How do I choose an AI consultancy in the UK?

Q: How do I choose an AI consultancy in the UK?

Choose an AI consultancy in the UK by scoring each shortlisted firm on evidence, not the pitch: a fixed scope and fee, a signed assignment of the code to you, UK GDPR data handling, independence from one model vendor, and human approval on consequential actions. M22 Consultancy, for example, prices audits and builds as fixed fees, starts with a diagnostic AI Audit from £1,500 before any build begins, and leaves you owning the code, data and documentation.

Score every shortlisted firm against the same nine criteria, using evidence you have seen rather than what the proposal says. Give each row 0 (no answer), 1 (a promise) or 2 (something you have watched working or can read in the contract). The maximum is 18. A zero on ownership or data location rules a firm out, whatever its total. This is the checklist M22 Consultancy built from the questions a buyer should ask any firm, itself included.

The scored checklist

Criterion	What good looks like	Question to ask	Red flag
Systems in production	A live system you can watch working, with usage and exception figures	"Show me something running today. Who uses it each week?"	Only demos, pilots or logos
Builds and runs what it delivers	The team that built it monitors, fixes and reports on it after launch	"Who is responsible when it breaks in month four?"	Support "by arrangement"; the builders leave at launch
Fixed scope and fee	A written scope, a fixed price and acceptance tests agreed before work starts	"What is in scope, and what happens to the price if it takes longer?"	Open-ended day rates for a build; a price before they have seen your operation
IP and data ownership	A signed assignment of code, documentation, prompts and configuration to you	"Does the contract assign copyright to us in writing?"	"You get a licence to use it"
UK GDPR and data location	Named sub-processors, stated regions and a written processing contract	"Where is our data processed and stored, and by which companies?"	Cannot name the model provider or the region
Model-agnostic	Models called behind a thin layer, so the provider can change	"What would it cost to move this to another model provider?"	One vendor's platform throughout, or an undisclosed resale or referral margin
Human approval	A person approves anything that reaches customers, money or official records	"Which actions can the system take without a person's yes?"	"Fully autonomous" sold as the benefit
Handover and exit	Code, accounts and keys in your name; a runbook; a stated notice period	"If we stop the retainer, what do we receive and how quickly?"	The system only runs inside their accounts

Two rows need more than a yes. On ownership, UK copyright law gives the code to whoever wrote it, or their employer, unless a signed, written assignment says otherwise: see who owns software built by an AI consultancy. On data, Article 28 of the UK GDPR requires a written contract with any processor, including deleting or returning your data when the service ends, and sending personal data to an organisation outside the UK is usually a restricted transfer that needs adequacy regulations, a safeguard such as the IDTA, or an exception (ICO). More on both in GDPR and AI for UK businesses and choosing a model provider. On ownership the position is simple: clients own the code, data and documentation that are built. On data, M22 is UK GDPR-native.

Check the firm in ten minutes

Companies House. Search the free Companies House register for the legal entity named on the proposal. It shows the incorporation date, registered address, officers and filing history. Check the name on the contract matches.
How long trading. A company in its first year has no accounts on file. That is not disqualifying on its own, but it moves the weight onto the people: ask what the directors have built and run before.
Directors. Open each director's other appointments and search the disqualified directors register. Late filings, or dissolved companies behind the same names, deserve a question.
Independent reviews. Clutch publishes reviews it could not verify as "Not Verified", and lists sponsors above non-sponsors, so position there is not a ranking. On Google, read the text: a real review names a project, a problem and a result.

How to read "best AI consultancies in the UK" lists

When we checked the top search results for "best AI consultancies UK" in September 2026, three of the ranked lists were published by consultancies that appear in them: one placed itself first, one second, and one disclosed its interest clearly. M22's own list of UK AI consultancies is a fourth, and it says who wrote it at the foot of the page. Use lists like these to build a long list, not to make the decision.

Find the author: the domain, the byline and the footer.
Look for a disclosure that the author appears in the list.
Check claims about other firms link to a source you can open.

A buying sequence that limits risk

Shortlist three firms from referrals, directories and lists, and score them on the table above.
Discovery call. Thirty minutes, free, the table's questions in order. A firm that quotes a build on this call fails the fixed-scope row; M22 uses it to find out whether an audit is the right next step at all.
Paid audit. A fixed fee, typically £1,500 to £15,000 in the UK depending on size, a market range from M22's cost guide. M22's own audit starts from £1,500 with no set length: it is sized to the business and follows the real work rather than a workshop. You keep the process map and costed plan, with no obligation to buy the build.
Fixed-fee build. Scoped from the audit, with acceptance tests, a copyright assignment and processing terms in the contract. Focused builds typically run £5,000 to £25,000, a market range from the same cost guide.
Run. A monthly retainer, commonly £1,000 to £5,000, measured against the audit's numbers, with a notice period you can live with.

If your problem is one bounded workflow, an automation agency may be enough. If you are a small business, the SME guide covers budgets and first projects.

How does M22 handle this?

M22 Consultancy is built to score well on its own checklist. The AI Audit is a fixed fee from £1,500 agreed before day one, with no set length: M22 sizes it to the business and follows the real work rather than running a generic workshop. The client keeps a process map, ranked opportunities and a costed build plan either way. If a build follows, it is a fixed fee too, and the client owns the code, data and documentation, not M22. M22 builds and runs what it delivers, on a monthly retainer that measures adoption and exceptions, and a person approves anything that touches customers, money or records. It is model-agnostic, building on whichever model fits, such as Claude, ChatGPT or Microsoft Copilot. See the kind of system this produces on the work page rather than take a pitch on trust.

Run any firm on your shortlist through the table above before a proposal reaches your desk. To score M22 against it directly, book a thirty-minute call at m22.group/contact.

Questions people also ask

How many AI consultancies should I talk to before choosing?

Three is usually enough. Score each on the same criteria after a thirty-minute discovery call, then pay only one for an audit. Talking to more firms rarely changes the answer and costs your team time; talking to one leaves you nothing to compare the scope and fee against.

Should I pay for an AI audit before commissioning a build?

Usually yes. A fixed-fee audit, typically £1,500 to £15,000 in the UK, prices each opportunity before you commit to a build and can conclude that something is not worth building. M22's own AI Audit starts from £1,500 and carries no obligation to buy the build. Make sure you keep the output whoever runs it.

How much does M22's AI audit cost?

From £1,500, a fixed fee agreed before day one. There is no set length: M22 sizes the audit to the business and follows the real work rather than running to a fixed number of days. You keep the process map, the ranked opportunities and the costed build plan either way.

Who owns what M22 builds?

The client. Clients own the code, data and documentation, so ownership sits with the business rather than the consultancy. Ask any firm on your shortlist to put ownership in writing before you sign.

Are online rankings of the best AI consultancies in the UK reliable?

Treat them as a starting list. Several of the most visible UK rankings are written by consultancies that include themselves, sometimes ranked first without a disclosure. Check who published the list, whether the criteria are stated, and whether claims about each firm link to a source you can verify.

Sources

Written and maintained by M22, a London-based AI consultancy. Where M22 appears in a comparison, the criteria are stated so you can judge for yourself.