On Tuesday, I met a potential customer to introduce our AI practice. Twenty minutes in, it was clear they had already designed an AI programme to rival ours. They had written it up as a request for information with four scopes, sent it to other providers and were already into the second round of reviewing responses.
We weren't part of it. Their enterprise architect was kind about it: the process was already under way, and it was a bit late for us to join.
That comment has stayed with me all week. Not because we lost a deal — you can't lose a race you weren't in. It's because of what it shows about who is now leading.
Here's the short version. Some buyers are now ahead of the providers who serve them. And being their managed service provider no longer guarantees you a seat when they buy AI work.
This came, unprompted, from an organisation that calls itself “not a tech company” and “very cautious, very risk-averse”. Their four scopes were:
Behind this sat a governance model with seven parts:
Now ask how many providers could deliver that fourth scope today. Few even have a slide on managing AI agents at scale, and almost nobody does it yet. The buyer asked for it anyway. He could see it coming, and he wanted the market to answer for it.
Two days later, a head of IT at a retail group told me he couldn't say whether anyone in his business was using an AI agent right now, and that it worried him. So the gap between buyers is huge. Some are now well ahead of the people who advise them. Others are years behind. The average tells you nothing, yet it's what most providers plan around.
First, they assume their current provider can do the work because it has a presentation about it. But saying you can do something and having done it are not the same. Only one of them has been tested in real delivery.
Every provider will say yes. The real test is whether they can tell you a true story about the scope you care about most: who they did it for, when, what went wrong and what it cost. One real project is worth more than five sales documents. A provider who has really done the work will tell you what went badly without being asked, because that's where they learned the most.
The second mistake is treating this as a question of loyalty. It isn't. It's about what kind of work is being bought. For twenty years, infrastructure work went to the infrastructure provider, because that was the shape of the problem. AI strategy doesn't follow that path. If you've stopped seeing your provider as strategic, they won't be on your next request for information, and nobody will tell them why.
This applies to providers like us too. The same week, I made a deliberate call to pause a data platform practice. Doing it properly meant hiring deep expertise first, and I'd rather focus than spread us thin. On AI, we're building our track record one project at a time, and every engagement teaches us something we take into the next.
That meeting didn't tell me we weren't trusted. It told me something bigger: buyers now see AI as its own kind of work, and they go looking for it separately. Being someone's managed service provider no longer gives you a first claim on their AI work. You earn that seat with real delivery and real stories. That's the shift every provider in our market needs to take seriously, and we're taking it seriously too.
Write your own scopes before you ask anyone for a proposal. If you can't name four, you'll end up buying whatever the first provider is best at selling. The four above are a better place to start than anything a provider will send you.
Ask each provider which of your scopes they've delivered from start to finish for another customer this year. Ask for the actual project, not a general description. Then ask what went wrong.
Ask what they've stopped doing this year. A provider that hasn't dropped anything hasn't made any real choices. And a provider that hasn't made choices is still hoping the old way of working will last.
If you're ahead of your provider, don't wait for them to catch up. And if you're a provider reading this, face the uncomfortable truth: the request for information may already be out, and nobody sends a copy to the provider they've quietly stopped seeing as strategic.