
How to practise difficult employee conversations with an AI avatar
A difficult conversation rarely goes wrong because the manager didn't know what to say. It goes wrong in the second or third minute, when the other person reacts in a way nobody planned for (defensive, silent, upset) and the prepared sentences stop fitting.
That is why reading about difficult conversations helps less than people hope. It is a skill you speak, not one you know. And the usual way to practise it, a role-play in a workshop in front of colleagues once or twice a year, is exactly the setting in which most people don't dare to try anything new.
Which conversations are worth practising
- Critical feedback on performance or behaviour
- Saying no to a request for a raise or a promotion
- Repeated lateness, absence or missed deadlines
- Mediating between two colleagues in conflict
- Announcing a change the team will not like
- Talking to an employee or a candidate who is upset
What they have in common: the other person's reaction decides the conversation, and you only get one attempt at the real one.
Prepare the conversation in four steps
1. Decide what should be different afterwards
Not "talk about the deadlines", but "we agree on a date for the next delivery and on what happens if it slips". A goal you can check at the end keeps the conversation from drifting.
2. Collect concrete observations
Two or three specific situations with dates, instead of judgements like "you are unreliable". Observations can be discussed. Labels only get defended.
3. Write your first two sentences
The opening sets the temperature. Name the topic and the reason in the first half-minute, without a long warm-up that leaves the other person guessing what is coming.
4. Expect the reaction
Denial, blame, tears, silence, a counter-attack. Decide in advance how you will respond. Usually that means listening first and coming back to your goal afterwards.
Why you should say it out loud
Preparation on paper stops exactly where the conversation becomes difficult. Saying the sentences out loud shows where you hedge, where you over-explain, and where you answer the question but not the objection behind it. Practice works when three things come together: repetition, fast feedback and a setting in which failing costs nothing.
How practice with an AI avatar works
In Persohap you talk to an AI human: a video avatar that speaks, listens and reacts. You choose one of ten avatars and a ready-made character such as a difficult manager, a frustrated customer or a negotiator, or you describe your own scenario in one line and the profile is drafted for you. You set a goal, press push to talk and speak, in English or German.
Every turn is scored on eight dimensions. Four are about the conversation: goal achievement, conversation flow, sentiment and the emotional curve. Four are about you: psychological safety, cognitive load, emotional stability and decision energy. Each score links to the message it came from, so the feedback quotes what you actually said instead of giving a general impression.
When the session ends, the avatar gives you spoken feedback first. The written report follows, with three strengths and three fixes, each tied to the turn it came from. Then you can run the same scenario again, as often as it needs and without an audience.
What stays private
Sessions belong to the person practising. Coaching results are not visible to HR, there is no league table, and there is no pool of other people's scores to be compared with. You see your own sessions and your own trend.
A short routine before the real conversation
- Write down the goal and two or three observations (ten minutes).
- Run the conversation once with the avatar and read your weakest turn.
- Rewrite that part and run the scenario again.
- Go into the real conversation with an opening you have already said out loud.

What belongs in an employee database: a checklist
Personal details, contract, organisation, payroll and tax, development: what an employee database should hold, what to leave out, and how to get from a spreadsheet to a working database.
Read article
We publish the review formula before anyone is scored
Goal progress 30%, kudos 15%, your department's own yardsticks 55% — and the reward scheme is fixed during setup, before a single person is graded. Here is why the order matters more than the numbers.
Read article
One yardstick at a time, across the whole team
Judging a whole person in one pass invites the halo effect, where a strong impression on one measure bleeds into every other rating. So the scoring board is built the other way round.
Read articleSee it working, not just described.
Every decision in these posts is visible in the product. We will walk you through whichever one you care about.

