Nobody regulates the job title. There is no register, no exam, no professional body you can ring to check somebody out. A person can finish a weekend course on Sunday and put “AI consultant” on their profile by Monday, and thousands have.
LinkedIn ranked artificial intelligence consultant the second fastest-growing job in its 2026 Jobs on the Rise list, behind only AI engineer. That is a lot of new supply arriving in a short window, and none of it comes with a way for you to tell the good from the bad before you pay.
You are not imagining the difficulty. The government’s own AI Adoption Plan for professional and business services names it directly: there is a lack of shared standards and no straightforward way for buyers to identify trusted suppliers, which slows procurement and drains confidence. When the people writing national policy say buyers can’t tell who is any good, you can stop blaming yourself for finding it hard.
So this post is the filter. What you’re actually buying, the questions that sort people quickly, the red flags worth ending a call over, and how to run the choice so you find out before the money goes. If you’re earlier than this and still working out what AI does inside a business your size, read the complete guide first and come back when you’re ready to hire.
First, work out what you’re actually buying
Most bad hires start here, before anyone is even shortlisted. Two very different things get sold under the same title, and buying one when you needed the other is the most expensive mistake available to you.
Advice is somebody telling you what to do. Workshops, opportunity assessments, a prioritised list, a governance framework. The deliverable is a document and the value depends entirely on you having someone able to act on it afterwards.
Delivery is somebody building the thing and leaving it running. The deliverable is software your team logs into on a Tuesday morning without thinking about it.
Plenty of firms sell the first and imply the second. You get a well-argued 40-page assessment, everyone agrees with it, and nothing changes because there was never anybody assigned to build anything. That is the single most common way this money gets wasted, and it’s exactly the pattern behind the pilots that never ship.
Neither is wrong. Advice-only is a reasonable buy if you have an internal team who can build and just need direction. If you don’t have that team, and most businesses under 250 people don’t, buying advice on its own means buying half a solution and paying twice.
Decide which you need before the first call. Then ask every candidate which they sell, and don’t accept “both” without evidence of shipped work.
The seven questions that sort people fast
You do not need to be technical to run this. Every one of these can be asked by an owner with no engineering background, and the answers separate people quickly.
1. What have you built that’s still running, and can I speak to whoever uses it?
The word “still” is doing the work. Anybody can point at a launch. Far fewer can point at something a real team has depended on for a year. Ask for a reference you can actually ring, and ask that person what broke and how it got fixed.
2. Who writes the code, and will I be talking to them?
If the person scoping your work is not the person building it, you’ve bought a translation layer. Every requirement now travels through a project manager, and detail dies in transit. Ask this early, because the answer changes the price and the timeline more than any other factor.
3. What happens in month four?
An AI system is not a bridge. Processes change, staff leave, the model provider updates something, and your data drifts away from what the thing was built on. Ask who maintains it, what that costs, and what the response looks like when it starts giving wrong answers on a Friday afternoon. Gartner found that 45% of organisations with high AI maturity keep AI projects operational for at least three years, against 20% of low-maturity ones. What separates the two is governance and measurement, and both of those are things your supplier either sets up or doesn’t.
4. What does the AI decide on its own, and what goes to a person?
Any decent answer includes a clear list of things the system never decides alone. Refunds, pricing, hiring, anything with a legal or safety edge. If a consultant tells you the agent handles it all, they either haven’t run one in production or they’re hoping you won’t ask again after go-live.
5. Who owns the code and the data when we’re done?
More on this below, because the default answer under UK law probably isn’t the one you assume.
6. What’s the number this is meant to move, and how will we know?
Hours per week. Cost per order processed. Days to respond. If nobody can name a measurable figure before work starts, nobody will be able to prove value afterwards, and the project quietly loses its budget at the next review.
7. What would you tell me not to do?
The best answer to this is specific and slightly against their own interest. Someone who has done this work has a list of things that don’t pay off. Someone who has read about it says everything is a good opportunity.
