How to choose an AI consultant: 10 questions to ask before hiring
Almost every founder we speak with has already had at least one bad experience with an AI or automation agency: a strategy deck with no working system behind it, a tool they don't own, a bill that kept growing after launch. These are the ten questions that would have caught each of those problems before signing anything.
Why this decision is harder than hiring a normal contractor
Most service categories have obvious quality signals: a portfolio, a licence, a track record you can verify in ten minutes. AI consulting doesn't yet, because the category is new enough that anyone can call themselves an AI consultant after a weekend with a chatbot. That's not a reason to avoid hiring one. It's a reason to ask sharper questions than you would for a website redesign.
The ten questions
01Who owns the infrastructure once it's built?
Ask directly: if this agency disappeared tomorrow, would the system keep running, and could I hand it to someone else? A good answer names the actual tools (which platform, whose account) and confirms you hold the keys. A bad answer is vague, or reveals the system only runs inside the agency's own proprietary platform that you can't take with you.
02Where does my customer data actually live?
If you're in the EU, this has a right answer: EU-hosted infrastructure, or a US provider with a valid transfer mechanism (see our data privacy guide for what that actually means). An agency that hasn't thought about this hasn't thought about GDPR either, which tells you something about how carefully the rest of the system was built.
03What happens if I stop paying you?
This is the single most revealing question on the list. Does the system keep running with reduced support, or does it stop working entirely because it's locked to the agency's account? Vendor lock-in isn't always malicious, but it's always worth knowing about before you're three months into a system you can't leave.
04Can you show me a system you've built that's still running today?
Not a case study slide. An actual, currently operating system, ideally with a specific number attached: leads processed, response time achieved, revenue tracked. Ask when it launched. A system that's been running unattended for six months tells you more than a demo built for the sales call.
05What's your pricing model, and why is it structured that way?
Fixed project fee, ongoing retainer, or usage-based: each is a legitimate model, but the agency should be able to explain why theirs fits the work rather than reciting a rate card. As a rough anchor from our own experience quoting against other agencies: a first working AI system for a small business commonly starts around €10,000 once agency overhead is priced in, though the number moves with scope. Be specifically wary of pricing based on vague "value delivered" with no defined deliverable.
06Is there human oversight on anything customer-facing?
An AI system that talks to your customers should have a defined escalation path for anything it isn't confident about, and someone should be reviewing outputs, at least in the first few weeks. If the answer is "it just runs itself," that's a system that hasn't been stress-tested against a genuinely odd customer message yet.
07What's the plan for when the underlying AI model changes?
Models get updated, deprecated, and replaced regularly. A system built with any awareness of this has prompts and logic that are reasonably portable across model versions and providers. An agency that's never thought about this question has probably built something more fragile than they realise.
08Have you worked in my industry, or will I be the first?
Being the first client in an industry isn't automatically disqualifying (every agency has a first client in every industry at some point), but you should know which one you are, and the price and timeline should reflect it. An agency that claims deep experience in twelve unrelated industries at once is usually overstating at least eleven of them.
09What does "done" actually look like?
A working system in production, handling real enquiries, that you've watched operate, or a strategy document and a set of recommendations? Both are legitimate things to sell, but they are not the same thing, and the gap between them is where a lot of AI consulting budgets disappear. Get this in writing before the contract is signed.
10Can I speak to a current client?
Not a testimonial quote on a website. An actual conversation, even a short one, with someone using the system today. Any agency confident in its work will arrange this without much friction. Reluctance here is itself an answer.
What we tell people who ask us these questions
We host on infrastructure we control (see our breakdown of the actual stack), a client can leave with the system intact, and we'd rather lose a prospective client to these questions than have them find out the answers three months into a contract. We built our own systems this way before we ever wired one up for a client, which is a large part of why we can answer all ten of these without checking anything.
The agencies that get defensive about these questions are telling you exactly why to ask them.
The honest summary
None of these questions require technical knowledge to ask. They require knowing what to ask, which is the actual gap most founders run into. Ask all ten before signing anything, and the answers will tell you more about the agency than any pitch deck will.