What an AI consultant actually does (for UK SMEs)

If you run a small or mid-sized business in the UK and you have been thinking about hiring an AI consultant, you have probably noticed the same thing I did before I started doing this work: the job description is all over the place.

One website tells you an AI consultant builds custom machine learning models. The next says they run ChatGPT training workshops. A third promises a "digital transformation roadmap" for 40,000 pounds and three months of slide decks. None of these are wrong, exactly, but none of them are particularly useful either if you just want to know whether AI can save your team ten hours a week.

So here is a plain explanation of what an AI consultant UK side actually does for an SME, what they should not do, what it costs, and how to spot a good one.

What an AI consultant actually does

The honest answer is that an AI consultant is part business analyst, part project manager, part technical translator, and part change manager. The technical bit is often the smallest part of the job.

A typical engagement covers five things.

Discovery. A few days spent understanding how your business actually works. Not how the org chart says it works. Where does time get burned. Where do mistakes happen. Where is the team frustrated. Which processes have grown organically over ten years and nobody has revisited. Most of the value in any AI project gets locked in here, before any technology is chosen.

Opportunity mapping. Taking what discovery found and ranking the possibilities. Each opportunity gets a rough estimate for effort, cost, time to value, and risk. Some get parked because the return is too small. Some get parked because the data is not there yet. One or two get picked as candidates for a pilot. A good map is usually two pages, not fifty.

Pilot scoping. Designing a small, reversible experiment for the strongest candidate. The scope should be tight enough that you can see results in 4 to 8 weeks, and the success criteria should be specific and measurable. "Reduce quote turnaround from 3 days to 1 day" is a pilot. "Improve customer experience" is not.

Vendor and tool selection. There are now thousands of AI tools, and most of them will not suit your business. A consultant who has done this a few times will know which off-the-shelf products are mature, which need heavy customisation, and which are essentially marketing. They should be vendor-agnostic, meaning they are not collecting referral fees from whatever they recommend.

Implementation oversight and change management. Getting the thing built and, more importantly, getting people to use it. This is where most pilots quietly die. The tool works, the team ignores it, the project gets shelved. A consultant earns their fee in this phase by training the team, handling objections, adjusting the workflow when reality bites, and making sure the new process becomes the actual process.

For more on what that looks like in practice, our post on what a good AI pilot looks like walks through a real example.

What an AI consultant should not do

Equally important is what the job is not. If a prospective consultant pitches you any of the following, treat it as a red flag.

Just reselling ChatGPT licences. ChatGPT, Claude, Gemini and Copilot are useful tools. They are not a strategy. If the entire engagement is "we will set you up with team licences and run a workshop", you are paying consultancy fees for something you could do yourself in an afternoon.

Building huge custom models from scratch. For 95% of SMEs, this is wildly unnecessary. The interesting work for small businesses is almost always combining existing models with your existing data and existing workflows. Custom model training is a six-figure project that takes six months and rarely beats the off-the-shelf option.

Drop in, drop a deck, disappear. A 60-slide strategy document with no follow-up is the classic consultancy trap. If there is no implementation phase in the proposal, you are buying a report, not an outcome. Reports are fine if a report is what you want, but be clear that is what you are paying for.

Charging enterprise rates for SME work. A Big Four firm will happily quote 100,000 pounds for what should be an 8,000 pound pilot. The methodology is the same, you are just paying for the brand and the office. For more on this, see our post on how to spot an AI vendor wasting your money.

The UK-specific bit

There is a reason a UK-based AI consultant tends to suit a UK SME better than an offshore generalist, and it is not flag-waving. It is compliance.

Any AI system that processes personal data about UK residents falls under UK GDPR and the Data Protection Act 2018. The Information Commissioner's Office publishes detailed guidance on AI and data protection that covers lawful basis, fairness, transparency, automated decision-making, and Data Protection Impact Assessments. These rules are not the same as the equivalents in the US, the EU AI Act, or any other jurisdiction, and getting them wrong can result in real fines and, more commonly, real reputational damage.

