How to Choose an AI Consultant (7 Questions to Ask First)

There is no shortage of people who will take money to help you with AI right now. The market is new enough that credentials mean very little and confident language means even less. So evaluate the way you’d evaluate any other vendor you can’t easily un-hire: on incentives, on evidence, and on how cheaply you can find out you were wrong.

The seven questions worth asking

Ask these on the first call. You’re not testing knowledge — you’re testing whether the answers are specific.

1. “When did you last tell a client not to buy something?” The most important question, and the hardest to fake. A real answer names the situation and what they recommended instead. 2. “How do you make money?” Fees only, or do they take commission or markup on software they recommend? Ask it flatly and watch how comfortable they are. 3. “Who does the actual work?”If the person on the call isn’t the person on the project, you’re paying senior rates for junior delivery. 4. “What number will this move, and by how much?”Hours saved, days off a cycle, error rate, cost per order. If the answer is “efficiency” or “transformation,” the engagement has no scoreboard. 5. “What’s the smallest version of this?”Good advisors have a cheap first step ready. People selling programs don’t. 6. “What happens in month three if it isn’t working?” You want a specific exit, not reassurance. 7. “What do I own at the end?” The automations, the prompts, the data, the accounts. Get it in writing.

The red flags

The recommendation always requires their product.Every finding conveniently points to the platform they resell. That’s a sales funnel wearing an advisory hat. The proposal leads with technology.If you hear which model, which vendor, which platform before you’ve been asked what’s actually slow and expensive in your business, the diagnosis is working backwards from the answer. Nothing is reversible. Long terms, full prepayment, data in their accounts, a two-year license underneath the advice. I have paid for that lesson in another business — $4.8 million in cheap money that cost $300,000 at the closing table and nearly strangled a profitable company. The sticker price was never the problem. The rigidity was.

What “experience” should actually mean

Not years in AI — the field is too young for that to sort anyone. What you want is someone who has run something: made payroll, carried risk, and had to decide whether a technology was worth the money with their own money on the line. That’s a different instinct than knowing which model shipped last month, and it’s the instinct that keeps a project pointed at a number instead of at a demo. Most failed AI projects were never tool problems — they were execution problems, and execution is a thing you learn by being accountable for it.

Ask what went wrong on a project and what it cost. Anyone with real operating history answers that easily. A polished record with no scars in it usually means someone else was carrying the risk.

Test cheaply before you commit

The best protection against choosing wrong is refusing to choose big. Buy one defined piece of work with a fixed price and a fixed end date — usually an AI Opportunity Audit or a single workflow — and judge the real thing instead of the pitch. Three weeks of actual work tells you more than three more proposals: whether they ask good questions, whether they hit dates, whether they tell you things you didn’t want to hear.

Before the first call, it’s worth knowing what shape of help you’re shopping for and roughly what it goes for — see what a fractional AI advisor does and what AI consulting actually costs. Going in with a range in mind changes the conversation.

The bottom line

Hire the person who is willing to make your project smaller. Enthusiasm about AI is cheap and mostly genuine; discipline about where it pays is rare and it’s the entire value of the engagement. If the first conversation makes your plan more expensive and more ambitious, keep looking.