The red flags
These are the ones worth ending a conversation over.
They start with the technology instead of your process
If the first meeting is about which model they use, which framework they’ve standardised on, or their partnership badges, you’re being sold a stack rather than an outcome. The right first meeting is boring: they ask what your team does all day, who touches what, where things get re-typed, and what happens when someone is off sick.
No fixed price, or a price that needs discovery to produce
Day rates put the risk on you. Every unforeseen problem becomes another invoice, and the incentive runs in the wrong direction the moment the work gets difficult. The UK contractor market makes the numbers concrete: YunoJuno’s 2026 rates report, built on more than 182,000 workforce data points, puts the average developer day rate at £438, with the top 10% of contracts averaging £654. Multiply either by an open-ended number of days and you can see the problem. YunoJuno publishes separate benchmarks for AI consultant contracts if you want a reference point before anybody quotes you.
A day rate isn’t automatically bad. It’s bad when it’s the only thing on offer and nobody will commit to a scope. Watch out too for the paid discovery phase that exists to produce a quote: you’re paying to find out the price.
If you want the wider picture before anyone quotes you, how much it costs to add AI to a small business sets out the five lines on the bill and what UK businesses have actually spent.
They don’t ask about your data
This one is nearly diagnostic. Gartner expects organisations to abandon 60% of AI projects that aren’t supported by AI-ready data through 2026, and found 63% of organisations either lack the right data practices for AI or don’t know whether they have them. Data is what sinks these projects, not model choice. If nobody asks where your records live, how clean they are, how far back they go, or who else has a copy, they have not built one of these before. The demo will work beautifully on sample data and fall over on yours.
Everything is an agent
Agents are genuinely useful for a narrow set of jobs and heavily oversold for everything else. What AI agents actually do sets out which jobs those are, so you can tell when the word is being used properly. Gartner expects over 40% of agentic AI projects to be cancelled by the end of 2027, citing rising costs and unclear business value. A good consultant will tell you when a rule, a form, or an ordinary bit of software beats an agent, and will sometimes tell you the honest answer is a Zapier automation that costs you £30 a month. Build vs buy for AI has the test you can run yourself before the call, so you know which answer you should be hearing.
They won’t show you anything working before you commit the full budget
This is the one that matters most, because it’s the only defence against a plausible pitch. Demos on the vendor’s own tidy sample data prove nothing. Ask to see something running on your data, on one narrow process, before the main money moves. If the answer is that it isn’t possible until the full engagement starts, you’re being asked to buy on trust alone.
Vague answers on who owns what
Under UK copyright law, whoever writes the code is the first owner of it. A contractor keeps ownership by default unless your contract assigns it to you in writing. Plenty of buyers assume that paying for something means owning it, and that assumption is wrong often enough to be worth a clause. Get a written assignment of the code on payment, and get it in the contract rather than in an email. The code is only one of eight things you can own or not own in an AI build, and who owns the code and the data works through the rest of them.
Data is the second half of the same question. Ask plainly whether your data, or anything derived from it, gets used to train or improve anything that isn’t yours. Under UK GDPR you stay accountable for what happens to personal data you hand over, so you need the answer in writing and you need to know whether your supplier is acting purely on your instructions or for purposes of their own. If that distinction gets waved away, that’s the flag. Worth asking the same question of your own house first, because the free accounts your team is already using came with terms nobody read: shadow AI covers what to do about that.
Their case studies have no numbers in them
“Improved efficiency across the client’s operations” means nothing. Hours saved per week, cost per transaction before and after, response time, volume handled without extra headcount: real work leaves numbers behind. If a consultant can’t produce them, either the work didn’t move anything or nobody bothered to measure, and both are reasons to keep looking.
Pressure and scarcity on a first call
Discounted if you sign this month. Two slots left. Prices going up in April. A firm with a full pipeline doesn’t need to rush you, and a decision this size deserves a week of thinking regardless.
