The 13% Problem
Here’s a puzzle hiding in plain sight: the channel has never been pushed harder to sell AI, and it has almost nothing to show for it.
Start with Kaseya. The company put out a direct call urging managed service providers to move fast on AI services — don’t wait, the window is now. And then, in nearly the same breath, it published the numbers that make the urgency look strange. Almost half of MSP clients are already asking for AI solutions. But only about thirteen percent of providers are generating any meaningful revenue from it. Thirteen. Kaseya’s own CEO, Rania Succar, pointed at what she called a governance gap behind the stall, and the supporting data is brutal: seventy percent of AI proof-of-concept projects never make it into production, and only one in five organizations gets from pilot to actual deployment. So the demand is there, the vendor is pushing hard, and the conversion to money is almost nonexistent.
And it isn’t just Kaseya saying move. The trade press is running the playbooks alongside it — pieces with titles like “five AI services every MSP should be offering right now,” handing providers a ready-made menu of exactly what to sell. And yet for most of the providers being handed that menu, AI still isn’t the revenue driver it’s sold as — they’re handed the list and still can’t make it pay.
Now the part that kills the easy explanation. You might assume the demand is just soft — clients dabbling, nobody serious. Gartner says the opposite. In its research, more than seventy percent of CEOs say their organization’s IT operating model is not fit for the age of AI. And only twenty-four percent of CIOs believe their current model can actually adapt to it. Read that carefully: the people at the very top already know the old way of running IT is broken for what’s coming. The need isn’t soft. It’s structural — and they’re telling you so.
So that’s the picture, before we’ve said a word about why. Enormous push. Real, admitted, structural demand. And almost no one turning it into revenue.
The Tool vs. The Work
The reason all that demand won’t convert comes down to a single mismatch: the thing being sold isn’t the thing that’s worth money. What gets pushed onto the MSP is a tool. What the client actually needs is the work — and a tool is not work.
Watch where the value really sits. Forbes looked at what happens when AI writes code, and the finding is the whole mechanism in miniature: security flaws show up in AI-generated code nearly three times as often, and logic errors about seventy-five percent more often, than in code a human wrote. So the machine produced the output, fast and cheap — and then left behind the part that was always the actual job. Someone has to read it, catch the flaw, decide whether it’s safe to ship. That judgment — the verdict on whether the output is right — is the work. It’s labor-intensive, it doesn’t come in a box, and it’s exactly what the MSP has been doing all along without ever naming it as the thing they sell.
And here’s why “just buy the tool” can never close that gap. A tool scales — that’s the entire reason a vendor builds one. You can sell the same product to ten thousand providers and a million seats. The work doesn’t scale like that; it has to be done, case by case, by someone accountable. So the vendor sells the thing that scales, because that’s the thing that makes the vendor money — and quietly leaves the part that doesn’t scale on your desk.
Listen to the vendors prove it on themselves. Anthropic — one of the companies setting this whole pace — now says about sixty-five percent of its own product team’s code is AI-generated. Sit with that. The company telling the channel to move fast is running machine-written code at exactly the scale where, by Forbes’s numbers, the errors pile up. They didn’t hand the judgment to the machine. They kept humans on top of it — because they know the output isn’t the value. The verdict on the output is.
Which means the question was never whether to adopt the tool. It’s whether you’ve figured out how to sell the work the tool can’t do — and whether anyone’s made you legally have to.
The Wrong Answer’s New Address
So aim that at the client, because the cost of an unchecked AI answer has stopped being hypothetical — and it lands on whoever deployed it.
Start with the size of the problem. Even a strong consumer AI runs at around ninety-one percent accuracy. That sounds fine until you turn it over: roughly one in every eleven answers is wrong, at the scale of every query a business runs through it. And the blame doesn’t float off into the ether. In the data on AI failures, half of consumers say that when a company’s AI gets it wrong, they hold the company’s leadership responsible — not the model, not the vendor, the business that put it in front of them. It already has a price tag, too: one car dealership’s chatbot talked itself into a mistake that cost the company about five thousand dollars on a single transaction. Wrong answer, real bill, and it landed on the deployer.
Now watch that consequence harden from a bad day into a legal fact. A German court ruled that Google can be held liable for its AI’s inaccurate summaries — treating what the AI said as the company’s own statement, not some neutral tool’s output. That’s the same direction an Air Canada case already went, when a court held the airline responsible for what its chatbot promised a customer. Put those together and the line is clear: the era where “the AI said it, not us” was a defense is closing. When your client’s AI is wrong, that wrong answer is becoming, in the eyes of a court, your client’s own words.
So here’s the choice, and it’s the most hopeful thing in this whole story. The work the vendor menu skips right past — reviewing what the AI produces, catching the errors before they reach a customer, standing behind whether the output can be trusted — is no longer a nice-to-have. Your clients structurally need it, and the courts are about to require it. You can package that work and price it as the service it is. Or you can keep selling the tool the vendor handed you, and stay the unpaid backstop holding the liability the day a wrong answer becomes someone’s legal problem.
Why Do We Care?
Because two MSPs are about to look identical on paper — same vendor tools, same AI menu, same pitch — and split into two completely different businesses. The one that can show a client a documented accuracy-review process, with its own name on whether the output is right, is selling something the shop down the street cannot match by signing the same vendor contract. The separator was never the tool; it’s whether you can stand behind the work — and the firm that can will be the one still in the room when the client’s lawyer asks who checked the AI.
What to Consider
- Build the proof a competitor can’t produce on demand. The vendor’s tool is available to every MSP in your market by the end of the week; a documented method for reviewing AI output — what you check, who signs off, what evidence you keep — is not. Write that method down as your own named standard, because when a prospect compares you to the shop reselling the identical platform, the deliverable they can actually see is your accuracy process, not the tool you both happen to license.
- Compete on the liability question, because that’s where rivals are exposed. The MSP down the street is selling AI adoption and quietly hoping nobody asks who’s accountable when it’s wrong. Make that the question you raise first in the room — walk the prospect to the exact moment a wrong AI answer becomes their problem, not the vendor’s, and let them sit in that gap before anyone mentions a product. You don’t win this by demoing a better tool; you win it by being the only one who made the risk visible.
- Price the separation so it can’t be undercut on tool cost alone. If your AI offer is priced like a product, a competitor reselling the same product will always undercut you. Price it like accountability — the review, the evidence, the contractual ownership of a wrong output — so the comparison stops being “whose Copilot license is cheaper” and becomes “who actually stands behind what the AI does.” That reframing is the competitive moat, and it’s one a box-reseller structurally can’t follow you into.
If this trend continues, within the next twelve to eighteen months, “who reviews your AI’s output, and will they put it in writing” becomes a standard question on the buyer’s checklist — and the MSPs who can answer it with a documented, priced process will be taking the accounts of the firms still competing on whose vendor tool is cheaper.