Five Problems Where AI Is the Wrong Tool

A commercially awkward article

We build AI systems and we turn down AI projects, for consistent reasons. The projects that should not have used AI are the ones that damage a business's appetite for the ones that should.

Five cases below. If yours matches one, the cheaper alternative is genuinely better.

1. The rules are knowable

If you can write the logic down — and for pricing, eligibility, routing and thresholds you usually can — then rules are faster, cheaper, perfectly explainable and impossible to talk into doing something odd.

Try to write the rules first. If you succeed, you have solved the problem for a fraction of the cost. If you fail, you have learned something useful about why it is hard.

2. The volume is small

Automating a decision made four times a week rarely justifies a project, whatever the technology. The build cost, the evaluation and the ongoing monitoring are largely fixed regardless of volume.

A person doing it carefully is usually better and cheaper below a few hundred instances a month.

3. The decision must be explainable

Where you must explain to a customer, a regulator or a tribunal exactly why a decision went the way it did, probabilistic output is a liability. Lending, employment, benefits and anything with a right of appeal fall here.

Use rules for the decision, and use AI at most for surfacing information a human considers.

4. There is no way to check the output

If verifying an answer takes as long as producing it, automation has moved effort rather than removed it. Long-form analysis where correctness cannot be spot-checked is the common example.

The four AI shapes that work — extraction, classification, retrieval-based answering, first drafts — all share the property that a human can check the output in seconds.

5. The real problem is upstream

• Data is incomplete or contradictory — no model infers what was never recorded
• The process changes monthly, so any encoding is obsolete on delivery
• Two departments disagree about the rule, and the project is being used to avoid the argument
• Nobody will own the result, so it will be abandoned regardless of quality

In each case AI is being asked to substitute for a decision the organisation has not made. It cannot, and the attempt is expensive.

What to do instead

Write the rules. Fix the data capture. Have the argument about the process. Buy something off the shelf. Or do nothing for six months and see whether the problem persists.

Any of those beats a five-figure AI system that nobody trusts.

Frequently asked questions