AI Built a Pricing Model. Every Unit Would Have Sold at a Loss.

A founder sent me an AI-built pricing model the night before his first retail meeting. It looked great. It took me about five seconds to find two errors, and either one would have crushed his margins for years.

Error one: one product had only been on sale for eight of the last twelve months. The model filled the other four with zeros, divided by twelve and overstated contribution margin.

Error two: freight wasn’t in the cost stack at all. For a heavy physical product going into retail, that’s not a rounding error.

He’s a smart founder who uses AI well. He still sent it to me, and he was right to. Retail pricing isn’t something you fix next quarter. The channel anchors on what you give them and holds you there.

I rebuilt the model with the same AI tool and got a better answer, because I’ve priced products into retail before and knew where to look.

Where AI helps and where it fails

AI is excellent at scaffolding. It drafts board commentary in seconds and compresses hours of variance analysis into minutes. But it’s confidently wrong about anything specific to your business: industry costs, contract structures, the things you only know from sitting in the seat.

And it won’t tell you. The output looks just as polished when it’s wrong as when it’s right. A messy spreadsheet invites questions. A clean AI-generated model gets trusted.

A review checklist for AI-built models

Before an AI-built model leaves the building, check:

1. Time periods. Does every product, customer or cost cover the same months? Partial periods are the most common error.
2. Missing costs. Freight, payment processing, returns, channel fees, support. AI includes what it’s told about.
3. Units. Monthly versus annual, per unit versus per case, gross versus net.
4. Hard-coded numbers. Find every typed-in number and confirm where it came from.
5. Sanity checks. Does margin look too good? Does it tie to last year’s actual results?
6. The question behind it. Is the model answering the decision you actually need to make?

The bar went up

Using AI well in finance takes what finance always took: the experience to know when the answer is wrong. AI doesn’t lower that bar. It raises it, because you’re reviewing more output, produced faster, that looks more finished than it is.

AI compresses the time to build a model. It doesn’t compress the years of pattern recognition that tell you which assumption usually breaks first.

Build the first cut with AI. Then check it with someone who has done it before.

Common questions

Should founders build financial models with AI?

Yes, as a first draft. AI is fast at structure, formulas and narrative. Just treat the output as a draft that needs review, not a finished model.

What kinds of errors does AI make most often in models?

Mismatched time periods, missing costs, wrong units, and confident assumptions it has no way to know, like your real freight or channel fees.

Who should review an AI-built model?

Someone who has built or used that kind of model before: a CFO, controller or advisor who knows your industry. The more costly the decision, the more experienced the reviewer should be.