21 questions to ask any UK AI consultancy before signing

Most UK SME owners go into a first meeting with an AI consultancy underprepared. Not because they are not smart. Because nobody has told them what to ask. The conversation drifts into use-cases, demos, and project ideas. The hard questions never come up. The contract gets signed on capability promises that nobody has tested.

This article is the list of 21 questions that closes that gap. Take it printed to the meeting. Read each one out loud. Tick each answer as it lands. The act of public asking forces specific answers and stops the conversation from drifting back into the consultancy's pitch.

The 21 questions split into four groups. Six on capability. Five on commercial. Five on compliance. Five on exit terms. Five of those 21 matter more than the rest, and we summarise them in their own section after the full list.

For the wider question of how to spot a good UK AI agency from a bad one, our agency guide covers the seven red flags and four traits. This article is the question list you take to the meeting.

Capability: six questions

These six tell you whether the consultancy actually knows how to build what you need.

1. Which specific tools will you use for our build, and why those rather than the alternatives? (A real answer names platforms like Make.com, n8n, Zapier, OpenAI, Anthropic, or Google and gives a reason tied to your business. A bad answer stays at "AI models" or "intelligent automation" with no named tool.)

2. Walk me through one similar build you have done. What did it do, what did it integrate with, and what saved time?

3. Which of your team will actually do the build? Will it be the person in this room or someone else?

4. How do you handle the integration with our specific stack? (Name your CRM, booking platform, accounting tool. They should either know it or admit they will need to learn it.)

5. What does the testing process look like before the build goes live? (Look for a real plan: staged rollout, parallel running, sign-off gates.)

6. Who handles the day-to-day maintenance after launch? (You want a named person, not a help-desk address.)

Commercial: five questions

These five tell you whether the contract is in your favour.

7. What is the scope in writing, with a fixed deliverable for each line item? (A scope is a list, not a paragraph. Each item has a deliverable. If the answer is "we will scope it as we go", the consultancy is selling time, not outcomes.)

8. What is fixed in the price and what is variable? Where would the variable parts kick in?

9. What does success look like in measurable terms? (Hours saved per week, error rate reduced, response time cut. The consultancy should know these answers without needing a workshop.)

10. What happens if the scope changes during the build? Is there a fixed change-request process?

11. What references can I speak to from clients who hired you for a similar build? (Direct conversations, not testimonials on a website.)

Compliance: five questions

These five matter for any UK business handling personal data, which since the Data (Use and Access) Act 2025 came into force on 5 February 2026 covers most SMEs running customer-facing or staff-facing automations.

12. Where will our data physically live? Which data centre, which country, which jurisdiction? (UK or EEA residency is the default. Anything outside needs an explicit reason.)

13. What lawful basis will the build operate under for processing personal data? (You are looking for a sensible answer using the post-2026 framework. The ICO published an updated lawful basis tool in April 2026 that any serious consultancy should be using.)

14. What happens to our data if the model provider has a breach? Who has the contractual obligation to notify us, and how quickly?

15. Will the build's logic be auditable? Can we see, for any decision the automation made, the inputs, the prompt, and the output?

16. What is your own data retention policy for our records during and after the engagement? (You want a clear period, not "indefinitely".)

Exit terms: five questions

These five protect you when the relationship ends. Exit is when consultancies most often let clients down.

17. Who owns the automations after we pay for them? (The right answer is the client. If the consultancy retains ownership of the workflow logic, you are paying for permission to use your own build.)

18. Who owns the data that flows through the automations, and the model outputs the build produces? (The right answer is the client. If the consultancy retains the right to use, train on, or aggregate your data, treat that as a separate commercial and compliance question.)

19. What handover documents do we receive on exit? (You want exported workflow configurations, credentials for any platforms registered in your name, written documentation of the build, and a brief technical handover meeting.)

20. How long do you continue to support the build after the contract ends? (You want a clear, defined window with bug-fix support included. A few weeks is too short. Open-ended is unrealistic. The right answer specifies the period and the support scope in writing.)

21. Can we take the build to another consultancy without re-implementation? (The right answer is yes, with the handover documents from question 19. If the consultancy says no, the build is locked to them and the contract is not in your favour.)

The five questions that matter most

If you only get clear answers to five of the 21, make sure they are these. Each one closes a specific commercial trap.

• Question 1: which tools and why. A consultancy that cannot name and justify the tool stack is not engineering, it is selling.
• Question 7: the fixed scope in writing. Without a written scope you have no contract, just a hopeful conversation.
• Question 17: who owns the automations. Without ownership, you cannot leave.
• Question 18: who owns the data. Without ownership of the data, the consultancy can use your operations to train someone else's tools.
• Question 21: can we take the build elsewhere. This is the test of whether everything else they have said is real or theatre.

If any of those five answers come back vague, keep looking. The other 16 questions matter, but these five are the ones that protect you from the worst of the contract traps.

What to do with the answers

Once the meeting ends, the answers should fit on one page. Three columns: the question, the answer they gave, and a one-word verdict (clear, vague, dodged). In our experience, more than a handful of vague answers across the 21 (or any dodge on one of the top five) is enough to walk away. The first conversation is the most honest a consultancy ever is. If they are not specific now, they will be less specific in the contract.

The questions that earn the most weight in the post-meeting review are the ownership questions. Many consultancies will give clear answers on capability and commercial, then quietly mishandle ownership and exit. The contract clause is usually the place to confirm. Read those clauses carefully.

If you would like a 30-minute conversation about how Reeve Consult would answer all 21, we are happy to do exactly that. Book a free 30-minute slot. We will go through the list with you, give you our answers in writing, and tell you honestly which questions on the list we are stronger on and which we would refer to a specialist.