team drill

• The situation: A customer collapses near the counter in a busy store. Three staff are on the floor.

The drill rehearses the store's current emergency plan with an appropriately qualified facilitator and a manikin where physical first-aid actions are practised. Use simulated calls; do not call emergency services for the exercise. The communication checklist below is not a first-aid assessment.

• The roles: The three learners take whatever jobs they decide on. One role-player lies on the floor as the customer, and a second plays a bystander who keeps asking questions. The bystander knows the customer said she felt dizzy a minute before she fell, and says so only if asked. The observer runs a stopwatch.
• Opening line: The bystander: "Is she breathing? Should I call someone?"
• What good looks like: One person takes charge and gives each of the others a named job, the call for an ambulance goes early, and the space around the customer is cleared.
• Watch for: someone taking charge and saying who does what; the time the call is made, from the stopwatch; a clear handover to the paramedics on what happened and when.
• Debrief question: "Who was doing what in the first minute, and how did everyone know?"

Drills like this are simulations more than role plays, and one hospital has measured them over years. Since 2000, Southmead Hospital in Bristol has run a one-day maternity emergencies course that includes a 30-minute hands-on drill for a rare birth emergency, shoulder dystocia. A 12-year follow-up by Crofts and colleagues found the recommended manoeuvre used in almost every shoulder dystocia birth by the later years, and nerve injury to the baby's arm (brachial plexus injury) in those births fell from 7.4% to 1.3%. Both reports are before-and-after studies at one hospital, so other changes over the period may have contributed.

8. Compliance and ethics: a request to backdate a deal

• The situation: It is the last day of the quarter. A senior sales manager asks a finance officer to record a deal as closed yesterday. The customer has agreed by phone, but the signed contract has not arrived.
• The roles: The learner is the finance officer. The role-player is friendly, senior and under pressure, because their quarterly bonus depends on the deal. The bonus comes out only if the finance officer asks why the date is so important.
• Opening line: "Pop it in with yesterday's date and we'll tidy the paperwork next week. They've said yes on the phone."
• What good looks like: The finance officer declines to enter an unsupported date, checks the organisation's recording requirements and refers uncertainty to the authorised finance lead. They offer to record the facts accurately and complete the next required step. The script does not decide revenue recognition from the arrival of a signature alone.
• Watch for: a clear no within the first two replies; the reason given as what the record needs, not a personal opinion; a next step offered.
• Debrief question: "At what point did you feel pressure to agree, and what did you say then?"

9. Healthcare: telling a family that surgery is delayed

• The situation: An elderly patient's hip operation, planned for this morning, has been moved to tomorrow because an emergency took the operating theatre. The ward nurse has to tell his daughter.
• The roles: The learner is the nurse. The role-player plays the daughter, who took the day off work and drove two hours. What worries her most is that her father has not eaten since midnight, and she says so only if asked.
• Opening line: "Nobody has told us anything. We've been sitting here since seven this morning."
• What good looks like: The nurse acknowledges how she feels before giving facts, explains what happened and what happens next in plain words, and asks what is worrying her most.
• Watch for: the frustration acknowledged before any explanation; what, why and when stated plainly; a question about her biggest worry.
• Debrief question: "What did she need from you first, the facts or the acknowledgement?"

Hard conversations in health care are where role play has the strongest evidence, including one trial that measured real consultations a year later.

Case study: UK oncologists who rehearsed hard conversations with actors

Senior oncologists from 34 UK cancer centres had little formal training in communication. A research team tested whether a three-day course could change how they talked with patients in real clinics.

• What they did: Each doctor's own clinic video was reviewed at the start. Each doctor then picked the communication problems most important to them and worked on at least one through role play with trained actors, called patient simulators, who gave feedback from inside the role. Video review and group discussion followed.
• Evidence and result: The 2002 randomised trial of 160 doctors coded videos of real consultations and reported more focused and open questions (27%), expressions of empathy (69%) and appropriate responses to patients' cues (38%) among course attenders than among controls; these are relative changes in coded counts. At the 12-month follow-up of 74 doctors, interruptions had fallen from 32 to 6, an 81% difference. Verbal expressions of empathy had dropped back since the three-month check, from a mean of 3.22 to 1.49; after re-watching the tapes, the authors judged that empathy was being shown nonverbally instead.
• Limitation: The control doctors were offered the course after three months, so the 12-month result compares each doctor with their own baseline, not with a control group. Three days with actors and facilitators is expensive.
• Lesson: Build the role play from each person's own real conversations, use role-players who can give feedback from inside the role, and check again later, because some behaviours fade.

