10 Questions to Ask Before Hiring an AI Consultant (2026)

What Questions Should I Ask an AI Consultant Before Hiring Them?

Ask questions only a practitioner can answer well. That's the whole method. You can't out-expertise a consultant in their own field, and you don't have to — generic consultants survive on abstraction, so every question below is built to make abstraction impossible. The ten that do it, in the order I'd ask them on a first call:

1. Walk me through an AI system you personally built, end to end. What almost killed it?
2. Which tools would you use for my workflow — names, not categories?
3. What's the fixed price for a defined deliverable?
4. Who actually does the work?
5. If we like the plan, will you build the first piece?
6. What do we own when you leave?
7. What happens when a tool you recommended gets repriced or discontinued?
8. What's the smallest engagement you've done that worked?
9. How do you keep AI output from reaching customers unreviewed?
10. What would make you tell us NOT to use AI for this?

A practitioner answers all ten inside a 30-minute call, in specifics, without a follow-up email. If three or more come back as methodology, you're talking to a presenter. The next section gives you the good answer and the bad answer for each one, so you can score the call while it's happening.

I sit on both sides of this table — I sell AI consulting, and I've hired consultants for my own businesses. These are the questions I'd want a client to ask me, because they're the ones a presenter can't fake through. (For the fuller selection framework, see how to choose an AI consultant.)

The 10 Questions to Ask Before Hiring an AI Consultant (With Good and Bad Answers)

1. "Walk me through an AI system you personally built, end to end. What almost killed it?"
   ✅ Good: immediate, specific, includes a failure — "the calendar connector kept dropping auth, we rebuilt the flow twice."
   ❌ Bad: "Our methodology begins with a comprehensive discovery phase." That's a no.

2. "What tools would you use for my workflow — names, not categories?"
   ✅ Good: "Claude with a project setup, your CRM connected, drafts routed for your approval — because your bottleneck is writing-heavy."
   ❌ Bad: "We're vendor-agnostic and would evaluate leading platforms." In 2026, that's a consultant who hasn't picked a lane.

3. "What's the fixed price for a defined deliverable?"
   ✅ Good: a number attached to an outcome. An assessment costs $X; this install costs $Y.
   ❌ Bad: open-ended hourly with no cap. Their inefficiency becomes your invoice.

4. "Who actually does the work?"
   ✅ Good: "Me" — or named people with stated hours.
   ❌ Bad: "team-based delivery model." Translation: the partner sells, associates you've never met deliver.

5. "If we like the plan, will you build the first piece?"
   ✅ Good: yes, with a price and a timeline. (This is why I structure my own assessment so the fee credits toward a build within 90 days — the plan should be a deposit on execution, not the product.)
   ❌ Bad: "we partner with implementation vendors." Now you're managing two firms and a handoff gap.

6. "What do we own when you leave?"
   ✅ Good: everything — prompts, configurations, handoff docs, accounts in your name.
   ❌ Bad: anything routed through their platform, their licenses, or their retainer. That's a dependency wearing a deliverable costume.

7. "What happens when a tool you recommended gets repriced or discontinued?"
   ✅ Good: an actual answer about swap costs and how they design for it. The AI tool market churns fast; anyone who's shipped has been burned and has a policy.
   ❌ Bad: a blank look. They haven't been in the game long enough to have lost a tool yet.

8. "What's the smallest engagement you've done that worked?"
   ✅ Good: a number within an order of magnitude of your budget.
   ❌ Bad: every example is 10x your size. Enterprise muscle memory will over-scope your business — see the sizing discussion in the hiring guide.

9. "How do you keep AI output from reaching customers unreviewed?"
   ✅ Good: a concrete approval flow. Mine is one sentence: AI drafts, you approve, nothing reaches a customer without your yes.
   ❌ Bad: "the models are very reliable now." They are — right up until the one time they aren't, in front of your best client.

10. "What would make you tell us NOT to use AI for this?"
    ✅ Good: a real answer with real conditions — low volume, judgment-heavy edge cases, data that can't leave your walls.
    ❌ Bad: "AI can add value in every area of your business." A consultant with no no is a vendor with a quota.

The 4 Questions Consultants Hope You Won't Ask

The ten above are the ones I'd want asked. These four are the ones I'd brace for — and almost nobody asks them, which is exactly why they work:

1. "Show me something you built that's live right now. Not a case study — a URL."
   Case studies are written after the fact by whoever is best at writing. A live system is a fact. Anyone who has actually shipped has three of these on their phone within ten seconds. The dodge to listen for: "most of our work is under NDA." Some of it always is. All of it never is.

