What does an AI consultant actually do?

A good one works out which parts of your business AI can help with, tells you which parts it cannot, and makes sure the tools you adopt do not quietly leak your client data.

Stopping the leak is the part that matters most, and it is the part nobody asks about.

We work with UK businesses of roughly ten to two hundred and fifty people, mostly construction, engineering and professional services. SMEs, where there is no IT department to hand this to and the managing director has ended up owning it by default.

The work itself is unglamorous. Finding out which AI tools your staff are already using, because they are. Deciding what data is allowed near them. Writing that down on one page. Picking one or two things worth automating and leaving the rest alone. Then watching it when the tools change, which they do every few weeks.


Do you need a consultant, or do you need someone in your own business?

For most businesses this size, you need someone of your own first, and an outside pair of hands second.

Somebody in your team should own AI. Not as a job title, as a responsibility. They try things, they keep a list of what is working, and they are the person staff ask before pasting a contract into a chatbot. Appointing them costs nothing, and until you have, nothing an outside consultant tells you will stick to anything.

Who it should be

Not necessarily the most technical person you have, and not necessarily the youngest.

  • A good understanding of the business. Spotting which problems are worth pointing a tool at is business judgement, not technical skill.
  • Enough technical confidence to set something up and to ask a supplier a hard question without being talked round.
  • Some project management. Things need finishing, not just starting.
  • A process mind. The wins come from noticing the same manual job being done forty times a month.
  • Genuine interest. This matters more than the rest. Somebody who finds it interesting keeps going after the novelty wears off. Somebody who does not will quietly stop.

For thirty people this is not a full-time role. It has to be somebody’s name, that is all.

What an outside consultant adds is what that person cannot do alone: knowing what the tools do with your data, what changed last month, and what is worth ignoring. Your own person knows your business. A consultant should know the tools. Neither is much use without the other.


When you should not hire one

Plenty of businesses asking this question do not need to spend anything, and it is worth saying so plainly.

  • You are under about ten people and hold nothing sensitive. Pick one tool, put it on a company account, agree two rules about what never goes into it, and get on with your work. That is the whole job at your size.
  • Nobody inside is interested. If there is no one willing to own it, an outside adviser produces a document that gets filed. Fix the ownership first or do not start.
  • You want AI without changing how anything is done. The savings come from changing a process. A tool dropped on top of an unchanged process usually adds a step.
  • You only need to answer a client questionnaire. If a customer has asked how you govern AI, what you need is a one page policy and an honest inventory, not a consulting engagement. Both are free to write and the templates are on our AI hub.

If two or more of those describe you, save your money.



Checking what a business is already paying for before buying more AI

What does an AI consultant cost?

There are three ways to buy this and they are priced completely differently. Knowing which one you are being sold matters more than the number.

How you buy itWhat it costsWhat you get
By the dayMore than a month of a managed service, and it ends when the day endsA report. Useful once, and out of date by the next quarter
A monthly managed serviceOurs starts at £100 per monthSomebody accountable month after month, still there when the tools change
Hiring in-house£90,000 to £170,000 a year in salary alone, before National Insurance, pension and recruitment feesA full-time Head of AI

A business of thirty people rarely needs the third and gets very little from the first. Day-rate consultancy suits a specific question with an end date. It is a poor fit for something that changes every few weeks.

What actually moves the price

Whoever you ask, the number will depend on four things far more than on your headcount.

  • How much of it you do yourselves. A business with somebody internal who owns this needs an adviser. A business with nobody needs an operator. Those are different amounts of work.
  • What state your permissions are in. If folder access has grown untidy over ten years, tidying it is the real project and AI is the thing that exposed it. Ask for that to be quoted separately so you can see it.
  • Whether anything is being built. Advice, policy and tool selection are one price. Automating an actual process is another, and it should never be bundled into a monthly figure without being named.
  • Who you answer to. A business bidding for public sector or main contractor work has evidence obligations that a business selling to consumers does not.

Be wary of a single monthly number covering all four with no breakdown. Ask which part you are paying for and what happens to the price when that part is finished.

Our own pricing is set out properly on the AI Manager page rather than repeated here, so there is only ever one place for it to be wrong.

UK, excluding VAT, correct as at August 2026.


How do you choose one?

Ask the awkward questions early. These are worth asking whoever you end up using, and they are just as useful if that is not us.



What will you tell us not to do?

If the answer is nothing, be careful. Most of the value in this work is in what you get talked out of. Plenty of AI projects are worse than the spreadsheet they replace.

A good answer names something specific they have recently advised a client against, and why.


Which of these tools are you reselling?

Anyone paid to recommend a platform will recommend that platform. We sell things too, so the question is fair to us as well.

A good answer is a straight yes with the commercial relationship named, and advice priced separately from products so you can take one without the other.


Who owns what we build, and what happens if we leave?

You should own your policy, your documentation and your configuration.

A good answer is that you keep all of it. If leaving means starting again from nothing, you did not buy advice, you rented a dependency.


Who actually does the work?

The person in the meeting is not always the person on the account afterwards.

A good answer names them, and it is the same name in month six.


Who is accountable when the output is wrong?

AI does not know things. It predicts what text should come next, and sometimes it is confidently wrong. Anything going to a client, into a contract or near a number that matters needs a person to check it.

A good answer agrees who that person is before you start, not after the first mistake.


What happens when the terms change?

They change quietly. Claude began training on user data by default in September 2025 and nothing on anyone’s screen changed to say so.

A good answer explains who is watching for that on your behalf and how you get told.


What does month four look like?

Most of the risk in AI adoption is not in the launch. It is six months later when the tool has changed, the person who championed it has moved on, and nobody has looked at the policy since.

A good answer describes the routine, not the project.


Can we speak to somebody you have done this for?

The ordinary question you would ask of any supplier, and it is worth asking here too.

An honest answer is sometimes no, because this work touches confidential material and not every client will be named. If so, ask for the shape of the engagement instead: how long, who was involved, what changed.


What about our client data?

It depends on the account it was signed into, not on which tool it is.

The same product behaves completely differently depending on how you reach it. A personal ChatGPT account and a company one are not the same thing. Microsoft’s free Copilot and Microsoft 365 Copilot share a name and very little else. That distinction matters more than which brand you pick, and it is the thing most often got wrong.

We keep a dated table of what each of the main tools does with your data, on personal accounts and on company ones, reviewed quarterly. Any consultant you speak to should be able to answer that question for the tools they are proposing, without checking.


The question worth asking yourself first: if a main contractor asked you in writing which third parties handle their information, could you answer honestly today? Most businesses cannot, and that is the real starting point rather than any tool.


What to do next

If you would rather not run all of this yourself, that is what our AI Manager
service is for. One person accountable for how your business uses AI: what is approved, what data
can go in, who checks the output, and what happens when the tools change.

If you would sooner just ask a question first, get in touch. We answer them whether or not you become a client.