How to choose an AI adoption partner.
Written by a specialist, which means you should read it with that in mind. It names the five types of provider you can buy from, the ten questions that separate them, and the four situations where we are the wrong answer. If a guide never tells you when not to hire the author, it is a brochure.
By Darren Boyle, founder of The AI Adoption Agency. Written from UK engagements, not from research. Last reviewed 29 July 2026.
Why is this choice harder than it should be?
Because almost every provider in the market is selling the same sentence, and the sentence is about the technology. You are not short of AI capability. Your business almost certainly already pays for more of it than it uses. What you are short of is people using it, and that is a different purchase entirely.
The research is consistent on this point, and so is the pattern inside the businesses we walk into. Licences are bought, a session is run, everyone is positive on the day, and eight weeks later three people use it properly and the rest have gone back to how they worked before. Nobody announces that. It just happens, quietly, and the return never appears.
So the question is not who can teach your team about AI. Several hundred UK firms can do that competently. The question is who is accountable for your team still working differently in six months. That narrows the field considerably, and the rest of this guide is about how to narrow it yourself.
What types of AI adoption provider can a UK business buy from?
Five, broadly. Each is genuinely the right answer for somebody. Knowing which one you are is most of the decision.
1. Large consultancies
Deloitte, PwC, EY, KPMG and the larger technology consultancies. Real depth, real methodology, and a name your board recognises.
Right for you if you are multi-site, running a transformation programme with a budget to match, and need the assurance a recognised name provides.
Watch for the gap between the partner who sells and the team who delivers, and a scope that grows.
2. Microsoft or Google partners
Resellers and implementation partners whose commercial model sits on one vendor stack. Strong on licensing, configuration and integration.
Right for you if you are committed to that stack and the problem is technical setup.
Watch for the recommendation always landing on the product they resell. Ask what they earn on the licence.
3. Independent AI trainers
A large and growing field, from experienced practitioners to people who read fast and started last year. Low outlay, quick to book.
Right for you if you need awareness across a lot of people and are not yet trying to change how work gets done.
Watch for the one-day session with no follow-up, which is the format most likely to produce nothing at all.
4. The platform vendors themselves
Anthropic, Microsoft, OpenAI and Google all publish enablement material and run their own sessions. It is accurate, current and free.
Right for you if you have capable, motivated individuals who will self-serve. Individuals reliably do. Whole teams reliably do not, which is why this rarely spreads past the enthusiasts.
Watch for the fact that it is written for everybody, so it is about the product rather than about your firm's month-end, your compliance position or your nervous employees.
5. Adoption specialists
Firms that go deep on one platform and are accountable for behaviour change rather than attendance. We are one of these.
Right for you if the problem is that people are not using what you have, or you are starting from nothing and want the deployment and the behaviour handled as one job rather than two.
Watch for the depth being real. Ask what they will not train you on. If the answer is nothing, they are a generalist wearing a specialist's language.
The pattern underneath
Types one and two are strongest when the problem is technical or structural. Types three and four are strongest when the problem is awareness. Type five is strongest when the people are the blocker, whether the technology is already in place or still to be set up.
Most UK firms who contact us are in that third situation. Some have already bought from one of the first four and want the return that never arrived. Others have bought nothing yet and would rather set it up correctly the first time than undo it later.
What ten questions should you ask any AI adoption partner?
Ask all ten of everyone on your shortlist, including us. The differences show up fast, and questions eight, nine and ten tell you more than the other seven together.
1. What will you measure, and when?
A good answer names numbers and dates: hours returned per head per week, how many people are still using the tool at week twelve, how many tasks now run without a person. A weak answer talks about satisfaction scores and attendance. Attendance is not adoption. Everybody attends.
2. How do you handle employees who think this is about replacing them?
The single most useful question on the list, because it is the real reason rollouts stall and most providers have no answer beyond reassurance. Listen for whether they raise it before you do, whether they have a method, and whether they will run that conversation in the room with your people or leave it to you.
3. Which platform do you specialise in, and why that one?
"Whichever you use" sounds accommodating and is a warning. Breadth across many tools produces awareness. Depth on one produces capability. A specialist should be able to argue their choice in detail, name what it is bad at, and tell you plainly when a different platform suits your workflow better.
