AI consultant: what they do, when you need one and how to choose

What does an AI consultant actually do?

The title covers at least four different professions, and the people in them rarely do each other's jobs well. Most UK firms that go looking end up talking to all four without realising it, because they describe themselves in near-identical language. Knowing which one you need is half the decision.

They also start in different places. An adoption consultant starts with your people and the tools you already pay for, whether that is Microsoft Copilot, ChatGPT, Claude or Gemini (we compare the main options in our guide to AI tools for business). The other three start with a system, a dataset or a document. Sorted by what each one leaves behind:

+

Adoption consultant

Gets existing teams using AI in their actual work. Works on habits, hands-on training, the rules for data and tools, and who owns the topic internally. This is what most companies of 20 to 200 people need, and the kind they ask for least.

+

Automation consultant

Builds workflows that remove repetitive work: reports that compile themselves, inbound enquiries sorted before anyone reads them, case files prepared overnight. Concrete and measurable, but it only fixes the processes it was built for.

+

Machine learning or data consultant

Builds custom models on your data. That needs the data, the volume and a problem that genuinely calls for a bespoke model. Far fewer companies need this than believe they do, because general-purpose models now cover much of what once required one.

+

Strategy consultant

Produces assessments, maturity scores and roadmaps. Useful when the leadership team genuinely disagrees about direction, but by definition it hands you the plan for a change and leaves the change to you.

When should you hire an AI consultant, and when shouldn't you?

Hire one when you know AI should make a difference but cannot say where to start; you have bought licences and nothing has stuck; a large group needs to move at once and nobody has time to learn it first; or nobody internally owns the question. The last is the most common situation in UK firms of 20 to 200 people. The managing director cares, IT is two people keeping the lights on, and AI falls in the gap between them.

Do not hire one when someone inside the business is already driving adoption well (back that person instead of parachuting in an outsider), when what you really want is a report to show the board, or when you are looking for someone to “do AI for us” while everyone carries on working exactly as before. That last brief fails whoever you hire.

The UK numbers are worth knowing here. In Anthropic's Economic Index for May 2026, the UK scored 3.35 on the AI Usage Index (its share of Claude usage relative to its share of the working-age population, where 1.0 is proportional), 16th of 121 countries. Yet 39.0% of that UK usage was work, against 43.4% globally, and personal use was above the global average (43.5% against 40.2%). Augmentation, a person collaborating with the AI, made up 55.2% of UK usage against 51.4% globally. It is a snapshot of one product, but the shape is familiar: plenty of people use AI on their own initiative, far fewer use it on the work they are paid for, and most of that use still involves a person working alongside the tool. Closing that gap is what an adoption consultant is for.

One pattern is worth recognising before you hire anyone. The companies that make AI stick almost always have a named person inside who owns it. It is rarely a full-time role and it does not have to sit in IT; as we argue in why the companies winning with AI are not the most technical ones, ownership and habit count for more than technical depth. A good consultant builds that role up instead of standing in for it. If the leadership team is still arguing about direction, read what an AI strategy for a business should contain first.

AI consultant vs AI agency vs AI consultancy: what is the difference?

Suppliers use the three terms almost interchangeably, but they signal different things. An AI agency usually means a team that produces work for you: they build something, you receive it. An AI consultancy usually means a firm with several consultants and a method, often with a discovery or strategy phase at the front. An AI consultant, whether an individual or a small team working that way, suggests someone who works with your organisation rather than for it.

So the useful question is not which word the supplier uses but what you want to have when the engagement ends: a delivered product, or an organisation that works differently. Too many AI consulting engagements are sold as the second and delivered as the first, for structural reasons we set out in why the AI consulting industry is broken.

+

Agency

Best for a defined build with a clear specification, such as an integration, an internal tool or a customer-facing assistant. Ask who maintains it after handover and what happens when the underlying model changes.

+

Consultancy

Best for a formal programme across several functions, where you need breadth, governance and a steering group. Check how much of the engagement goes on discovery and how much on people changing how they work.

+

Consultant

Best when the goal is capability inside your own team, quickly. Check that the person who sells the work is the person who delivers it, and that they have done it inside a real organisation.

