Your clients have started asking about AI. Not in a vague, read-an-article way, they are asking what you are going to do about it, and they are asking before they ask about security. Most MSPs have no billable answer. That gap between what clients want and what the channel actually sells is now the clearest growth opportunity in managed services, and it is closing.
The gap in one number
48% of MSPs rank AI and automation as the top client need for 2026, ahead of security and backup. Just 13% are currently generating meaningful revenue from it. (Kaseya 2026 State of the MSP Report, survey of more than 1,000 managed service providers worldwide)
Read those two figures together and the picture is unusual. Demand signals in this channel are normally trailing indicators: clients ask for what they have already been sold. This one is inverted. Clients have arrived at AI on their own, ahead of the providers, and they have put it above the two categories MSPs have spent a decade learning to sell.
The 35-point spread between demand and monetisation is not a marketing problem. MSPs are perfectly capable of marketing a service they can deliver. The spread exists because most MSPs genuinely cannot deliver this one yet.
Why the gap exists
You automated your own business, not a product
Kaseya found 53% of MSPs already use AI to automate ticketing, patching and monitoring. That sounds like readiness. It is not, it is cost reduction. Internal automation shows up in your margin, never on an invoice. And it is shallower than it looks: more than half of MSPs have automated only about a quarter of their workload.
The distance between “we use AI to triage our tickets” and “we sell an AI service” is the entire distance between operations and product. One is a tool you configured. The other needs scoping, a defined outcome, a support model, a price, and something to hand the client when they ask what they are paying for.
AI services are engineering work, and MSPs staff for support
The service your client is actually asking for, make our documents searchable, stop us re-keying this data, get Copilot working properly is a build. It needs someone who can model the client’s data, wire systems together, handle permissions and identity, and maintain the result when a vendor changes an API.
That is not the skill set an MSP hires for, and the hiring market is not helping. The share of MSPs reporting difficulty hiring skilled technicians nearly doubled year over year, from 9% to 16%. Recruiting your way into an AI practice in 2026 is slow, expensive and uncertain.
The failure rate is real, and owners know it
MSP owners are not being timid without cause. Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027, on escalating costs, unclear business value and inadequate risk controls. MIT’s Project NANDA, looking at more than 300 enterprise deployments, found 95% of generative AI pilots delivered no measurable profit impact at all.
An MSP that stakes its reputation on an AI rollout that quietly fails does not just lose the project. It loses standing on the account it spent years building. Caution here is rational. It is just no longer cost-free.
Why it is urgent this year
If the economics of the core business were healthy, waiting would be defensible. They are not.
| Pressure | Finding | Change |
| New client acquisition | 71% of MSPs call it their top challenge | Now the number one constraint |
| Deal size | 41% have typical customer spend above $25,000/year | Down from 75% the prior year |
| Demonstrating value | 19% struggle to show value to prospects quickly | Almost double the previous year |
| Technical hiring | 16% report difficulty hiring skilled technicians | Up from 9% |
All four figures come from the same Kaseya survey. Together they describe a market where winning new logos is harder, existing accounts are worth less, and the thing that used to differentiate you being competent and responsive, is no longer enough to make the case quickly.
Meanwhile the categories that are growing are the ones providers committed to early. 71% of MSPs reported year-over-year revenue growth in cybersecurity; 50% reported growth in business continuity and disaster recovery. Security paid off for the MSPs who built the practice before it was obvious. AI is sitting in exactly that position right now, with a shorter runway.
Four AI services that actually sell
Ignore the demos. These four have real SMB demand, a definable scope, and an outcome a client can verify.
1.Microsoft 365 Copilot readiness
The highest-demand and most under-served of the four. Clients buy Copilot licences, switch it on, and discover it surfaces documents people were never supposed to see, because permissions have been accumulating unmanaged for a decade. Copilot does not create that problem, it exposes it.
The service is permission and governance remediation before rollout: audit sharing and access, fix oversharing, impose a sensible SharePoint structure, apply sensitivity labelling, then enable Copilot and train users. It is concrete, it is billable as a fixed-scope project, and it has an unmistakable before-and-after. For MSPs already managing a Microsoft estate, it is the shortest path from where you are to a saleable AI service.
2. AI ticket triage, sold as a client service level
If you have already pointed AI at your own queue, productise it. Rather than quietly banking the efficiency, offer it as a tier: faster first response, out-of-hours triage, automated categorisation and routing, with the response times written into the agreement.
You are selling the outcome, not the technology. The client does not buy “AI triage”, they buy a response-time commitment you can now afford to make. This is the one service where internal automation converts directly into a priced tier.
3. Document and workflow automation on the stack they already own
Every SMB has three or four processes that run on re-keying: invoices retyped into the finance system, form submissions copied into a CRM, reports assembled by hand each month. These are unglamorous, high-irritation, and genuinely solvable with the Microsoft or Google tooling the client already licenses.
Scope one process, automate it, measure the hours returned, then use that number to sell the next one. This is the most reliable land-and-expand motion in the category, because the first engagement pays for itself in a timeframe the client can see.
4. AI governance as a retainer
Staff are already using AI tools, licensed or not. Most SMBs have no policy, no list of approved tools, no view of what company data is being pasted into them, and no answer when a customer or insurer asks. Only 21% of organisations have a mature governance model for agentic AI, according to Deloitte’s 2026 survey of 3,235 leaders.