10. Onboarding: a new starter's first angry customer

• The situation: It is a new starter's first week on the support line at an internet provider. A customer calls because their connection has dropped for the third time this week.
• The roles: The learner is the new starter, who has been told when to hand a call to a senior colleague. The role-player works from home and has a client video call at 2 pm. They mention the call only if asked what they need today.
• Opening line: "This is the third time this week. I want to speak to someone who actually knows what they're doing."
• What good looks like: The new starter acknowledges the frustration without arguing about their own experience, asks what the customer needs today, and uses the handover rule if the fix is beyond them.
• Watch for: the frustration acknowledged before any troubleshooting; a question about what the customer needs today; the handover rule used, with the customer told what happens next.
• Debrief question: "What did the customer need most, and when did you find out?"

Running this in week one means the first angry customer a new starter meets is a colleague with a brief. For a full set of service conversations at different levels of difficulty, use our 25 customer service role play scenarios, and see how they fit a wider programme in our guide to customer service training.

Simulations beyond role play

Some skills need more of the job than the conversation: the equipment in your hands, the system on the screen, the clock running and the team around you. Four kinds of simulation are common at work:

• Drills: a rare emergency rehearsed with real equipment and real roles, such as a customer collapsing in a store or an emergency on a maternity ward.
• Simulated customers: a trained person visits the real counter or calls the real line, and staff do not know which customer it is.
• Equipment and system simulations: a training copy of the booking system, or a machine run in training mode, so people practise the steps without real consequences.
• Screen-based scenarios: a branching scenario that a learner works through alone.

More technology does not automatically mean more learning. A 2020 meta-analysis of simulation-based learning in higher education by Chernikova and colleagues put it bluntly: "Typical computer-and-screen-based simulations do not outperform well-organized no-tech role plays and simulated patients or simulated students." Pick the format for the skill, and spend the effort on the brief and the debrief.

That debrief is not optional in simulation practice either. The healthcare simulation standards published by INACSL say that all simulation-based activities "must include a planned debriefing process", and they call for a prebriefing so learners know the ground rules before they start.

How to run a role play in 45 minutes

Back at the removals company, Dana runs a 45-minute role play every fortnight at the Geelong depot, first thing on a Monday, with one brief per session. This session uses Joan's price objection: Marco has quoted $3,400, and Joan says another company quoted $2,600. Every name and figure is the same illustrative one.

The four roles

• The learner: Marco, who plays himself.
• The role-player: the second estimator, who plays Joan from her brief.
• The observer: Tevita, the office coordinator, who ticks behaviours and writes down what Marco says.
• The facilitator: Dana, who keeps time and runs the debrief.

Brief each role separately

Hand out three briefs, and keep the two private ones apart:

1. The situation, for everyone: Joan, the house in Highton, the unit in Torquay, the $3,400 quote for Friday on the kitchen table.
2. Joan's brief, for the role-player only: the opening line; the piano, the stairs and the other company's silence about the piano, revealed only if Marco asks; when to ease off (once Marco has asked what the other quote includes and found the piano and the stairs, start asking about Friday); when to push (if he discounts straight away, say "If you can drop it that fast, what's the real price?"); and the stop rule (8 minutes, or when Joan agrees to a next step or says no).
3. Marco's brief: what the $3,400 includes, the $420 kitchen packing he may drop, the rule that he may not discount the rate without Dana's approval, and his goal: an accurate comparison and a next step, not a win at any price.

The session, minute by minute

The debrief takes longer than the round, on purpose. A structured medical-school programme described by Nestel and Tierney used a similar split: "5 minutes preparation, 5 minutes in role-play and 10 minutes in feedback".

My free role play kit has Joan's three briefs, the observer checklist and this run sheet filled in, with blank versions for your own scenario.

The observer checklist

The brief lists three core behaviours. In a full session Tevita watches for two more, five in all, and writes down Marco's words for each one he sees:

Tick only what was said or done, and write the words down. A quote such as "What does their price include?" gives the debrief something concrete to discuss; a tick on its own does not. Our guide to observation checklists covers how to write behaviours someone else can observe.

The debrief, learner first

Dana runs the debrief in the same order every time:

1. Marco first: what went well and what he would change.
2. Tevita: the ticks, each with the words Marco used.
3. The second estimator, answering as Joan for one question: the moment she would have booked or walked.
4. Marco again: the one thing he will do differently in round 2.

The learner speaks first because he usually knows where it went wrong, and saying it himself is easier to hear than being told. When Dana does need to raise something, she pairs what she saw with what she was thinking and then asks, an approach from simulation training called advocacy and inquiry: "I noticed you went straight to the discount. I was worried Joan would push for more. What was going through your mind?" If you want the theory behind reflecting on an experience, our guide to Kolb's experiential learning theory covers the cycle.