4. What happens after the training ends?
Behaviour needs a second and third contact to hold. If the engagement finishes on the last day of delivery, you have bought an event. Ask what the follow-up looks like, who runs it, and what triggers a return visit. A provider whose model depends on never coming back is not carrying any of the risk.
5. Who owns governance, and is it in this engagement?
The use policy, the boundaries on what AI may touch, the approval points for anything leaving the business, and the record of who signed off what. If this is a separate proposal, ask why. The legislation shows the direction of travel clearly, and firms that can evidence their approvals will find the next two years easier than firms that cannot.
6. How do you price this, and against what?
A fee compared to other fees tells you nothing. A fee compared to the return tells you everything. The arithmetic is simple: hours returned per head per week, multiplied by the number of heads, multiplied by your internal hourly value. Any provider unwilling to be judged against that number is asking you to buy on faith.
7. Who actually delivers, and have they done the job my team does?
Ask to meet the person who will be in the room. In larger firms the seller and the deliverer are frequently different people with different depth. Your finance team will know within ten minutes whether the trainer has ever run a month-end, and their attention follows that judgement.
8. What will you not do?
A specialist has a clear edge to their scope and can state it without hesitating. A generalist gives a vague answer because saying no has never been part of the model. This question separates the two faster than any credential check.
9. Which of our workflows would you leave alone?
The best question on the list, and almost nobody asks it. Some work should stay human: the judgement calls, the sensitive conversations, the things carrying professional liability. A partner who wants to automate everything you show them has not understood your business and will damage trust with your team on day one.
10. Will my team take to the person delivering this?
The least businesslike question on the list and one of the most predictive. Question seven asks whether they know your job. This one asks whether your people will let them near it. Knowledge is the entry requirement, not the difference between a programme that lands and one that does not. Meet the person who will be in the room, then ask yourself whether your team would want three months of them. If the answer is no, the material will not save it.
Which answers should worry you?
"We can train you on any tool." Optimised for winning the work, not for your team using anything afterwards.
A proposal built around a number of training days. Days are an input. You are buying an outcome. If the engagement is priced and described in days, that is what you will get and no more.
No mention of employee resistance anywhere in the proposal. Either they have not met the problem, or they have decided not to mention the hardest part before you sign.
Governance sold separately, later. Boundaries are straightforward to set before adoption and painful to set after an incident.
Case studies with percentages and no arithmetic. "Forty percent more efficient" means nothing without the baseline. Ask how it was measured. A real answer exists or it does not.
Urgency about falling behind. Competitive advantage is a real reason to move. Fear is a sales technique. You can tell them apart by whether the provider can explain, specifically, what your competitors are doing that you are not.
When is a specialist like us the wrong choice?
When you need software built, not adopted. If the requirement is a custom system or a bespoke integration, you need a development firm. We work with what exists and make people use it well. That is a different discipline. Worth knowing that the two are sequential rather than exclusive: the build finishes, and then somebody still has to get the team using it, which is the point most projects underestimate.
When the requirement is licensing and configuration, not behaviour change. Tenant setup, licence assignment, security baselines and integration work sit better with a Microsoft or Google partner. Being committed to Microsoft is not the dividing line, and we should be clear about that: Claude is generally available in Microsoft Foundry and in Microsoft 365 Copilot, and it works across Word, Excel, PowerPoint and Outlook, so a Microsoft-committed firm is one of the easier deployments we do rather than a conflict. The dividing line is whether your problem is technical or human.
When you need a hundred people made aware, not a team changed. Broad awareness sessions are a legitimate thing to want, and plenty of good independent trainers deliver them well. Do not pay adoption-programme fees for an awareness outcome. If awareness is the first step and changed behaviour is the destination, say that out loud, because the order you buy them in decides whether the second one works.
When leadership has not decided this matters. The one we turn down. Adoption follows attention, and attention starts at the top. If the MD is delegating this because it is on a list rather than because it is a priority, no provider will produce a return, including us. Fix that first and the rest becomes straightforward.
We would rather tell you this now than take a fee and produce nothing. It also happens to be the reason our own selection process is short.
What does good look like ninety days after you sign?
Set these expectations with whoever you choose, in writing, before work starts. They are reasonable, they are measurable, and a provider who resists them is telling you something useful.
