AI for team coaching and conflict handling

Updated · Thomas A. Thejn

This is the use where the potential for harm is highest, so it is worth being blunt about the boundary before discussing the benefit.

The boundary

Rehearsal for you: legitimate and valuable. Analysis of them: a data-protection problem and an ethical one.

When you write "Anna has been defensive in the last three steering meetings and I think she is out of her depth on the technical side" into a chat window, you have:

  • Processed personal data about an identifiable colleague, without a lawful basis and without her knowledge
  • Possibly processed special category data, if the situation touches health, stress or wellbeing
  • Created a record that would be seriously damaging in a grievance, a disclosure exercise, or a subject access request

And conflict situations are precisely the ones that later attract exactly those processes. That is not a hypothetical risk profile; it is the normal life cycle of a workplace dispute.

What works instead

Abstract the person, keep the situation

Almost all the value survives abstraction:

Two workstream leads disagree about who owns integration testing. One believes it sits with the vendor, the other with the internal team. It has been unresolved for three weeks and is now blocking a go/no-go decision. What is likely going on beneath the stated positions, and how would you structure a conversation to resolve it?

That gets you everything the named version would, and creates no record about identifiable people. If your description genuinely stops working once names are removed, it is worth asking what you were actually seeking.

Rehearse the conversation

The strongest use in this guide. Before a difficult conversation:

  • Draft three ways to open it and compare how each is likely to land
  • Have your reasoning argued against, so you meet the objection first in private
  • Check whether feedback you have drafted reads as specific and actionable rather than personal
  • Prepare for the responses you would find hardest

Difficult conversations mostly go wrong in the first ninety seconds, and rehearsal genuinely helps. Doing it with a model is faster and less awkward than doing it with a colleague, and unlike a colleague it costs nothing to try the version you are embarrassed by.

Understand the pattern

Conflicts in programmes rhyme. Unclear ownership at a boundary. A decision made informally that one party never accepted. Two people held to incompatible objectives by different bosses.

Describing the shape and asking what usually underlies it is useful, and it is entirely generic — no personal data involved.

Get the wording right

Feedback that lands is specific, behavioural and about impact rather than character. Most people write feedback that is vague and evaluative instead. Drafting it, then asking whether it is specific and behavioural, reliably improves it.

Keep the subject abstract while drafting. Put the name in when you deliver it, in person.

What it cannot do

Be in the room. Conflict resolves because someone with standing sits down with both parties and does not leave until something is agreed. That is the intervention. Everything else is preparation for it.

Know the history. Two people who worked badly together on a previous programme carry that in. It shapes everything and appears in no document.

Take the risk. Naming a conflict makes you responsible for resolving it. That exposure is exactly why conflicts persist unaddressed, and no tool absorbs it for you.

Hold the relationship. Team members take feedback from people they trust. Trust is built by attention over time, which cannot be delegated.

A quick reference

UseVerdict
Rehearsing a difficult conversation, abstractedYes — the best use here
Pressure-testing your own reasoningYes
Understanding a conflict pattern genericallyYes
Drafting feedback wording, subject abstractedYes
Writing a named person's behaviour or performance into a toolNo
Anything touching health, stress or wellbeing of a named personNo — likely special category data
Building a case against someoneNo — and go and talk to HR
Replacing a conversation you are avoidingNo — that is the actual task

If your organisation runs AI tooling that touches HR matters, it needs a lawful basis, a retention policy, a data protection impact assessment and transparency with staff. That is your HR and legal function's decision to make, not a project manager's. Until it exists, keep names out of it.

Where such tooling is sanctioned, an EU-hosted model is the floor rather than a refinement. This is the most sensitive category of data a programme touches, and processing it under a foreign jurisdiction is difficult to defend to a works council, a data protection officer, or the colleague it concerns.

The honest summary

This category offers a real benefit and it is narrower than the marketing suggests. AI makes you better prepared for hard conversations. It does not have them for you, and the having is where the job is.

The failure mode to watch for is not a compliance breach. It is using the tool to feel productive about a conflict you are avoiding. Drafting an analysis of the situation can feel like progress; it is not. The conversation is the progress.

Frequently asked questions

Can I describe a conflict to AI to get advice?
Describe the situation, not the people. 'Two workstream leads disagree about who owns integration testing and it is delaying a decision' gets you the same useful response as naming them, without creating a record about identifiable colleagues. If your description only works with names in it, that is a signal about what you are really asking.
Why is this higher risk than other uses?
Because the content is personal data at its most sensitive — conduct, capability, relationships, sometimes health or wellbeing, which is special category data under GDPR. It is also the material most likely to be damaging if it surfaces, and conflict situations are exactly the ones that later attract grievances, disclosure and scrutiny.
Is it useful at all, then?
Very. It is an excellent rehearsal partner. Trying three ways to open a difficult conversation, having your reasoning challenged before you commit to it, checking whether feedback lands as intended — these genuinely improve outcomes, and none of them require naming anyone.
Can AI coach my team members directly?
Some organisations are trying it, and it is a poor substitute for a manager who pays attention. Coaching works because someone credible knows the person's situation and is invested in them. A tool that does neither can offer reflection prompts, which has some value, but do not mistake it for the relationship.

← Back to Transformation and project management

Programme not behaving?

A programme review is a contained piece of work: a few weeks, an honest status picture, the risks that matter, and a prioritised list of what to do now.

thomas@thejn.dk +45 2048 3147

Copenhagen, Denmark · Nordic coverage · Independent & platform-agnostic