You can run the round from written observations and the learner's own words. Cheng and colleagues' healthcare-simulation review did not establish a clear additional benefit from video-assisted debriefing in the included comparisons; that does not prove video and written notes equivalent. Choose recording when it serves the task and agreed handling requirements.

When someone does not want to role play

Some people dread role play, and that is normal. In Nestel and Tierney's study of a structured role-play session for medical students, the authors wrote that role play "almost always meets with resistance and/or anxiety from some students", and 63 of 284 students (22.2%) had found role play unhelpful before. The structured session still went down well:

That is self-report from students after one structured session. It describes perceived helpfulness, not measured learning, and the reported totals do not show that each person who previously disliked role play changed their view. The practical options below are facilitation suggestions.

• Start as the observer. Someone nervous can watch and tick a round first, then take the learner's chair next time.
• Anyone can call a pause. Say so in the brief, before round 1.
• Keep the brief in hand. A learner who freezes can stop, read it again and restart.
• Swap roles. Playing the customer lets the learner hear their own lines from the other side.

What the research says about role play and debriefing

The evidence for practising conversations is strongest in health care, because that is where most of the trials were run. An overview of 12 systematic reviews of communication training for qualified doctors, by Berkhof and colleagues in 2011, found that effective programmes lasted at least a day, were learner-centred and focused on practising skills, and that the best strategies included role play, feedback and small group discussion. The authors advise using lectures and written information only to support that practice.

The debrief has some of the largest measured effects in this research. Keiser and Arthur pooled studies of after-action reviews, a structured form of debrief, across teams and individuals:

An effect size of 0.92 is large: it is measured in standard deviations, not a percentage. The detail that helps a trainer is where the effect is largest. In Keiser and Arthur's analysis, debriefs did most in tasks with "high complexity and ambiguity in terms of offering no intrinsic feedback". A sales call or a complaint is that kind of task: the customer rarely tells you what you did wrong, so without a debrief nobody finds out. The same paper found that add-ons such as "a canned performance review" did not meaningfully improve the effect. An earlier meta-analysis by Tannenbaum and Cerasoli, across 46 samples, put the average effect of debriefs at d = .67, which the authors describe as improving effectiveness over a control group by approximately 25%.

You do not need actors to start. A Cochrane review of communication training for medical students, covering 76 studies and 10,124 students, found that personalised feedback improved communication skills (moderate-quality evidence), and could not tell whether trained simulated patients beat peers playing the role; the evidence on that comparison was of very low quality.

After the session: the manager's follow-up

From the COO's chair, the role play is the cheap part. It costs 45 minutes of four people's time. What decides whether it was worth it is the conversation a week or two later, when I can see whether the new behaviour turned up on a real call.

The research on behaviour modelling backs this up. A meta-analysis of 117 behaviour modelling training studies by Taylor, Russ-Eft and Chan found that skills carried over to the job best when, among other things, trainees saw poor examples as well as good ones, set goals, and had managers who were also trained. Effects were larger on what people learned than on how they behaved at work. The gap between the two is where the manager comes in.

One public service built the follow-up into the programme from the start.

Case study: US probation officers who role played each conversation skill

US federal probation officers meet clients one to one, and those sessions rarely used structured behaviour-change skills. A training programme called STARR set out to change that and measured whether clients were re-arrested or had their supervision revoked.

• What they did: A three-and-a-half-day course demonstrated each skill, then officers practised it by role playing the interaction with another officer and received feedback. Afterwards, trainers gave feedback on audio recordings of the officers' real sessions, and ran four booster sessions by phone with individual coaching, about 40 hours of training in all.
• Evidence and result: In Robinson and colleagues' 2011 evaluation (67 officers analysed, 41 trained and 26 not; 513 clients after training), 26% of trained officers' clients failed within 12 months against 34% of the comparison officers' clients. For moderate-risk clients the figures were 16% and 32%. The comparison for high-risk clients was not statistically significant; that does not establish no effect.
• Limitation: Clients were not randomly assigned to officers, two districts declined randomisation, and the programme's developers evaluated it themselves. A 24-month follow-up reported that the overall effect partly faded.
• Lesson: Short peer role plays work best when the coaching continues on real conversations afterwards, and averages can hide groups for whom nothing changed.

At the removals company, the follow-up is a short question. A week after the session, Dana asks Marco about one real quote where a customer mentioned a cheaper price, and uses the same five behaviours from Tevita's checklist: what did the other quote include, and when did he find out? If Marco did not ask, that becomes the brief for the next fortnight.

Live, peer or AI role play?

