AI for stakeholder analysis

Stakeholder analysis is where project management quietly becomes the processing of personal data about colleagues, and almost nobody treats it that way.

That matters more now, because AI makes it frictionless to produce exactly the document you should not have.

The line

Structural mapping is fine. Who holds which role, what each role needs from the programme, where interests conflict structurally, who must be consulted before which decision. This is organisational analysis and it is entirely legitimate.

Individual assessment is not. Writing "Lars is resistant to change and territorial about his team; needs managing carefully" into a tool creates a record of your opinion about a named colleague's character. That is personal data. Lars has rights over it, including the right to ask what you hold. And a subject access request or a disclosure exercise finding that text is a genuinely bad day.

The line is not always obvious in the moment, which is why the test is useful:

Would you be comfortable if this person read exactly what you typed?

Not a legal test, but it tracks the legal answer closely and you can apply it while typing.

What to use it for instead

Structuring a map you built

Do the thinking yourself, then use AI to organise it. Give it roles rather than names — "the finance director", "the operations lead in the largest region" — and ask it to structure interest and influence, identify where consultation is missing, or spot decisions with no clear owner.

You get the structural benefit and never write a character assessment.

Audience translation

Genuinely valuable and completely safe. The same programme update needs to say different things to a CFO, an operations director and an engineering team. Drafting three versions is tedious and AI does it well.

Give it your content and the audience's concerns by role. You are describing what a finance director cares about, not what your finance director is like.

Preparing for a conversation

Before a difficult stakeholder meeting: what questions should I be ready for, what objections are likely from someone in this role, what would I want to know if I were them?

This is rehearsal, and it is one of the better uses of AI in the whole discipline — it makes you readier without recording anything about the person.

Interview questions

Starting a programme review or a stakeholder discovery round, AI will draft a solid interview guide in minutes. Edit for your context. Low risk, real time saving.

The judgment it cannot make

Stakeholder work is almost entirely about things the model has no access to: history between two directors, who owes whom a favour, which sponsor is genuinely committed versus nominally supportive, what happened on the last programme that everyone remembers and nobody mentions.

This is the substance of stakeholder management, and it comes from conversations, not analysis. If you find yourself asking a model to profile a named colleague, the honest read is that you are avoiding a conversation you should be having.

A working pattern

If your work genuinely requires stakeholder records naming individuals — and some governance frameworks do — keep them in your organisation's sanctioned systems under its retention policy, with an EU-hosted tool if AI touches them at all. Not in a chat window with a US provider.

Why this matters beyond compliance

Stakeholder analysis that reads as manipulation damages you if it surfaces, and it surfaces more often than people expect — a shared screen, a document left in a folder, a colleague picking up the wrong laptop.

The best stakeholder work has never needed hiding. It is understanding what people need and taking it seriously. Keep your written record to that, and the compliance question mostly answers itself.