Forty-eight per cent of managed service providers say AI and automation is the thing their clients want most in 2026, ahead of security and ahead of backup. Thirteen per cent are making any real money from it.

That gap is the whole opportunity, and it is unusual. Demand is not the question here, and neither is whether the work exists. The question is why the people best placed to sell it are not selling it.

Those two figures come from Kaseya's 2026 State of the MSP Report, a survey of more than a thousand providers published in April. The same survey is worth reading for the three numbers nobody quotes from it.

The numbers underneath the headline

Deal sizes are falling. The share of providers whose typical client spends more than twenty-five thousand dollars a year dropped to 41%, from 75% the year before. That is not a slow squeeze. That is the average contract changing shape inside twelve months.

Seventy-one per cent say winning new customers is their hardest problem, and 19% say they struggle to show a prospect their value quickly, which is nearly double the year before. Meanwhile the share reporting difficulty hiring skilled technicians went from 9% to 16%.

Read those together and the strategy writes itself. You cannot hire your way out, the contracts you already have are shrinking, new logos are getting harder, and the service your clients are actively asking for is one that almost nobody in your market has packaged yet.

The market opportunity, with the caveat attached

Analysts put the AI consulting and support services market at around fourteen billion dollars in 2024, growing to roughly seventy-three billion by 2030 at a compound rate above 30%.

Treat that as direction rather than fact. Other houses forecasting the same category land nearer twenty billion for the same year, which tells you the definitions are doing as much work as the research. When two credible forecasts differ by a factor of three, the honest reading is that the category is growing fast and nobody can size it properly yet.

The more useful number is the one in your own accounts. You already know what your clients pay you, how many of them there are, and roughly what a project engagement is worth in your market. The addressable opportunity for most providers is not a slice of a global forecast; it is a second line of revenue inside a client base you already hold.

Why an MSP wins this work against a consultancy

Every AI consultancy in the country is trying to get into the room you are already sitting in.

You have administrative access to the systems. Not a slide about the systems, not an interview about the systems. The actual tenant, the actual identity structure, the actual permissions tree, and a support history going back years that tells you exactly where the work jams.

You have the contract. There is no procurement cycle to win, no security review to pass, no reference to supply. The commercial relationship exists and the invoice already goes out every month.

You have the trust, which is the expensive part. When a business hands a consultancy its process documentation it is taking a risk on a stranger. When it hands you the same thing, you already hold its passwords.

And you have the context nobody can buy. A consultancy discovering a client's operations starts at zero and bills for the privilege. You have engineers who could tell you, unprompted, which department still does something by hand that should have been automated three years ago.

The catch is that none of this is in a sellable form. It sits in ticket histories, in engineers' heads, and in the specific frustration of whoever looks after that account. The work of the pivot is turning what you already know into something a finance director can sign.

The build, in three phases

Nobody makes this shift in a quarter. A realistic plan runs about a year, and the first third of it produces no client revenue at all.

Months one to three, fix your own house first. Automate your own service desk before you sell automation to anyone. Pick the three highest-volume ticket types, automate them properly, and measure what happens to first response. You will learn more about what breaks in a live automation from doing it to yourself than from any vendor deck, and you will free the engineering hours you are about to need.

Standardise your own data while you are there. Twelve months of consistent ticket categories is the difference between a triage model that works and one that embarrasses you. Nominate one person as the automation lead. Not a committee.

Months four to six, sell diagnosis rather than implementation. Package a fixed-fee assessment and run it on three friendly accounts. This is the move that pays for all the others, because it converts an open-ended conversation about AI into a scoped, priced, finite engagement with a deliverable at the end.

The assessment should answer three questions for the client. Where is AI already being used here, sanctioned or not? Which processes are worth automating, ranked by hours and cost? And what has to be fixed before any of it is safe to switch on, which in practice usually means permissions nobody has audited since the tenant was built.

Sell the assessment on its own merits and let the implementation work follow from what it finds. An assessment that only ever recommends your own services stops being a diagnosis and becomes a sales call with a worksheet.

Months seven to twelve, build and then run. Start building automations for the processes the assessments surfaced. Price each build as a fixed fee. Then, and this is the part most providers skip, sell the ongoing management of what you built.

Automations break in ways servers do not. The API changes shape, the model's behaviour drifts, an edge case nobody modelled turns up in week six. Somebody has to watch that, and the client cannot. This is the recurring revenue and it is the reason an MSP should be doing this work rather than a project shop that disengages at handover.

What to charge

The bands being quoted across the market at the moment: a fixed-fee assessment in the low tens of thousands, individual automation builds from around five thousand to twenty-five thousand depending on how many systems they touch, and a monthly retainer per live automation for the ongoing management.

Two pricing rules matter more than the numbers. Stop billing this work hourly, because you will use AI to build faster and your invoice will shrink while the client's benefit grows. And pass model usage costs through to the client rather than absorbing them, because a workflow that suddenly runs ten times more often should not quietly eat your margin.

Where a client has a baseline you can both agree on, the strongest position is to price from the saving rather than the effort. Capturing a fixed share of verified annual savings aligns you with the outcome and takes the argument about day rates off the table entirely.

Who should not do this

If you are firefighting, do not start here. A provider whose own service desk is underwater will deliver a bad assessment, and a bad assessment is worse than none because it spends the trust you would have needed for the second conversation.

If your clients are almost all under twenty people, be careful too. The arithmetic on a bespoke automation rarely works at that size, and you will do better reselling and managing platform tools than building anything custom.

And be honest about the first engagement. It will take longer than you scoped, the client's documentation will be wrong, and the process you were told about will turn out to be three processes. Price the first two deliberately low and treat them as training.

Where to start this week

Pick one account. Ideally a good relationship, a renewal far enough away that nobody suspects a manoeuvre, and a department that is not the one your sponsor runs.

Map how that client's business actually works: the departments, the systems, the processes running between them, and the hours each one consumes. That map is the assessment deliverable, it is the thing that makes your findings specific rather than anecdotal, and it is what turns "their invoicing is a mess" into a number with a name attached.

abi. Clone for MSPs builds that map from interviews with the client's own team rather than from your engineers' time, exports it under your firm's name, and gives you five client maps free to start with. Given that 19% of providers now say they cannot demonstrate value to a prospect quickly, walking into the room with a working model of the client's own business is not a bad answer to that problem.

Sources: Kaseya, 2026 State of the MSP Report, April 2026. AI consulting and support services market forecasts, Technavio and Tech Market Experts, 2024 to 2030.