TL;DR
- The AI consulting market mixes genuine builders with slideware sellers wearing the same vocabulary — interview questions expose the difference faster than references do.
- The twelve questions below cover five areas: build skill, measurement, security, ownership, and pricing.
- The single fastest filter: "Show me something you built that's running in production today." Builders demo; advisors deflect.
- Insist on written answers about code ownership and data handling before signing — those two clauses decide how expensive it is to ever leave.
Hiring an AI consultant in 2026 is hard for an unusual reason: everyone's website says the same thing. "AI transformation," "proven frameworks," "responsible AI" — the language of firms that ship and firms that don't is identical. What isn't identical is how they answer specific questions. Here are the twelve we'd ask any firm — including us.
Build Skill: Can They Actually Make Things?
1. "Show me something you built that's running in production today."
Not a demo environment, not a case study PDF — a live system with a real user base. Builders light up at this question. Firms whose product is advice change the subject to methodology. If AI ends up automating your invoicing, you want the people who wired up someone's invoicing, not the people who wrote a white paper about it.
2. "Who exactly will do the work, and what have those people shipped?"
Larger firms commonly sell with senior people and deliver with junior ones. Ask for the names on your engagement and what each has personally built. At a small studio the answer is short because the sellers are the builders; either answer can be fine, but you should know which one you're buying.
3. "What's in your own stack — and how do you use AI in your own operations?"
A firm advising you on AI adoption should be conspicuously good at using it themselves. Ask how AI accelerates their delivery (the honest answer in 2026 involves heavy use of coding agents — see our piece on AI coding agents in software development) and what it does for their own back office. "We're still exploring internally" is a disqualifying answer from someone selling exploration.
Measurement: Will You Be Able to Prove It Worked?
4. "What baseline will we record before the build starts?"
Serious firms measure the current state — hours, error rates, turnaround times — before automating anything, because without a baseline no result can ever be demonstrated. If measurement isn't in the proposal, the vendor is planning to declare success by vibes.
5. "What would failure look like, and what happens then?"
A firm that has shipped real pilots has also had some miss their targets and can tell you how that was handled — descoped, rebuilt, or refunded. A firm that claims a perfect record either hasn't shipped much or doesn't measure.
6. "Have you ever told a client not to automate something?"
The best consultants say no regularly: the workflow was too variable, the data too thin, the volume too low to repay a build. Specific stories of advising against work are among the strongest trust signals available in this market — the same test we recommend for readiness audits.
Security: Where Does Your Data Go?
7. "Which models and vendors will touch our data, under what terms?"
You want a specific answer: which providers, whether your data is used for model training under the applicable API terms, what's retained and for how long, and where processing happens. Hand-waving here is the most dangerous red flag on this list — the full checklist lives in our guide to enterprise AI governance and security.
8. "What access will the system have, and how is it limited?"
An automation that reads your CRM does not need write access to your accounting system. Good firms talk naturally about least-privilege access, scoped credentials, and audit logs. Firms that say "we'll just need admin access to everything" should not get it.
Ownership: What Happens After They Leave?
9. "Who owns the code, prompts, and configuration when we're done?"
The right answer: you do, in a repository you control, with documentation. Vendor-retained ownership or systems that only run inside the vendor's accounts convert a project fee into a permanent dependency.
10. "What does maintenance look like, and what does it cost?"
AI systems need tending — models get deprecated, APIs change, your processes evolve. Ask what breaks first, who fixes it, and what ongoing support costs. Published retainer pricing (ours starts at $5,000/month) is a good sign; "we'll figure it out later" is not.
Pricing: Is the Money Story Honest?
11. "What does this engagement cost, and what does the next one cost?"
Get the full path priced: the audit, the pilot, and indicative build pricing beyond that. Firms with published prices — compare our AI consulting cost guide for how the models differ — make this easy. Firms that defer all pricing until after a paid discovery phase are selling the discovery phase.
12. "What do you guarantee — and what won't you promise?"
Beware guaranteed savings percentages invented before anyone has seen your workflows. The honest version of confidence is a fixed price, a defined deliverable, and a measurement plan — not a prophecy.
Scoring the Answers
| Area | Strong answer sounds like | Walk away when you hear |
|---|---|---|
| Build skill | A live demo and the names of who built it | "Our framework has been applied across industries" |
| Measurement | "We record your baseline in week one" | Guaranteed ROI percentages before seeing your workflows |
| Security | Named vendors, data terms, least-privilege access | "Don't worry, it's all secure" |
| Ownership | "Code, prompts, and docs transfer to your repo on payment" | Systems that only run inside the vendor's accounts |
| Pricing | Published or fixed prices for each stage | All pricing deferred until after paid discovery |
No firm will be perfect on all five — but security and ownership are the two where "good enough" doesn't exist.
Frequently Asked Questions
What should I look for when hiring an AI consultant?
Four things above all: hands-on build capability, not just advisory slides; a measurement habit, meaning every proposal includes a baseline and success criteria; a serious answer on security and data handling before you ask; and clean terms on code ownership and handoff. Certifications and brand names matter far less than whether the team can show you working systems they built and explain exactly how those systems are doing in production.
Should I hire a big consulting firm or a boutique AI studio?
It depends on the job. Large firms suit multi-country programs that need armies of change managers and procurement departments demand a famous logo. Boutique studios typically win on speed, price, and the fact that the people who sold the work also build it. For a first audit or pilot, paying big-firm overhead rarely makes sense; you can graduate to a larger program later if the scale genuinely demands it.
What are red flags when evaluating an AI consulting firm?
The reliable ones: no working product they can demo, only decks; guaranteed savings percentages invented before seeing your workflows; vagueness about where your data goes or refusal to sign confidentiality and data-processing terms; pricing that only becomes clear after a long discovery phase; insistence that everything must run on one specific platform they happen to resell; and no story for what happens after handoff — maintenance, updates, and who owns the code.
Who should own the code an AI consultant builds?
You should. The contract should state that custom code, prompts, configurations, and documentation transfer to you on payment, with source in a repository you control. Consultants may reasonably retain their pre-existing tools and libraries, and that is fine if the boundary is written down. If ownership terms are vague, or the system only runs inside the vendor's accounts, you are renting, not buying — price the exit before you sign.