Rehearsing it in your head is not practice

By the time Thursday comes you will have had this conversation eleven times. In the shower, in the car, at 2am with the ceiling. In every one of those versions you were articulate and she took it well.

Then you sit down opposite her, she says something you had not accounted for in the first fifteen seconds, and all eleven rehearsals go out of the window at once.

Somebody measured this

In 2024 a group at Stanford built a tool called Rehearsal: a simulated person you could argue with, replay, and try differently. They ran a proper comparison. One group studied the conflict resolution theory as lecture material. The other practised against the simulation. Then all 40 participants were put into an actual argument with someone they believed was another participant, and their behaviour was coded.

The group that had practised used escalating, competitive moves 67% less often, and used cooperative ones about twice as often, than the group that had studied the same ideas.

The sentence from that paper worth pinning above your desk is not the statistic. It is the observation that reading the strategies did not translate into applying them, and that the people who had only read them believed they were ready.

One study, 40 people, a staged conflict. It is not the last word on anything. But it is a direct test of the thing you are about to do on Thursday, and the result went one way.

Knowing and doing are two different skills

You can hold every rule in your head. Describe the behaviour, not the person. Ask before you tell. Do not fill the silence. Under pressure, none of those are knowledge problems any more. They are production problems, executed at speed, while your own heart rate is up and somebody’s face is doing something you did not expect.

Recognising the right move when you read it is easy. Producing it in the second after someone says “I think that is quite unfair, actually” is a different act, and the only known way to get better at an act is to perform it.

What makes practice actually transfer

Role-play has a bad name, most of it earned in away-day breakout rooms. The research on behaviour modelling training is more specific than “role-play works”. Taylor, Russ-Eft and Chan pooled the studies and found the design features that separated the training that reached the job from the training that did not:

• Show the bad example as well as the good one. Watching only the polished version teaches you what right looks like and nothing about how wrong starts.
• Practise on your own situations. Not a generic case study. The actual person, the actual thing they did.
• Set a goal before you start, and make it a behaviour you could tell afterwards whether you did.

Their other finding is the one that matters for a busy manager: the effect on skills held or grew as time passed, even where other outcomes faded.

And what makes it stick

The biggest meta-analysis of leadership training, 335 studies by Lacerenza and colleagues, found training works better than the cynics assume: the average programme moved a leader from the 50th to about the 78th percentile, with the strongest effect on behaviour back at work.

More useful than the headline are the conditions attached to it. Training worked better with a needs analysis first, with feedback, with practice among the delivery methods, and with sessions spaced out rather than stacked into one day.

Why the one-day course in March is already gone

Arthur and colleagues reviewed the skill decay literature and put numbers on non-use: a small loss immediately after training, and an effect of around d = −1.4 after more than a year without practice. Cognitive skills decayed faster than physical ones, and tasks that needed accuracy faded faster than tasks that needed speed.

Difficult conversations are cognitive and accuracy-based, and most managers have four of them a year. That is close to the worst case in the literature: exactly the kind of skill that fades, practised at exactly the interval that guarantees it will.

What to actually do, with or without an app

Practice, in the sense that any of this research means it, has four parts.

1. Out loud. Your mouth does something different from your head. The sentence that was crisp in the car comes out with three apologies attached.
2. Against someone who does not cooperate. The rehearsal in your head is worthless precisely because the other person behaves. Brief a colleague: hold your position, go quiet once, and at some point say “I feel like I am being singled out here.”
3. With feedback on what you did, not on what you should have said. How long before you got to the point. Whether you asked. What you said after the first push back.
4. More than once, with a gap. Twice this week and once next week beats five times tonight.

If you have nobody to practise with, a voice memo is not as good but it is not nothing. Say your opening out loud, play it back, and time how long it takes you to say what the conversation is about. Most people are shocked twice: once by the number, and once by how many times they said “just”.

The script is the thing to give up

The point of any of this is not to arrive with better lines. It is to have already been somewhere near this feeling once, so that the first unexpected sentence is not also the first time your mind has gone blank.

Go in with three things you know and one question. Everything after that is a conversation, and conversations cannot be memorised. They can only be practised.