Most of the team, not the enthusiasts, using the tool weekly without prompting. Named tasks that used to take a person and now do not. Hours returned per head that somebody has actually counted rather than estimated. A written use policy that your people can name the contents of. At least one person in the business who has become the internal expert and no longer needs you. And a clear answer to what those returned hours were spent on, because capacity that is returned and not redeployed is capacity that quietly disappears.
That last point is where most programmes fail on the business case despite succeeding with the individuals. People genuinely save the time. Nobody decides what the time is for. Six months later the firm feels no different and the return cannot be found, because it was never directed anywhere.
What buyers ask us at this stage.
How much should AI adoption training be in the UK?
Judge the fee against the return rather than against a training budget, because they are different purchases. The arithmetic is hours returned per head per week, multiplied by team size, multiplied by your internal hourly value. Businesses that adopt AI correctly typically see 6-10 hours returned per head, per week. Run that number for your own team before you speak to anyone, and you will find every conversation afterwards is shorter and clearer.
Should the partner specialise in one platform?
For most non-technical teams, yes. Depth on one platform produces capability; breadth across many produces familiarity with none. The exception is an organisation genuinely running several platforms at scale, where a broader partner earns their place. Ask any specialist to argue their platform choice and to name its weaknesses. If they cannot name the weaknesses, the specialism is marketing.
Do we need to fix governance before we start?
You need it running alongside, not finished first. Waiting for a perfect policy is how firms spend a year not adopting anything while their people quietly use consumer AI accounts with company data in them. Set the boundaries, name the approval points, start, and tighten as you learn what your team actually does.
Can we do this internally instead?
Sometimes, and you should consider it seriously. If you have someone credible internally with the time genuinely protected, an internal programme can work and will hold better than an external one. It fails when the role is added to a full job, because adoption work is the first thing dropped when the day gets busy. Be honest about whether the time is real.
How long should an engagement run?
Long enough for a second and third contact after the initial delivery. Our own programme runs four months for that reason. Anything under a month is an event rather than a programme, and events produce a good day and very little else.
We are a Microsoft house. Does that rule Claude out?
No, and it is now one of the easier starting points. Claude is generally available in Microsoft Foundry and in Microsoft 365 Copilot, and it works across Word, Excel, PowerPoint and Outlook, so your people stay in the applications they already use. A Microsoft commitment is a question about where the work happens, not about which assistant is allowed to do it.
What if we have already tried and it did not stick?
That is the most common position we are called into, and it is a better starting point than it feels. Your people have used the tool, so the fear is specific rather than abstract, and you know what did not work. The diagnosis is usually the same: nobody addressed the job-security question, and nobody measured anything, so the programme quietly ended without a decision being made.
Darren Boyle, and why he takes this position.
I founded The AI Adoption Agency in Scotland and work with UK organisations on one platform, Claude by Anthropic, rather than on all of them. That is a deliberate limit and it is the thing I would most want a buyer to interrogate, which is why it is question three on the list above.
The position underneath this guide is one I will defend: return from AI is a behaviour problem, not a technology problem. Almost every firm I speak to already has the capability sitting unused inside a licence somebody is already paying for. What stops it is not the software. It is a room full of people who have quietly worked out that a tool which does part of their job is a tool that raises a question about their job. Nobody says that in the session. Everybody thinks it. A programme that does not deal with it directly produces a good day and no change.
There is a part of this that has nothing to do with knowing the platform. Getting buy-in is not a teaching problem, it is a people problem, and it turns on whether the person standing in front of your team can meet them where they are and move them on from there. Ten people in a room are in ten different places. One is already using it at home. One is quietly worried about their job. One has decided it is beneath them. The rest are waiting to see which way the room goes. Reading that, and pushing the dial without pushing anyone into a corner, is a different skill from knowing the software, and it is the one that decides whether anything holds after you leave. Your team has to trust the person and, ideally, enjoy the journey with them. Nobody changes how they work for someone they are enduring. That is why question ten is on the list, and I would ask it of us as hard as you ask it of anyone else.
That is also why this guide tells you when not to hire us. If the four situations above describe you, we will say so on the call. It is a shorter conversation and a better outcome than a proposal that was never going to work.