Each format buys you something different:

The evidence on AI role play is early. A 2026 systematic review of generative AI virtual patients in health care education found 15 studies with 645 participants, often single sessions under 10 minutes, and almost no long-term follow-up. Its authors concluded that the findings "support feasibility and acceptability more strongly than demonstrable learning gains or transfer to clinical practice". In one study of 40 adults recruited online, people who practised with an AI conflict tool called Rehearsal reduced their use of escalating competitive strategies by an average of 67% compared with a lecture-only group, but that was measured in a role play straight after one session, not at work. So far, AI role play is good for repetitions, and I found no study I could read that shows it changes how people behave on the job.

In the support threads inspected since May 2026, customers asked whether a choice could change what happens next and how to edit an AI-generated scenario. Those search-selected examples are useful design questions, not a measure of the most common requests. Both questions came up in the Joan test.

Building Joan's objection as an AI role play

On 26 October 2026 I gave the Coursebox course agent a brief for a 15-minute course called "When the customer has a cheaper quote", with Joan's objection as an AI role play at its centre. Coursebox is our product, so weigh my account with that in mind. The brief carried the $3,400 quote and what it covers, Joan's opening line, her hidden concern about the piano and the stairs, the $420 kitchen packing option and the rule against discounting the rate. It also told the scenario not to mark choices correct or incorrect.

The facts came through. The scenario setup read: "Your quote is $3,400 for a crew of three for about 8 hours, packing the kitchen, taking apart and reassembling one wardrobe, and moving her upright piano with a fourth crew member and a stair trolley." Joan opened word for word: "Honestly, the other company said two thousand six hundred. Why would I pay eight hundred more?" The setup and the first turn said nothing about the narrow stairs, her late husband or the other company ignoring the piano. When I picked the choice to ask what the other quote included, Joan said: "They never asked about the piano. I don't know whether it was included, and they didn't ask about the stairs." After every choice she replied in character, with no correct or incorrect labels.

The first turn offered four reasonable moves: asking what was most important about the move, asking about heavy or special items, asking what the other quote included, or explaining the discount authority limit. These are paraphrases of the original options. The tempting early discount only appeared at turn 2. That first turn did not test the intended choice between asking and conceding under pressure.

In the AI editor on that page, I asked it to change only turn 1: keep the two questions about the other quote and what Joan cares about most, and replace the other two with moves people make under pressure, "Offer to take 10% off the hourly rate if Joan books today" and "Tell Joan the other company will probably cut corners". The editor replied: "Updated only the first turn ... No correct/incorrect labels are shown; the rest of the scenario is unchanged." I reloaded the learner view and started again. Turn 1 showed the four new choices, and when I picked the discount, Joan said: "If you can do 10%, you can do 20." I checked turn 1 again; I did not replay the later turns after the change.

That was one scenario and one and a half playthroughs: the good path to turn 3 and the discount path to turn 2. It shows the scenario kept the facts and the hidden concern I checked, and that one turn could be fixed, not that anyone quotes better for it. Between Dana's sessions, Marco could run it twice, once discounting early and once asking first. What he cannot practise there is his own wording, so the live round and the debrief stay. If you are planning where AI practice fits in a wider programme, our page on AI for employee training shows the other formats.

The agent also reported that the "After the practice" page repeated the introduction, behaviour list, scenario and wrap-up after reaching its revision limit. That page was not independently checked or repaired in this run. End-of-scenario feedback was not captured because the browser froze, so I cannot assess it from this test.

Adapting these role plays for a classroom

The same structure works with secondary students and with teachers. Keep three roles (two players and an observer with three behaviours to tick), a timed round, and a debrief where the student who handled the conversation speaks first. Change the stakes to suit the class:

• Business studies or a language class: one student plays a shop owner and another a customer comparing two quotes for a school event, such as a bus hire or a sound system. The customer's hidden concern might be a fixed budget from the school.
• Teacher professional development: teachers rehearse a difficult conversation with a parent, with a colleague playing the parent from a brief.

There is evidence for drama-based teaching in schools. A meta-analysis of drama-based pedagogy by Lee, Enciso and Brown found a moderate average effect on literacy achievement (d = .64 across 37 comparisons), and effects were larger when the class teacher led the work than when a visiting artist did. That is school drama, not workplace role play, but it supports the same habit: the person who knows the group runs the practice.

Putting role play to work

Role play pays off at work when the brief and the debrief carry the teaching. The acting is the cheapest part. Give the other person a hidden concern, watch for three to five behaviours and write down the words, debrief with the learner speaking first, and check a week or two later whether the behaviour showed up in a real conversation. The research is strongest for debriefing and for health care. For corporate sales and service it is thinner, so treat the follow-up as your own evidence.

Your next step: pick the one conversation your team gets wrong most often, write a brief for both sides with a hidden concern and an opening line, and run a 45-minute session with an observer next week.