The Most Expensive AI Project Is the One That Should Have Been a Spreadsheet

The most expensive AI project is the one that should have been a spreadsheet.

I watched a company spend weeks scoping an AI solution for detecting anomalies in their data. Expected vs. actual quantities. Flagging outliers. Building models.

Then someone asked: "What if we just compared the numbers?"

That's it. A basic comparison. Regional averages vs. individual reports. Anything 25% above the norm gets flagged. No training data. No model tuning. No six-figure vendor contract.

It worked.

Not because AI couldn't have done it. AI absolutely could have. But the simple version shipped in days, cost almost nothing, and solved 80% of the problem.

This happens more than anyone admits.

Towards Data Science published a piece last year saying the same thing: start simple. Linear regression, pre-trained models, or plain heuristics. You learn about the problem. You find out why the simple version fails. Then you have a foundation for something smarter.

The companies winning with AI aren't the ones throwing it at every wall. They're the ones who know when the wall just needs a coat of paint.

Try AI on everything. Absolutely. You'll be surprised what it can and can't do.

But don't skip the obvious solution because it isn't impressive enough to put on a slide deck.