When NOT to use generative AI (and save the project)

The concept

The question that saves money is the opposite one

In almost every meeting I hear "how do we put AI into this?". The question that saves money is the opposite: "does this problem really need AI?".

Generative AI is extraordinary at demos: in fifteen minutes it produces something that looks like magic, and that feeling pushes people to approve projects out of enthusiasm, not analysis. In production it has to work ten thousand times, with real data, without erring on what matters and at a cost that adds up.

It shines at drafting, summarizing or conversing; it's risky when you need verifiable exactness.

The key distinction

Generative AI doesn't compute: it predicts the most likely. It's brilliant for drafting, summarizing or conversing, and dangerous when you need an exact, verifiable answer. Confusing the two uses is the most expensive mistake I see.

Red flags

When NOT to use it

You need absolute accuracy: financial calculations, inventory, amounts. A system that "almost always" gets the money right is a liability.
A clear rule solves the problem: if it can be stated as "if this happens, do that", traditional software is cheaper and doesn't err.
You can't tolerate it making things up: these models state false things with confidence; in legal or medical contexts that's unacceptable without controls.
You don't have the data —or the permission— to use it.
The volume doesn't justify the cost: each query costs, and at scale the bill surprises.

The other side

When it is a great investment

It pays off when the problem is about language, volume and a reasonable tolerance for error:

Summarizing, classifying or extracting information from thousands of documents or emails.
First-level customer support, with a human for the hard cases.
Assisting your team —drafts, searching internal manuals— so they produce more.
Tasks where a result that's "very good 90% of the time" is already a huge improvement.

Rule of thumb

Before approving any AI project, demand three answers in business terms: what decision or task it improves, what happens when it's wrong, and how much each use costs at real scale. If those three aren't clear, there's no project yet.

In short

The edge isn't "having AI"

AI isn't an end, it's a tool: it shines at some jobs and gets in the way at others. The competitive edge is knowing exactly where to apply it and where not.

Conclusion

What to do about it

The expensive mistake isn't trying AI: it's falling in love with the demo without analyzing the case. Applying judgment before investing is what avoids the projects that get cancelled halfway through with the budget already spent.

Have an AI idea and want to know if it's worth it?

In a free assessment session we evaluate your case honestly: if AI is the way, how to do it right; and if it isn't, I tell you before you spend.

Book your free assessment