Not every problem is an AI problem.

Why does everything suddenly look like an AI problem?

Because the market is holding a very expensive hammer. Vendors need every problem to be an AI problem, boards are asking for an AI strategy before anyone has defined a problem, and teams under pressure reach for the tool with the budget attached.

The result is familiar from every previous wave: solutions in search of problems, pilots chosen for optics, and a growing quiet suspicion inside the organisation that none of it is landing.

What are the problems AI cannot fix?

The sorting question that comes first is simple and it is not about technology at all: what becomes better?

The structural ones, and there are more of them than anyone likes to admit. Unclear ownership. A process that is broken by design. Data nobody trusts. Two teams that do not talk to each other. Put AI on top of a broken process and you get the same mess, faster.

Then there are the traditional ones. A surprising share of what lands on our desk as an AI brief is solved better by boring technology: an integration, a report, a workflow tool, a rule that runs every night without a model in sight. Boring technology is cheap, reliable and already understood by your teams. It should win whenever it can do the job.

How do you tell the difference?

We use four questions, and they take an hour, not a quarter:

• What decision or outcome actually gets better if this succeeds?
• Would a simpler tool do the same job? If yes, why isn’t it?
• Is the underlying process worth automating at all, or does it need fixing first?
• Who owns this after the project team leaves?

A genuine AI problem survives all four. Most briefs do not, and finding that out in an hour is a far better result than finding it out at the end of a six-month engagement.

What happens when you get this right?

Money lands where the value is. Staff stop rolling their eyes, because the initiatives they see make obvious sense. And your AI credibility, the scarcest resource in most organisations right now, compounds instead of eroding.

There is a test for advisers hidden in here too. A partner who will look at your brief and say “you do not need AI for this” is arguing against their own interest to tell you the truth. That is the partner worth keeping.

Common questions

Is this an anti-AI position?

No. We build and deploy AI, and have done since before the current wave. AI applied to the wrong problem wastes money and credibility, and both are hard to win back.

What proportion of problems genuinely need AI?

No honest fixed number exists. The pattern is consistent though: many problems are traditional, more are structural, and the genuine AI problems are real but fewer than the market suggests.

What should we do before choosing any technology?

State what is actually wrong, precisely. The technology choice becomes obvious once the problem is named properly.