2. "What percentage of your revenue comes from implementation versus strategy documents?"
   This one is uncomfortable because it's diagnostic. A firm that makes its money on decks will describe your install as a "phase two" that somehow never gets quoted. You want someone whose revenue depends on the thing working after the invoice clears.

3. "Tell me about an engagement that failed. What did you refund or fix?"
   Everyone who has done twenty installs has blown at least one. The good answer is specific and slightly painful to tell — wrong workflow picked, client's data was worse than described, tool got deprecated mid-build — and it ends with what they did about it. "We haven't had one" means either not enough reps or not enough honesty. Neither helps you.

4. "What will this cost me to run every month after you're gone?"
   The install price is the number everyone quotes. The run cost is the number that bites: seats, API usage, the two subscriptions the workflow now depends on. A practitioner has this on hand within a range — a Claude Team plan starts at $125/month for five seats, most small-business AI stacks land between $100 and $400/month all-in. A presenter has never had to answer it, because they were never around at month three.

The Reference Call: 3 Questions to Ask Their Last Client

Ask for one reference — a client from the last six months, same size as you, not their flagship logo. Then ask these three, in this order, and stay quiet after each:

• "What's still running today that they built?" The gap between what was delivered and what survived contact with the business is the only quality metric that matters. If the honest answer is "we stopped using it in March," you just saved yourself the fee.
• "What did your team have to do that you didn't expect?" Every install has client-side labor — intake calls, cleaning up data, learning a new approval habit. A good consultant scopes it out loud in advance (mine runs about 3–6 hours of the owner's time). A bad one discovers it on your calendar.
• "Would you buy the same thing again at the same price?" Not "were you happy." References are trained to say happy. Making them re-underwrite the purchase gets you the real answer, and the hesitation before it tells you as much as the words.

What This Should Cost in 2026 (So You Can Price the Answers)

Questions are only as good as your ability to price the reply. Current market, as of July 2026:

• Solo consultants: $150–$400/hour; project engagements $5K–$50K.
• Boutique firms: $300–$800/hour.
• Big-firm AI practices: $500–$2,000/hour, six-figure minimums.
• Productized fixed-fee — the small-business entry point: a written readiness assessment around $2,500, a single-workflow install from $4,500, ongoing support commonly $1,500/month and cancellable.

If the quote you're holding is 3x one of those bands, that's not automatically wrong — but it's a question. Ask what's in it that the band doesn't include. The answer is usually "people," and people you'll never meet are the most expensive thing on any consulting invoice. (Full cost breakdown here.)

What Are the Key Benefits of Hiring an AI Consultant?

Assuming the candidate survives the ten questions — here's what you're actually buying, without the brochure gloss:

Speed to a working system. Most owners spend 3–6 months poking at tools between real work, and end with five half-configured subscriptions. A practitioner compresses that into a 2–4 week install because they've made the mistakes already, on someone else's clock.

Avoided spend. The unglamorous benefit nobody markets: killed subscriptions, prevented wrong-tool purchases, consolidated overlap. A decent tool audit routinely pays for itself before anything new gets installed.

Judgment transfer. Every install a consultant has done before yours is pattern recognition you inherit — which workflows resist automation, which tools die within a quarter, which "AI features" in software you already own are worth switching on. You're not renting intelligence; you're skipping expensive experiments.

For a small business, all three add up to one thing: capacity without payroll. The definitional guide covers this mechanism in detail.

When You Should NOT Hire an AI Consultant

The honest section, since question 10 cuts both ways:

• Under ~$500K revenue. A $25/month subscription and one focused weekend gets you most of the value. Spend the consultant money on demand generation.
• No nameable workflow. If you can't point at a process that eats 5+ hours a week, there's nothing to install yet. Measure first.
• The problem is demand, not capacity. AI makes operations faster. It does not make a weak offer strong or conjure customers. Wrong tool for that job.
• You want autopilot. Anyone selling unattended AI running your customer relationships in 2026 is selling risk with a subscription attached. Walk.

Do the Free Homework First

Before any sales call, get your own baseline: the free AI Readiness Checklist scores your business in five minutes and tells you whether your gap is tools, workflows, or data. Walking into a consultant conversation with your score — and one sentence about the workflow that hurts — flips the power dynamic of the entire call.

And if you'd rather sanity-check a shortlist with someone who does this daily: book a free 30-minute strategy call. No pitch. If the answer is "you don't need a consultant yet," you'll hear that too.

Related guides: how to choose an AI consultant, AI consultant: the full hiring guide, what is AI consulting, AI implementation consultant, do you need an AI consultant at all?.