How to evaluate an AI consultant before you sign

Judge an AI consultant on three things: whether their answers are specific, whether you own what they build, and whether their pricing has a defined deliverable attached. Everything that impresses in a sales call, the slick demo, the confident tone, the jargon, is easy to fake. Specifics, ownership, and a clear price for a clear scope are not.

The best filter is how a firm answers a hard question. A reseller deflects to a demo or a buzzword. An engineer gets concrete: here is who would write it, here is what we shipped for a business like yours, here is what happens when the model is wrong. The rest of this guide is the questions that pull those answers out.

Three questions separate the engineers from the resellers: who actually writes the code, what have you shipped for someone like me, and what happens when the model is wrong. Each one is hard to answer well without real engineering behind it, and easy to dodge when there isn't.

Ask all three, and listen for concrete answers:

A consultant's pricing shape tells you how they think about risk and deliverables. A fixed quote after a scoping step is the healthy shape: it means they understood the work before they priced it. An open-ended monthly fee with no defined deliverable is a flag. A guaranteed financial return is disqualifying.

Here is how to read the three shapes you will see most:

A guaranteed return is a reason to walk away because nobody can honestly promise one, and regulators treat the promise as a warning sign. Deceptive earnings claims aimed at small businesses are exactly what the Federal Trade Commission polices, and in 2024 it opened Operation AI Comply, a law-enforcement sweep against companies making misleading AI claims.

A firm promising you a guaranteed result is either naive about their own product or comfortable with a claim regulators have gone after. Neither is who you want building the software your business will run on. Specific past results are worth a lot. Promises about your future returns are worth nothing, and a little bit less than nothing when they come with a guarantee.

You should own all of it: the code, the cloud accounts, and the data, in your name from day one. If a consultant builds on infrastructure only they can reach, or keeps the code where you cannot get to it, they have made leaving them expensive on purpose. Good firms make the exit easy because they plan to keep you by being worth keeping, not by trapping you.

Ask it directly: when this is done, whose accounts is it in, who holds the code, and what happens if we part ways. The right answer is that everything is already yours and any competent engineer could pick it up. Anything else is a lock-in you are being asked to accept quietly.

Ask for specific past work, not promises about your project. A firm that has done this before can describe a real engagement: the industry, the problem, what they built, and what it does now. Specifics you can check beat any assurance about the future, because anyone can promise a result and only someone who has shipped can describe one.

References count for more when they are recent and relevant. A reference doing similar work in a similar business will tell you what the firm is like when something goes wrong, which is the part that matters most. Be wary of a firm that can talk at length about AI in general but goes vague the moment you ask what, exactly, they have shipped.

Run it as a few short calls with two or three firms, followed by a small paid scoping step before you commit to any build. You do not need a long procurement process. You need enough contact to compare how each firm answers the hard questions, and one low-risk paid step that shows how they actually work before real money is on the line.