What good actually looks like
Positive signals are less discussed than red flags, so here they are plainly.
- They ask about your worst process before they mention any technology
- They can name work they turned down, and why
- They put a number on the current cost of the process before proposing to change it
- They show you something running on your own data early
- They tell you which parts a person will keep deciding
- They give you one price in writing and it holds
- They’re straight about what happens after launch, including what it costs
That third one is worth dwelling on. If nobody has counted what a process costs you today in hours and pounds, nobody can tell you whether fixing it is worth doing. That maths is not hard and you can do it yourself before you speak to anybody: our manual work cost calculator will get you a defensible figure in a few minutes, and the cost of manual work explains what to count. Walking into a sales conversation already holding that number changes the conversation entirely.
How to run the choice
Four steps, and none of them require a procurement department.
Pick the process first, not the supplier. One repeatable, high-volume, low-judgement process with a cost you can state. Which processes to automate first covers how to rank them, and there’s a ranking tool if you’d rather work it through on screen.
Brief three suppliers identically. Same process, same current cost, same question: what would you do, what would it cost, and how quickly could I see something running. Identical briefs are the only way to compare answers rather than sales ability.
Score the answers, not the pitch. Did they change the brief after understanding the process better? Good sign. Did they quote fast without asking anything? Bad sign, whatever the number.
Buy a small thing first. A scoped, fixed-price piece of work that ends in something running beats a large engagement that ends in a document. If it goes well you have a supplier and a working system. If it goes badly you’ve learned that for a fraction of the cost of learning it later. An audit is the usual shape of that small first purchase, so it’s worth knowing what an AI audit is and what you should get from one before anybody quotes you for one.
Where we sit on all of this
We should be straight that we sell this, so read the following as a statement of how we work rather than neutral advice.
Our answers to the seven questions are fixed and public. The person who scopes your work writes the code, so there’s no project manager in the middle. Every price is fixed and agreed in writing before anything starts, with no day rates and no paid discovery phase. The audit runs two weeks and ends with every repeatable process mapped and costed in hours and pounds, ranked by what to do first, plus one working proof of concept running on your own data rather than a tidy sample. The fee comes off a build in full if you start one within 90 days. On a build you see a working prototype inside 60 days with a full refund if you don’t approve it.
The numbers behind that: Founderise took 12 hours a week off its founder and lifted margin 3.5x, with an MVP live in 9 weeks. MidShift has served more than 20,000 professionals and moved people through career progression 92% faster.
Hold every other supplier you speak to against the same list. If one of them answers better than we do, hire them.
The short version
You cannot verify an AI consultant’s credentials, because there aren’t any to verify. What you can do is ask for evidence of shipped, still-running work, insist on one fixed price in writing, refuse to commit the full budget before you’ve seen something working on your own data, get code and data ownership written down, and start with one small process you’ve already costed.
Do those five things and the field narrows fast. Most of the people who arrived in this market last year cannot pass them.
Related reads
- Who owns the code and the data when you hire an AI consultant?
- Shadow AI: what your team is already doing with company data
- AI agents for business: what they actually do
- AI for small business: the complete guide
- What an AI audit is and what you should get from one
- Why most small business AI pilots never ship
- What AI can actually do for a small business (and what it can’t)
- The cost of manual work: what repetitive admin really costs
- How much does it cost to add AI to a small business?
- Which processes to automate first (and how to rank them)
- Build vs buy for AI: when off-the-shelf is genuinely fine
- Build vs buy: when custom software is worth it
- Case study: how MidShift built an AI career guidance engine for 20,000+ professionals
- Case study: how Founderise automated delivery
Still not sure who to trust with this?
Book the call and use it as a test. Bring your worst process and we’ll tell you what we’d do with it, what we wouldn’t touch, and whether you need us at all. If wiring two tools together solves it, we’ll say so and you’ll have lost nothing but half an hour.
Book a free call, or read what the AI audit covers and what you walk away holding.