Where AI doesn’t fit

Brenda Frisk
  • A manager stands alone in an office of empty desks, every computer screen glowing gold.
    Image concept by Brenda Frisk, generated with AI.

There’s no shortage of advice on where to use AI. I see far less on where not to use it, and I think that’s the more useful place to start.

A lot of money is going into AI right now with not much yet coming back. In PwC’s 2026 Global CEO Survey, 56% of CEOs said they’d seen no significant financial benefit from AI so far (PwC).

I don’t think that’s because the technology doesn’t work. I think it has more to do with where it’s being put, and what’s being taken out to make room for it.

One of the biggest examples I’ve seen is boards and companies cutting roles and staff to bring in the savings of AI, without knowing the cost of losing those people. I’ve experienced this myself, and I’ve seen it in several organisations.

The savings are easy to count. What leaves with those people is much harder to see. A finance person once told me redundancy wasn’t personal, it was just a numbers game. And that’s the trouble. The numbers show what each person costs. They don’t show what each person knows. Valuable expertise gets overlooked. People leave. Knowledge disappears. And sometimes the organisation ends up rebuilding capabilities it already had. Others may never recover, or understand the opportunities and capabilities they have lost.

Underneath it is something that was there long before AI. Many organisations struggle to understand what their business knows, and what their people are capable of.

So when someone asks me whether AI belongs in a part of their business, there are some questions I’d start with. These four are a sample, not the full list. Here’s each one, and how I’d go about answering it.

1. What does this role know, and not only do?

A position description tells you the tasks. It doesn’t tell you what the person has learned over time.

To find out, I’d sit down with the people in the role and ask:

  • What do you check that isn’t written down anywhere?
  • Who comes to you for help, and what do they ask?
  • What goes wrong when you’re away?

Write down what you hear. A page per role is enough. For most organisations it’s the first time that knowledge has been recorded anywhere.

If the only description you have of a role is the position description, you don’t know enough yet to change it.

2. What would we lose, as well as save?

Most business cases I’ve seen for AI have the savings worked out in detail. Very few have the other side.

I’d put these next to the saving:

  • Who relies on this role, inside and outside the organisation, and what would they do without it?
  • What relationships sit with this person and not with the organisation?
  • What would it cost to rebuild this capability if we got it wrong?

Then ask who in the organisation can answer each one. If nobody can, that’s your answer, and until someone can, the business case isn’t ready for a decision.

3. Does the work need understanding and capability, or only access to information?

AI is very good at finding and pulling together information. But access to information isn’t the same as understanding it. And knowing something doesn’t mean being capable of doing it.

A simple way to tell them apart is to ask: if the answer was wrong, who would notice, and how?

If anyone could spot the mistake, AI may fit well. If it would take someone with years in the work to see it, then the understanding is the job, and that’s the part to protect.

4. What do the people closest to the work say?

They know where things break. They also know where the time goes on work nobody values.

Don’t ask these questions with the aim of having AI take over their role. Ask because you’re interested in who they are, what they do and the value they bring to the organisation.

I’d ask them before the decision, not after:

  • What slows you down that you’d hand over tomorrow?
  • What would you never hand over, and why?

In my experience they’ll tell you where AI would help, more precisely than anyone else in the organisation.

This is also the question that gets skipped most often, and I think it’s the most important of the four.

I’ve seen what happens when leaders don’t trust the people closest to the work. The decision gets made above them, and they hear about it afterwards. By then the expertise that could have shaped it has been ignored, and the people who hold it have started to leave.

An organisation that doesn’t trust its own people has no way of finding out what it knows. And if it doesn’t know that, it can’t know where AI fits.

What to do with the answers

Start small. Pick one role where a change is being considered, and ask these four questions before the decision is made, not after.

Where the answers are clear, AI may well have a place. Where they’re not, I’d hold off until they are. That doesn’t mean never. It means there’s more to understand first.

These questions are a place to start. In practice it’s more complex than four questions, and every organisation will have its own to add.

This isn’t an argument against AI. It opens up possibilities. But the possibilities aren’t the same for every organisation, and there’s a lot to understand before deciding what should change. That includes working out what’s worth keeping.

If your organisation is working through these questions, this is some of the work we do at Department of Future. Message us on LinkedIn.

(By Brenda Frisk, Department of Future)

Making trust visible. For your now and next.