How to Choose an AI Consultant (and Not Buy a Dependency)

I went looking for what an AI consultant costs in this country, the same way you probably did. Every guide on the first page had a confident table of day rates. Not one said where its numbers came from, and when I lined them up they contradicted each other flatly: one put a large consultancy at a thousand pounds a day, another at four thousand. Same job, same year, same country.

That tells you something useful before we start. The entire first page of Google on this question is written by AI consultancies selling AI consulting, which is a bit like asking a barber whether you need a haircut. I sell this work too, so read me with the same suspicion. The difference I will try to earn over the next thousand words is that I think most of you should be hiring less of it than you are being sold, and I will give you the test that tells you which kind you are being offered.

First, do you actually need an AI consultant?

Probably not for your first automation, and the government’s own numbers back that up. The Department for Science, Innovation and Technology surveyed three and a half thousand UK businesses for its AI adoption research and found the single biggest barrier was not money or vendors. It was that 71 percent had not identified a need.

That is the whole problem in one statistic. If you cannot name the task you want gone, no consultant can find it for you at a price that makes sense, because discovery is the most expensive hour they sell.

Look at what businesses actually do with AI once they adopt it and the point sharpens. Among UK adopters, 85 percent use it for text generation and 70 percent for admin support. Nobody is building a moon rocket. They are drafting the same reply for the ninth time this week and moving figures from one system into another.

So before you shortlist anyone, write down the job. One task, and roughly how many hours a week it eats. If that sounds too simple to be worth doing, notice that 71 percent of the country has not done it. Our AI Opportunity Scorecard is a two minute version of exactly this exercise if you want a starting point.

Why the pricing advice you just read was made up

Because there is no independent UK benchmark for AI consulting rates, and the pages claiming otherwise invented their numbers. I fetched them and checked. One presents its figures as proprietary market data with no external source of any kind. Another describes “one UK survey” without ever naming the survey. Their ranges do not merely differ, they overlap in ways that cannot all be true at once.

I am not saying this to be clever about competitors. I am saying it because a fabricated day rate table is the thing you are most likely to anchor your budget on, and it will anchor you wrong.

Here is a real anchor instead. Look at what the software underneath actually costs to run.

None of that means a quote of several thousand pounds is dishonest. Skilled time is worth paying for, and the software is the cheapest part of any build. But you should be able to see the gap between the running cost and the fee, and hear a straight answer about what fills it. A consultant who cannot explain why their number is their number, in language you follow, has told you what you needed to know.

The one test: can you change it yourself?

Every good AI consultant fails in the same specific way. They make themselves unnecessary.

That is the test. Not the case study wall, and not whether they can say “retrieval augmented generation” without reading it off a slide. Ask whether the engagement ends with you able to change the thing without ringing them.

Notice how few pages will tell you this. Read the guides on page one and you will find “knowledge transfer” listed as a green flag by nearly all of them, always as an aspiration and never as a mechanism. No clause. No artefact. No way to check. It stays a talking point because the people writing it bill retainers, and a client who can edit their own automation is a retainer that quietly ends.

Go back to the government research for a moment. The second biggest barrier UK businesses reported was limited AI skills and knowledge, at 60 percent. Sit with that. The main thing standing between British businesses and this technology is capability, not suppliers. An industry that treats your capability gap as its revenue model is not solving your problem. It is farming it.

How to choose an AI consultant, step by step

Six moves, roughly in order.

1. Write the task down first. One process, and the hours it swallows. Do this before any call, because it is the thing that stops you buying a strategy when you needed a script.
2. Ask for the boring project, not the showcase. Anyone can demo a chatbot. Ask what they built that saved somebody four hours a week and never got a case study, then ask what it cost to run afterwards.
3. Make them name the tools in the first conversation. If you hear “LLMs and automation platforms” instead of Claude Code, n8n, Make or Zapier, you are talking to a salesperson. Vagueness this early is a choice.
4. Ask what happens in month thirteen. Specifically: if you stop paying, what stops working? A good answer is “nothing, you own it”. A bad answer takes a while to arrive and involves the word “partnership”.
5. Buy a small piece first. One scoped build you could pay for out of cash flow, with a defined output. Not a discovery phase that produces a document.
6. Watch it being built. Sit on the screen share. You will learn more about whether this person is worth their fee in forty minutes of watching them work than in any proposal.

That fifth step is most of the risk management. It is also why our AI Opportunity Audit is deliberately a fixed fee session that builds a real automation live rather than a retainer, because I would rather you saw the work happen than read about it later.

The questions that expose a dependency

Ask these four late in the conversation, when the rapport is warm and the answers are less rehearsed.

“Can I log in and change this myself?” Listen for whether the thing lives on a platform you can access or inside code only they can read.

“Whose name are the accounts in?” Yours, on your billing, or you are renting your own automation back from them.

“Who owns the code and the prompts?” Under UK law the contractor keeps the intellectual property unless your contract assigns it to you. This is not a detail to sort at handover. It is a clause to sort before work starts.

“Walk me through a project that failed.” Anyone who has done this work for real has one. A consultant with no documented failures is either new or editing.

So when is an AI consultant genuinely worth paying for?

When the problem is scoped and still hard. That is the honest boundary.

Messy data that needs judgement to untangle. Two systems that were never designed to talk and now must. A process where the rules live in someone’s head and have to be pulled out before anything can be automated. Compliance work where being wrong is expensive. These are real, and they are worth a fee, and no amount of enthusiasm replaces someone who has done it before.

What is not worth paying for is the thing most of the market is actually selling: a retainer that stands between you and software you could operate yourself, priced against a day rate table somebody invented. The gap between those two is the whole game.

Choose the one who wants to leave

I would rather you read this and decided not to hire me. Genuinely. Most of the business owners I meet do not need a consultant so much as they need a fortnight of learning and one honest hour with someone who will tell them which of their ten ideas is worth building.

So choose the consultant who is trying to work themselves out of a job. Ask the month thirteen question and watch their face. Ask who owns the code and see whether the answer arrives fast or slow. The good ones will not flinch at any of it, because the whole point of the work is to leave you with something you can run without them. If you want that conversation with no invoice attached, start here and tell me what the task is. If the answer is that you should go and build it yourself, that is what I will say.

Sixty percent of British businesses say the thing stopping them is that they do not know how. I do not think that is a market to be farmed. It is a gap to be closed, and whoever you end up hiring, me included, should be measured on how fast they close it.