Why I Built an AI Platform After Three Decades of Leadership Training

In 1997, on his first corporate training course, Bernhard Kerres learned an icebreaker he still uses today - and almost nothing else. Three decades of the same Monday-morning frustration, one painfully honest follow-up call with twelve senior consultants, and a summer at an Austrian mountain lake led to RolePlays.ai. The founder's story, told straight.

Published 2026-06-22 by Bernhard Kerres

In 1997, I was on my first corporate training course. Booz Allen & Hamilton had invited all the summer associates to spend a week together in upstate New York. I remember the smell of the woods. I remember one exercise: Two Truths and One Lie.

We stood in small circles of six. To get to know each other, everyone made three statements about themselves. Two were true. One was a lie. The rest of the circle had to guess which was which.

Our facilitator was in my group. One of his statements was that he had once landed a plane on a main street. We were certain that was the lie. It turned out to be true.

I still use that exercise as an icebreaker, almost thirty years later.

Here is what I want you to notice about that memory. I remember the woods. I remember the plane on the main street. I remember the structure of the icebreaker well enough to still run it.

I remember almost nothing else from that week. And I am fairly sure it changed very little about how I actually worked for our clients that summer.

That gap - between what we remember from training and what changes in our behavior because of it - has been the central frustration of my professional life. It is the reason RolePlays.ai exists.

---

Thirty years of the same Monday morning

Since that week in upstate New York, I have been on countless training courses. I have hired facilitators. I have sent colleagues and team members on programs. And eventually I started running trainings myself - first as a CEO, then as an executive coach and educator.

The same thing kept happening. The training would be good. People would be engaged. They would leave energized, full of intentions. And then Monday morning would come, and within a few weeks everything was back to exactly the way it had been before.

Let me give you one example, because the general version is too easy to nod along to.

A few years ago I was running a development program for senior consultants at a global strategy firm. As consultants get more senior, their focus has to shift - from delivering the work, to managing the project, to bringing in new business. That last shift, from project management to business development, is the hardest one. It requires letting go of the work you are good at and doing something that feels exposed and uncertain.

We ran a session on exactly this, in Italy. It went well. It was clear to everyone in the room what had to change: protect time in the calendar for business development, delegate more of the delivery to the actual project managers, and put real energy into BD every week. We planned it in detail. Everyone worked through their own calendar. Everyone blocked weekly BD time. Everyone chose an accountability buddy. It was, by the standards of corporate training, a well-designed session with a concrete action plan.

I ran a follow-up call three months later.

Of the twelve participants, exactly one had protected their business development time - and even that one person had managed it only about half the time. The other eleven had not protected it at all. Three months of good intentions, a detailed plan, accountability buddies, and the result was effectively zero.

That is pure frustration. But it is also a clue, and it took me too long to take the clue seriously.

In that Italy session, we talked about business development. We discussed it, planned it, structured it. What we never did was practice it. Nobody in that room ever sat across from a skeptical prospect and tried to open a conversation. Nobody felt, in their body, the discomfort of asking for the business. We had worked on their calendars. We had never worked on the conversation itself.

---

The polite roleplay problem

The obvious answer to "they never practiced" is roleplay. And I used roleplays for years. You play the seller, I play the buyer. We have our short scripts. We swap roles. It is genuinely useful - there is real value in getting into someone else's shoes, even briefly.

But there is a ceiling, and anyone who has done a lot of roleplay training knows exactly what it is. In the end, we are two colleagues being polite to each other. I cannot fully be the hostile CFO, because I like you and we are in a workshop and there is a coffee break in twenty minutes. You cannot fully fail, because I will, gently, help you out. We act. We are kind. We learn a little.

The buyer who actually tests you - who pushes back, goes quiet, gets irritated, offers you the easy way out to see whether you take it - almost never shows up in a workshop roleplay. And that buyer is precisely the one you need to practice with.

For years I assumed that was just the natural limit of the method. You cannot manufacture a genuinely difficult conversation partner on demand. So you settle for the polite version and hope some of it transfers.

---

A lake in the mountains

Every summer I take what I call creative time. In 2025 I was at a lake in the Austrian mountains, in the nature, thinking about the trainings I had coming up in the autumn.

I started wondering whether AI could make roleplays more realistic. Not as a gimmick - I had no interest in a chatbot with a training skin - but whether it could solve the specific problem I had been stuck on for decades: the partner who does not go easy on you.

So I started tinkering. Playing around. Experimenting at the edge of the lake and then for weeks afterward. And at some point the experimenting turned into something else, because I realized that roleplay with AI does not have to be a worse version of roleplay with a person. Done properly, it can be a different thing entirely.

An AI persona can hold a character for the whole conversation without breaking to be kind to you. It can push back when you are doing well, not cave. It can carry a hidden concern that only surfaces if you earn it. It can be the hostile CFO, the grieving stakeholder, the consultant who has been auditioning for a promotion for two years and cannot say so. It can let you fail completely, in private, and then tell you exactly where you failed - referencing the actual words you used.

That was the moment RolePlays.ai was born. Summer 2025, at a mountain lake, out of thirty years of frustration with the same Monday morning.

---

What I do differently now

I went back to that business development problem. The senior consultants who need to shift from project management to BD. This time, instead of only planning their calendars, I built practice into the work - scenarios where they actually have to open a business development conversation with a prospect who is busy, skeptical, and not especially interested in being sold to.

I piloted it. Then I started using it everywhere. The difference is not subtle. People walk into real BD meetings having already had the conversation a dozen times. They are not improvising the hardest part under pressure for the first time. They are far more secure, because security comes from practice, not from a plan written on a flipchart in Italy.

That is the whole thesis, really. Knowing what to do has never been the problem. The eleven consultants who did not protect their BD time knew exactly what they were supposed to do. The gap was never knowledge. The gap was that they had never practiced the thing itself often enough for it to feel possible.

---

Two Truths and One Lie

I still run that icebreaker, almost thirty years after a facilitator told a circle of summer associates that he had once landed a plane on a main street.

The difference is that now the icebreaker is not the only thing people remember three months later. The practice is. We have even built Two Truths and One Lie as a small scenario on the platform - you can try it yourself, as a way in.

I do not think the woods, or the plane, or the icebreaker were the problem in 1997. They were lovely. The problem was that a lovely week never touched what I actually did on Monday.

That is the thing I have spent the last thirty years trying to fix, and the thing I finally built a tool for at a lake in the mountains. Training should not just be remembered. It should change what you do when it counts - and that only happens if you have practiced it before it counts.

That is why I built RolePlays.ai.

---

If you want to feel the difference for yourself, try a scenario (https://roleplays.ai). Start with whichever conversation you would least like to have for real. That is usually the one worth practicing.