Every few days someone who runs a sales team or a contact center asks me some version of the same question: does role play actually work? I've started answering it with a question of my own. Does shooting baskets make you better at basketball?
Nobody has ever told me no. You get better at a skill by doing it yourself, clumsily at first and then less so, until the motion becomes yours, and watching other people do it on video does almost none of that work. A sales call is the same kind of skill, and a rep who has already worked through a hundred practice conversations walks into the real one warm, which shows in the first thirty seconds.
For years there was a fair objection to AI role play, which was that the bots felt fake. The voices were robotic, the responses were canned, and the pretend customer never once went off script, so reps humored the exercise and learned nothing from it. That objection has finally disappeared.
Realism stopped being the problem
The models we have today, and the voices layered on top of them, are good enough that a practice call now feels like a real one. The prospect interrupts, loses the thread, and throws an objection nobody scripted, and when the rep's answer is weak, the bot keeps pushing instead of letting it slide. A couple of minutes in, most reps forget they are talking to software at all. For years realism was the barrier between practice and real improvement, and that barrier has come down.
The hard part is doing enough of it
Once a practice call feels real, whether a program works comes down to something far less glamorous than the AI itself, which is how often the practice actually happens. One impressive role play a quarter changes nobody, while the teams that move their numbers practice constantly and practice with intent.
In my experience three things separate a program that compounds from one that stalls:
- Targeting. Each rep should practice the specific lines they handle worst, which means the practice has to be driven by what their real calls reveal rather than a generic script the whole team runs.
- Frequency. Short weekly reps beat one long quarterly session, because fluency is built through repetition spaced close together in time.
- Verification. A rep should prove they can handle the scenario in practice before taking it back to a live customer, so the improvement is confirmed rather than assumed.
All of that is a logistics problem rather than a technology problem. Someone has to decide which rep practices which call, build the scenario, set the standard it is scored against, assign it, and then run the whole cycle again the following week for every rep on the team. Done by hand, it collapses inside a month, because the manager in charge of it already has a full-time job. This is the same discipline behind call center simulation training, applied every week instead of once, and practice at scale lives or dies on how hard it is to set up.
So we made setup a conversation
This is the part most people underestimate, and it is where we put our energy. We built Itero API-first, so every scenario, assignment, and scoring rubric is configurable through the API. You can point a coding agent like Claude Code at it and stand up an entire practice program by describing what you want in plain language. Tell it the call your reps need to master, the objections that should come up, and the standard to score against, and it configures the whole thing for you. Setup that used to cost a manager a week of fiddling now takes a conversation.
That is the shift that matters. Practice has always worked, but making enough of it happen for every rep, without exhausting the person in charge of it, never did. Once that cost drops close to zero, you can finally give every rep as many reps as they need, which is something that simply was not possible before.
And then people don't get a little better
When practice becomes this easy to run, the gains stop looking like ordinary training gains. A national auto-finance lender we work with put their contact center agents on assigned weekly practice and cut average handle time by 31%, and because handle time drives almost everything downstream in a contact center, their margins rose 17% behind it. A different customer, selling rather than servicing, watched their win rate climb 23%, which is the kind of jump you almost never see from a change in training.
Those are not the numbers you get from a better training video or a sharper deck. They are the numbers you get when people arrive at the real conversation already fluent, because they have had the reps, dozens of them, against a customer who fought back, before any of it counted. Fluency compounds, and a rep who is a little sharper on every call turns into a much stronger one by the end of the quarter.
So when someone asks me whether role play works, I no longer think that is the question worth arguing about, because it works the way practice has always worked. The question I care about is whether you can make practice happen often enough, and easily enough, that it stops being an event and becomes a habit. That is what we have been building at Itero, and it is what actually moves the numbers.
If you want to see what that looks like for your own team, come take a look at Itero.