A consultant who knows this material can build it into the design from day one. A consultant who does not will either ignore it (your problem when something goes wrong) or panic about it (your project gets stuck in legal review for six months).

The UK SME landscape itself matters too. According to the Federation of Small Businesses, there are around 5.5 million small businesses in the UK, accounting for over 99% of all private sector firms. The vast majority of them have fewer than 50 employees, operate on tight margins, and cannot afford a failed technology project. The right consultant designs around that reality.

How engagements are typically structured

Most sensible AI consultancy UK engagements follow a three-phase shape, and the budgets are reasonably predictable.

Discovery: 1 to 2 weeks, £500 to £2,000. A handful of interviews, a process walkthrough, and a written summary of the opportunities. This phase should always exist, even if it is short. Skipping it is how consultants end up building the wrong thing.

Pilot: 4 to 8 weeks, £2,000 to £8,000. One opportunity, taken from concept to a working solution that handles real volume with real users. Includes implementation, training, and a measurement period. If it works, you keep it. If it does not, you have spent less than a single hire to find out, and you have learned something useful.

Rollout: 3 to 6 months, varies. Scaling what worked across more of the business, integrating with other systems, building governance and review processes. Cost depends entirely on scope, but should always be priced against a clear expected return.

If a proposal lumps all three into one big fixed-fee project before discovery has happened, push back. The whole point of starting small is that you reserve the right to stop if the early signals are bad.

How to pick a good one

If you are interviewing AI consultants for small business work, ask six questions. The answers will tell you more than any case study.

1. What does your typical first engagement look like? A good answer is specific. A bad answer is "it depends entirely on your needs."

2. Can you show me two case studies in businesses my size? If the only references are FTSE 100 logos, this is not the right fit. SME work is genuinely different.

3. What happens if the pilot does not produce the result we hoped for? Look for someone who has clearly thought about this. "We adjust and retry, or we stop" is a healthy answer.

4. Who actually does the work, you or a junior? Many consultancies sell with their senior people and deliver with their juniors. Worth knowing upfront.

5. How do you handle UK GDPR, ICO guidance, and the Data Protection Act? If the response is hand-wavy, that is your answer.

6. What does success look like, in numbers, and will you put it in the proposal? A good consultant will. A vague one will not.

Three other quick signals. They should be willing to say no to work that is not a fit. They should be cautious about overpromising on speed. And they should be more interested in your business than in their own methodology.

One more thing worth checking: do they have an opinion. The best AI consultants I know have firm views on what works and what does not, and they will tell you when they disagree with something you want to do. A consultant who agrees with every idea you float is either trying to win the work or has not done enough of it to know the difference. You are not paying for a yes-person. You are paying for someone who has seen ten of these projects and can tell you which of your three ideas is the one most likely to pay off, and why.

A word on timing

A lot of UK SMEs ask whether now is the right moment. The honest answer is that the technology has been good enough to deliver real value to small businesses for about two years, and the cost of using it has fallen significantly in the last twelve months. The tools that were experimental in 2024 are now boring infrastructure. That is usually the right time to adopt something: not at the bleeding edge, not after everyone else has finished, but in the steady middle where the kinks are worked out and the price has come down.

What has not changed is the failure rate of projects that skip discovery, skip pilots, and try to do everything at once. Those still fail at roughly the same rate they always have. Good consulting is mostly about not making those mistakes on your behalf.

Our services page and about page lay out how we work, if you want a concrete example of what good answers to these questions look like. And if you are still not sure whether you are ready for any of this, the post on three signs your business is ready for AI is a useful gut-check.

A sensible next step

If you have read this far, you are probably the kind of UK SME owner who wants to make an informed decision rather than a panicked one. That is the right instinct.

The cheapest way to find out whether an AI consultant could earn back their fee in your business is to get an outside view of where the opportunities actually are, before you commit to anything. That is what our free AI opportunity report does. You tell us about your business in a short form, and we send you a personalised analysis of where AI could realistically help, what it would likely cost, and what the returns might look like. No sales call, no obligation, and written in plain English for business owners rather than technologists.

Get your free AI opportunity report here and find out what an AI consultant would actually do for your business before you hire one.