How to Choose an AI Consultancy in the UK

If you search for an AI consultancy in the UK you will get thousands of results, and a striking proportion of them look identical: the same stock imagery, the same promise of transformation, the same absence of anything you could actually verify. This is a practical guide to telling them apart, written by a firm that is obviously one of the options and is therefore not neutral. Read it with that in mind.

First, work out what you are actually buying

The word consultancy covers at least four different products, and confusing them is the most common procurement mistake we see.

• Advisory. Strategy, roadmaps, use case identification, operating model design. Produces documents and decisions. Valuable when the organisation genuinely does not know where to start.
• Delivery. Building and deploying working software. Produces systems. Valuable when you know what you want and need it to exist.
• Staff augmentation. Renting engineers who work under your management. Valuable when you have the leadership and the process but not the headcount.
• Managed service. Someone else runs the thing forever. Valuable when you have no intention of building internal capability.

Many firms sell one and describe it as all four. A useful question early in a conversation is simply: at the end of this engagement, do I have a document or a system? Both are legitimate answers. Only one of them logs in.

Questions that separate real delivery from theatre

Who exactly will do the work?

Ask for the names and ask whether you will meet them before signing. The classic consulting failure is that the people in the pitch are not the people on the project. In a small studio this is easy to answer. In a large firm it may still be answerable, but you have to ask.

Can I see something you built that is still running?

Not a case study PDF. A system, ideally with a client reference who will take a call. Anyone can write a case study. Very few AI projects from the last three years are still in production, which is exactly why this question is diagnostic.

What happens to the code?

You want a clear answer that the source code, infrastructure configuration and documentation are yours, in your repositories, with no licence required to keep using them. Some firms genuinely do build reusable platform components and licence them, which is fine if it is stated upfront and priced separately. What you do not want is discovering at handover that the useful part is not transferable.

How will we know if it worked?

A serious partner will push you towards a measurable definition of success before the build, and will be visibly comfortable with the possibility of being measured against it. Vagueness here is a tell.

What would make you tell me not to do this?

The most useful question on the list. Everyone has a version of the answer prepared, but the specificity of it tells you a great deal. A firm that has genuinely walked away from work can describe when and why. A firm that has never said no to a project will not say no to yours either, including when it should.

On price

UK day rates for AI work vary enormously, roughly from six hundred pounds to well over two thousand, and the correlation with outcome quality is weaker than you would hope. What matters more than the rate is the shape of the commercial arrangement.

Fixed-fee discovery with a fixed deliverable protects you at the point of maximum uncertainty. Milestone pricing with a written definition of done protects you during the build. A monthly arrangement you can exit at the end of any month protects you afterwards. Open-ended time and materials with a vague statement of work protects nobody except the supplier.

Be suspicious of very low quotes for production systems specifically. The production requirements described in our piece on why AI pilots fail do not become optional because the quote was cheap. They just get discovered later, as change requests.

Size: large firm or small studio

Large firms bring procurement familiarity, insurance and indemnity limits that satisfy risk committees, sector benchmarking, and the ability to surge capacity. They cost more, move slower, and the seniority of the people actually typing tends to be lower than the pitch implied.

Small studios bring the people you met, faster decisions and lower cost. They carry key person risk, may not clear procurement thresholds for large public bodies, and cannot absorb a sudden tripling of scope.

Neither is correct in general. If your project is a two hundred person programme across five business units, do not hire a studio of eight people. If your project is one well-defined system that needs to exist by autumn, a large firm is an expensive way to get it.

Sector and regulatory fit

If you are in financial services, health, legal or the public sector, the governance requirements will shape the build more than the technology will. Ask specifically what the partner has done about data protection impact assessments, model documentation, human oversight design and audit logging. If the answer is that they would bring in a specialist, that is honest but it needs pricing. Our own approach is set out under AI governance and compliance.

A reasonable process

Speak to three or four firms of deliberately different types. Give them all the same short brief. Ask each the five questions above. Pay one of them a small fixed fee for a discovery before committing to anything larger. The discovery is the cheapest audition available and you keep the output regardless.

Where we fit

We are a small UK studio that does delivery, based in Glasgow and working across the country including London, Manchester, Birmingham and Edinburgh. We are the right choice for well-defined systems that need to reach production, and the wrong choice for large multi-workstream transformation programmes. If that sounds like your problem, book a chat.

Related reading

• The Dashboard Trap
• The Data Puddle That Works
• Machine Learning Beyond the Chatbot
• Your Data is Scattered Across a Dozen Systems

How we can help

AI Applied is a UK software studio building production AI systems. See our AI services, including AI readiness assessment, AI implementation, machine learning development, data engineering, AI automation and AI governance and compliance. Or just get in touch.