How to choose: questions that show who can deliver

Cost is usually the first thing buyers ask about, and on its own the least revealing. What drives it, format by format, is covered in how much an AI consultant costs and what you get for it.

The market for AI consulting grew faster than the expertise to supply it, and plenty of providers have a polished deck with very little delivery behind it. These six questions sort them quickly, and none of them requires technical knowledge to ask.

+

“What have you deployed yourselves?”

The difference between having read about AI adoption and having pushed it through in a real organisation shows within minutes. Ask for an example where it went wrong and what they did about it.

+

“What will we have when you leave?”

A good answer describes capability: who can do what, which ways of working changed, who owns the topic internally. A weak answer lists deliverables.

+

“How will we know it worked?”

Satisfaction scores collected at the end of a session measure very little. Ask what they check four to six weeks later, and how.

+

“What do you do when the team doesn't use it?”

The best single signal of experience. Anyone who has really done this knows it happens, and has an answer more specific than “more training”.

+

“Who actually does the work?”

In larger consultancies the person who wins the work is often not the person in the room. Ask directly, and meet them before you sign.

+

“How do you handle our data?”

They should be able to say which tools and account types are suitable for client or employee data, and how that fits with your UK GDPR obligations. A vague answer here predicts vague governance later.

What should you be left with when the engagement ends?

Judge any engagement by the capability it leaves behind. If all you receive is a report or a finished build, expect the effect to fade within a couple of months, because nobody inside the business was changed by it. Write the end state into the brief before anyone starts, and ask every supplier how they will get you there.

The direction worth aiming for is what we call AI-native: AI built into how ordinary work gets done, rather than a side project for a few keen people. We describe what that looks like in practice in what AI-native means for a 50-person company. No single engagement gets you all the way, but by the end you should be able to point to five things:

+

People who use it on real work

Staff applying the tools to their own tasks every week, well beyond the enthusiasts who would have done it anyway.

+

Processes that changed

Specific pieces of work now done a new way: the weekly client report, the first draft of a tender response, the triage of inbound enquiries.

+

Rules people know

Short, written guidance on which tools are approved, what data can go into them and who to ask when unsure. Short enough that people read it.

+

A named internal owner

Someone inside who keeps it going, spots the next use case and is the first port of call for colleagues.

+

A way to measure it

A handful of signals you can check without the consultant: hours returned, adoption by team, use cases that went live and stayed live.

Hiring an AI consultant in the UK: data, regulation and location

For most UK firms the regulatory side is manageable, but it belongs in the first week of the engagement, where it shapes everything that follows. UK GDPR applies whenever personal data goes into an AI tool, and the Information Commissioner's Office (ICO) publishes guidance on AI and data protection that any consultant working with UK clients should know. If you sell AI-enabled products or services into the EU, the EU AI Act may apply to you as well, even though the UK sits outside the EU. None of this is legal advice.

In practice the most useful output is a short set of rules agreed before people start pasting client files into chatbots: which tools are approved, which data stays out, who signs off on new uses. A recruitment firm handling CVs or an accountancy practice handling client records needs this more urgently than most, and agreeing it should take days, not a quarter.

Geography matters less than it used to. The part that benefits from being in the same room is the first session, where a whole team shifts at once and the sceptics see their own work done differently; ongoing support works well remotely. We run sessions on-site across the UK and remotely for distributed teams (more in how we work with UK companies of 20 to 200 people), and in the same way with Nordic companies from our Stockholm and Gothenburg offices.

How we work: hands-on first, then ownership

We almost always start with Deployed Kickstart, a half-day session where the whole team builds something for their own role on their own real work, and the ground rules are agreed in the same room. It gives everyone a shared starting point and means adoption does not hinge on a few enthusiasts. What a session like that should include is covered in our guide to AI training for business.

Where the direction is unclear, an AI Audit & Discovery runs over a few weeks: a survey, interviews and a leadership workshop, with every use case checked against data, process, technology, people and governance before anything is built. For organisations that want the change to last, we continue as an AI Partnership, the embedded or fractional AI lead most companies need but do not have in-house. The aim throughout is to build up your own capability so that you need us less.

The full sequence is on our advisory page. It is one version of the hands-on model described in this guide, and the questions above apply to us as much as to anyone.

Frequently asked questions