AI adoption MSP SMB: from switching a licence on to really putting it to work
AuthorMSP Late NightPublished on:AI adoption MSP SMB is currently the conversation coming up in every managed service provider boardroom, and at the same time the conversation almost nobody has an honest answer to. Everyone wants to do something with it. Clients are asking for it. And when you dig into it with that client, they do not know themselves what they want from it. Meanwhile the clock is ticking, because the competitor round the corner has the same story.
You probably recognise it. A mid-sized client of around 40 staff calls and says they want “something with AI” as well. You sell Copilot licences, you switch them on, and three months later you see in the tenant that usage is nearly zero apart from a few enthusiasts. The licence costs keep running. The client is disappointed. And you are empty-handed at the next quarterly conversation.
The problem does not sit in the technology. The technology is already further along than that client organisation. Data governance was not in order before AI existed, and now that it becomes relevant, it suddenly shows how much deferred maintenance is there. Add to that the fact that as an MSP of five, ten or twenty people you can barely oversee the whole security field, and that you now also have to know which data may go into a model at all before anything comes out. It is no wonder many MSPs are having the internal discussion about whether to step into this. The conclusion of that discussion is nearly everywhere the same: we have to, because the client is counting on it.
What MSPs have to do for mid-sized business: from switching on to putting to work
What works in practice is a shift in how you define your role. No longer the party that supplies a licence and switches it on. Instead the party that sits down next to the owner, in language they understand, and helps them actually put to work something that is switched on. That is a different craft from traditional MSP work and it asks for different people at the table.
Concretely that means: making roadmaps with the client. Not about which endpoint protection you roll out next year, but about what is changing in their sector, what legislation asks of them, and which processes they can redesign step by step with AI in them. You are accountable as a supplier for matters such as NIS2, so take that client along in it instead of throwing a report over the fence. That is the skill that will make the difference over the coming years between the MSP that keeps margin and the MSP that competes on price.
Why the small MSP has to make a choice
If you are three to ten people, it is nearly impossible to do this on the side. The security field alone asks for full-time attention. Building an AI practice on top of that, in which you know which data may and may not go into which model, and translating that knowledge to an owner who has only just got their bookkeeping in order, is not realistic for most small outfits. There are then roughly two ways out.
“You have to move from switching on to putting to work.”
Specialise or get under an umbrella
The first way out is choosing a very small niche and becoming the best in it. Accountancy practices for example, or orthodontic practices, or a specific niche in manufacturing. Then you can leave the whole broader field alone because your clients all resemble each other. You know the software, you know the legislation, and you can make an AI roadmap that is right for ten clients at once. The second way out is becoming part of a larger umbrella, where the back office, the purchasing discussions with vendors, recruitment, HR and compliance are arranged centrally. You then stay close to the client, your account manager still calls the owner in their own region, and the rest is done elsewhere. What you want to avoid in both cases is the middle position: too big for the niche, too small for the complete package. That is exactly the place where margins evaporate and where you will take the first blows over the next two years.
What works well in that second model in practice is a central vendor and partner programme. Most small MSPs simply never get round to looking seriously at partner programmes, earning the right certifications or using the marketing campaigns vendors co-fund. There is a lot of money left on the table there. And the real game with vendors, by the way, is no longer about two per cent extra discount through scale. It is about which vendor has open APIs, so that you can get the data out of their tool and build something clever in your own warehouse. The IP of the future MSP does not sit in a self-built PSA package, but in the layer underneath where you combine data from several vendors cleverly. That layer is what you will sell as an adoption partner, not the individual licence underneath it.
What the MSP role is in three years
The role that is coming looks much more like that of an orchestrator. Just as you used to keep an overview of a client install base, before long you will keep an overview of their agentic workforce: the collection of bots and AI agents working for them. Governance over that workforce, knowing what they do with which data, and being able to step in when it goes wrong. That comes with a branching into specialisms as well: cybersecurity, data and governance, and adoption. Those three together decide whether a client is still workable in three years.
It also means the client becomes considerably more self-sufficient, because before long they will have an agentic workforce of their own to solve things they still call you for today. That sounds like a threat to your service desk revenue, and in the short term it is. But the revenue you lose on tickets you win back on orchestration: who builds and guards that workforce, who makes sure agents do not get at each other data, who explains to that client accountant why an automated invoice approval meets data protection law? That is not a threat to your margin, provided you move to that orchestrating role in time and take your people with you. The engineers solving tickets now have to be sitting in on adoption work today, otherwise in two years you have a team that cannot do the new work.
“The MSP becomes the orchestrator of the agentic workforce.”
Start with the menu, not with the licence
The practical start is much simpler than the hype suggests. Make sure your menu is complete, so that you can support a mid-sized client on the basics of AI, data governance and adoption, yourself or through a partner you orchestrate. Anyone with that in place can walk into every quarterly conversation with their head up. Anyone still waiting until the client knows exactly what they want will lose that conversation more and more often over the coming years.