Acceptable-use policy, an approved tool list, access review, quarterly usage reporting, and a named person accountable for it. It is recurring by nature, it is low-delivery-cost once the framework exists, and it pairs with every other service on this list.
How to package and price it
Two rules carry most of the outcome.
Separate licence resale from service work. Licence margin is thin and gets thinner as vendors compete. Service margin is yours. Keep them on separate lines so pressure on one does not quietly erode the other, and so the client can see what they are paying you for versus what they are paying Microsoft for.
Sell a project first, then a retainer. The project establishes scope, proves value on something measurable, and gets you paid while you learn the client’s environment. The retainer is where the recurring revenue lives. Do not attempt to sell an open-ended monthly AI engagement to a client who has not yet seen you deliver something.
| Stage | What it is | Commercial shape |
| Readiness | Assessment, permission and governance remediation, pilot group | Fixed-fee project |
| Rollout | Enablement, configuration, user training, adoption tracking | Fixed fee per user band |
| Govern | Policy, access review, usage reporting, optimisation | Monthly recurring |
| Expand | Next automated process, next department, next site | Project, then added to retainer |
Price the service work against the outcome rather than the hours. A permissions remediation that prevents one oversharing incident is not priced by how long the audit took. And do not price against what the licence costs, the two numbers are unrelated, and anchoring to the licence is how MSPs end up giving away the engineering.
Build, buy or partner
The instinct is to hire a developer and build a practice. The evidence argues against it as a first move. MIT’s Project NANDA found buying capability from specialised vendors and building partnerships succeeded about 67% of the time, while internal builds succeeded roughly a third as often.
The split that works is straightforward: own the parts that are actually yours, the client relationship, the scoping conversation, the governance framework, the ongoing service and partner for the engineering underneath, at least until a service line has proven it sells.
This keeps the economics honest. You find out whether clients will pay for Copilot readiness by selling three of them, not by carrying a developer’s salary for nine months to discover they will not. And when a line does prove out, you have delivery history to hire against instead of a guess.
What to look for in an engineering partner: they work under your brand, they can evidence comparable work rather than describe it, and they will scope a single project before asking for a retainer. If a prospective partner wants a long commitment before delivering anything, they are managing their risk by transferring it to you.
The next 30 days
Narrow beats comprehensive. One service, one client, one proof point.
- Pick your five best Microsoft accounts and run a sharing and permissions review on each. You will find oversharing. That finding is the sales conversation, it is a risk they did not know they had.
- Choose one service from the four above and scope it properly: what is included, what is not, what the client gets at the end, what it costs. One page.
- Sell it once, at a price that makes you slightly uncomfortable. A first engagement priced to win teaches you nothing about whether the service is viable.
- Measure the outcome in the client’s units, hours returned, incidents avoided, days saved per month and write it down. That number is what sells the second one.
- Decide the engineering question before you sell the third, not after. Two successful projects is enough signal to choose between hiring and partnering.
The MSPs who built security practices early did not wait for the category to be proven. They took a position while it was still uncomfortable, and by the time it was obvious they had the delivery history, the references and the pricing power. AI is at that same point now, with the difference that clients are already asking. That does not usually last.
Questions MSPs ask
Q. What AI services can an MSP actually sell to SMB clients?
A. Four productise reliably today: Microsoft 365 Copilot readiness (permission and governance remediation before rollout), AI-assisted ticket triage delivered as a client-facing service level, document and workflow automation built on the client’s existing Microsoft or Google stack, and an ongoing AI governance retainer covering policy, access review and usage reporting. All four bill as a fixed-scope project followed by a monthly recurring engagement.
Q. Why do most MSPs fail to make money from AI?
A. Most have used AI internally rather than built anything to sell. Kaseya found 53% of MSPs already use AI to automate ticketing, patching and monitoring, that reduces cost but produces no billable line item. Turning internal automation into a sellable service requires scoping, packaging and engineering work that support-staffed teams are not structured to do.
Q. Should an MSP build AI services in-house or partner?
A. Build the parts that touch your clients and your process, scoping, governance, the client relationship. Partner for the engineering underneath. MIT’s Project NANDA found vendor partnerships succeed roughly 67% of the time while internal builds succeed about a third as often, and Gartner expects over 40% of agentic AI projects to be cancelled by the end of 2027. Hiring a development team to find out whether a service line works is the most expensive way to run the experiment.
Q. How should MSPs price AI services?
A. Separate licence resale from service work and price the service work on outcome, not hours. A typical structure is a fixed-fee readiness or rollout project to establish scope and prove value, followed by a monthly governance and optimisation retainer that grows as the client adds users, locations or use cases. Keep licence margin and service margin on separate lines so pricing pressure on one does not erode the other.
Q. Is it too late to start an AI practice?
A. No. Only 13% of MSPs are earning meaningful revenue from AI services, so the category is still early by any reasonable measure. The window that is closing is the one where clients are asking unprompted, that demand makes the first sale dramatically easier than it will be once every provider in your market is pitching the same service.
You may also like : MSP Partnership: Scaling IT Services
