How to Vet an AI Consultant Company Before You Sign

Vet an AI consultant company on what it has already built, not on what it says it can build. Before you sign, see a system one of its clients still runs, find out who at the firm will actually do your work, and read the contract for who owns your data and the finished workflows. The rest is sales.

TL;DR

• See a live system a client still runs, then call that client yourself.
• Find out who does the work and whose platform it runs on.
• Read the contract for data use, ownership, and an exit plan before you argue about price.

What should you check before the first call?

Ten minutes on the firm's site tells you what kind of company it is. You're looking for specifics: a named task, the kind of business it was built for, and a result with a time frame. "We help you win with AI" is a slogan. "Built an intake system that sorts service requests for a 12-person plumbing company" is a claim you can check later.

Look at how the examples are described, too. If every case study is about a platform the firm partners with, you may be talking to a reseller. That isn't automatically bad, but you want to know it before you hear the recommendation.

And be careful with lists of "top AI consulting companies." Check who published the list. If the author's own firm sits on it, you're reading an ad.

What should you ask on the first call?

A good first call is mostly the firm asking about your process. Save these five questions for the second half.

1. "Can you show me a system you built that a client is still using?" Live, on screen, not a recorded demo or a slide. If nothing they've built is still running, that's your answer.
2. "Who will do the work?" The person selling you may not be the person building. Ask for the builder's name, whether any of the work is subcontracted, and who takes over if that person leaves.
3. "What will this run on, and does that platform pay you?" Some firms earn referral fees or reseller margins on the software they recommend. A firm that's paid by a platform has a reason to recommend it. Ask plainly and write down the answer.
4. "What happens when it breaks in month three?" You want a name, a response time, and a price, not "we'll take care of you."
5. "What results do you promise?" The right answer is a measurable target you agree on together, like hours saved on one task. A guaranteed revenue number is a warning sign, and the section on promises below explains why.

Once a proposal arrives, score it against what a good AI consultant should leave you with: a system running on your real work, accounts in your name, and a document someone else could maintain it from.

How do you check references without getting a sales pitch?

Ask the firm for two clients whose work looks like yours, and one engagement that didn't go to plan. The third request matters most. How a firm talks about a project that went sideways tells you how it would handle yours.

Then ask to speak with the person who uses the system every day, not the executive who signed. The owner remembers the kickoff. The office manager knows whether it still works on a Tuesday afternoon.

Questions worth asking the reference:

• Is the system still running, and on whose accounts?
• Who fixes it now when something changes?
• Did the project finish on the date and budget in the proposal? If not, why?
• What surprised you after launch?
• Would you hire them again for the next process?

A reference who hesitates on the last question has told you more than the rest of the call.

What should the contract say about your data and the finished work?

This is the easiest section to skim, and it's the one that decides whether you can ever leave.

Start with your data. A law firm's guide to AI vendor agreements puts it simply: the contract should restrict the vendor's use of "customer data, prompts, inputs, and outputs for training or improving models unless the customer has expressly permitted such use." The same guide says the contract should, at the least, "grant the customer a broad license to use outputs for any lawful business purpose without restriction" (Ward and Smith).

Then look for an exit. Gartner's advice to large companies hiring outside engineers to build AI agents applies at any size: "Establish key contractual requirements beyond procurement's scope, including deliverables, knowledge transfer, intellectual property rights and transition or exit responsibilities" (Gartner). In plain terms, the contract should say what you get back if you end the relationship: the workflows, the prompts, the logins, and the documentation, in a form another builder can use.

If the contract is silent on any of this, ask for it in writing. A firm that pushes back on your owning your own data is telling you how the relationship will go.

Which promises should make you suspicious?

AI is new enough that some firms sell certainty they can't deliver. Four promises deserve a hard look:

• Guaranteed revenue or earnings. In March 2026 the FTC settled with Air AI, a company that sold AI to small businesses, over charges including that it "falsely claimed that people who purchase their services will or are likely to make substantial earnings" (FTC). That's an extreme case, but the pattern to watch for is the same: a specific dollar outcome promised before anyone has seen your process.
• A refund policy that only exists on the call. If it matters, it goes in the contract.
• "Our proprietary AI." Ask which AI model it's built on and who owns that model. A firm that won't say may be hiding a markup or a dependency you'd inherit.
• "Fully autonomous from day one." Anything that spends money, emails a customer, or changes records should start with a person approving it. A firm that skips that step is optimizing the demo, not your business.

How do you compare two firms side by side?

Once you've had two calls and a reference conversation for each, fill in a table like this. Start your decision with the row where the answers differ most.

Price belongs at the bottom of this list, not the top. The cheaper firm that keeps your logins is the expensive one.

Before you book any calls, settle whether you need an AI strategy consultant or someone who builds, because you'd vet those two differently. And once you pick a firm, know what AI implementation should look like from first call to handoff, so you can tell early if the work is drifting.

Frequently asked questions

Plan for two calls and at least two reference conversations before you see a proposal. If a firm pushes you to sign before you've done that, ask why the rush.

Not by default. The risk with a small firm is depending on one person, so ask who covers if that person is out. The risk with a large one is paying for layers of staff you never talk to.

It can be worth it if it's fixed-price, short, and ends in a scoped proposal you're free to take to another builder. If the discovery output only works with that firm, you've paid for a sales process.

Yes, and you should. Until the contract is signed, share a small set of sample records rather than a full export of your customer data.