Forecasting AI, Skipping The Redesign
The channel has decided AI is its next big revenue line. The evidence about what makes AI pay hasn’t caught up.
Start with Informa’s MSP 501 survey, more than six hundred managed service providers. Fifty-seven percent expect AI to be among their biggest revenue gainers over the next twelve months. A year ago, that number was thirty-seven. Ninety-one percent now offer or use AI, and about a third expect their AI revenue to grow by more than twenty percent. Only managed security ranks higher.
Now look at how those same providers describe their own use of it. Fifty-four percent name employee productivity as a key internal use. Thirty-five percent name reducing repetitive work. Those are tools in people’s hands. They are not a business that runs differently.
That difference is the one IHL Group president Greg Buzek put in front of channel partners at BlueStar’s VARTECH event, as Managed Services Journal reports. He cited McKinsey research. Only eleven percent of leaders surveyed said their organizations had reached what McKinsey calls reinvention: redesigning roles, workflows and operating models around AI. In that group, forty-eight percent report realizing enterprise value. Among companies that simply give employees AI tools, it’s thirteen percent. The ones that rebuilt the work report value nearly four times as often.
And McKinsey notes it deliberately recruited more advanced companies, so it warns that eleven percent isn't a market-wide estimate. Even in a sample tilted toward the leaders, about one in ten did the redesign. Nearly half of those got value. Most of everyone else didn't.
The chief executives say the same thing from the top. PwC’s annual survey of 4,454 CEOs across ninety-five countries found fifty-six percent have seen no significant financial benefit from AI so far. One in eight said it delivered both lower costs and higher revenue. Those were two to three times more likely to have embedded AI extensively across the business.
So put the three side by side. The channel’s AI forecast is climbing fast. The redesign that produces measurable value is rare, even among the leaders McKinsey went looking for. And most CEOs can’t point to a return.
The forecast is built on adoption. The returns show up where the work was rebuilt. Those are not the same number.
So if the returns come from rebuilt work, what are all those AI programs actually measuring?
When Usage Stands In For Results
When the work hasn’t been rebuilt, there’s no new outcome to point at. So leaders point at the one thing they can see, whether people are using the tool.
That substitution is the whole problem. A redesigned process gives you a result to measure: a proposal out in a day instead of a week, a ticket closed without a second touch. A tool handed to the old process gives you a usage number. And once usage is what leadership watches, usage is what people produce. McKinsey reaches for an old comparison. When factories first swapped steam engines for electric motors, they kept the same floor plans, the same line, the same management, and productivity barely moved. The gains came only after they rebuilt the factory around the motor.
Visier surveyed a thousand full-time US workers on exactly this. Visier sells workforce analytics software, so weigh it with that in mind. Forty-eight percent said they have exaggerated their AI usage or expertise to colleagues or leadership. Forty-five percent said they feel pressured to use AI even when they’re not confident using it well. Nearly sixty percent are unsure what their employer’s plan for AI is. Visier’s own conclusion is that the driver of what it calls performative AI use is a lack of trust in leadership.
Pressure to use it, no plan for what the use is for, and nearly half of the people surveyed shading the number. That’s what a usage target produces.
You might reasonably say that some usage is inflated but most of it is real, and real usage is still productivity. That’s the assumption Forrester takes apart. Writing about content and creative teams, a Forrester analyst describes AI making an individual employee more productive while making the organization less efficient. The illustration: if AI saves one employee twenty minutes but creates thirty minutes of work for others downstream, that’s not a productivity win. The first phase of adoption, the analyst writes, was about individual productivity. The harder phase is the whole workflow, including the work AI created for everyone downstream.
So even honest usage measures the wrong thing. It counts the twenty minutes at one desk and misses the thirty at the next.
In plain terms, without a redesigned process there’s nothing to measure but activity. Activity gets performed when it’s watched, and it misleads even when it isn’t.
Which makes the tool that watches activity the easiest AI product there is to sell.
And somebody now sells that tool to you, so you can sell it to your clients.
The Usage Meter Hits Your Catalog
Start with Insightful. At MSP Summit, it launched a partner program that gives managed service providers the full customer lifecycle for its workforce analytics platform: installation, onboarding, support and billing, at a thirty percent partner discount. As Telecom Reseller describes it, the platform shows a business how employees spend their time and which applications they use. On its own product page, Insightful lists an AI Adoption Report, a view of how AI tools are being used across teams. That’s the usage number, packaged, discounted and ready for your catalog. Telecom Reseller’s caution is a fair one: there has to be enough customer demand to justify adding it. The supply side is betting the demand arrives.
Now look at where that number lands. The OECD surveyed more than six thousand firms across six countries on software that instructs, monitors or evaluates workers. Sixty-seven percent of US firms use software to sanction poor performance. In the four European countries surveyed, it’s four percent. Seventy-two percent of US firms track how fast people work. Two cautions: the fieldwork dates to 2024, and managers answered, not workers.
But the finding that matters most for you is a different one. The OECD says most of this software isn’t AI and isn’t labeled as monitoring at all. It’s the ordinary business stack, the HR system, the project tracker, the time-tracking tool, that picked up a scoring function along the way.
That’s the stack you already run. And in the American workplace, a measurement in front of a manager is very often a measurement used on an employee. So a usage report you install doesn’t necessarily stay a report. It can become a ranking of who performed AI most convincingly. And the OECD found that among US managers who acknowledge collecting worker data, nine in ten say workers can’t opt out, and more than half say workers can’t request corrections. If you install it, you’re the one holding that data.
So here’s the choice. Sell the usage meter. Clients who never rebuilt the work will ask for it, it’s discounted, and it renews. Accept that you are now keeping score on a number nearly half of workers admit they shade. Or decline to sell AI measurement that isn’t tied to a process someone actually rebuilt, and sell the before-and-after on one changed workflow instead: how long it took, what it cost, and how often someone had to fix it.
Before you make that choice for a client, there’s a test to run on your own shop first.
Because the same survey that shows the channel forecasting AI revenue shows most providers using AI the way their clients do, as a productivity tool on an unchanged process. Before you sell anyone a before-and-after, produce one. Pick one recurring workflow in your own shop, proposals or onboarding, and record how long it takes and how often it gets corrected. Rebuild it, measure again, and that number is the one demo a client can’t get from a vendor.
What to Consider
- Pick the workflow by what it costs you, not by what AI can touch. Buzek’s test, as Managed Services Journal reports it, is any recurring activity that eats more than four hours a month. Choose one with an outcome you can count, like proposal turnaround, days to onboard a client, or how often scope or pricing gets corrected. Write the baseline down before anyone changes anything.
- Count the minutes at the next desk. Forrester’s twenty-minutes-saved, thirty-minutes-created illustration is the trap for your own measurement too. Time the whole workflow, including the review and rework the next person does, and include what the tools cost. If you skip that, you’ve just built the activity number you’d decline to sell a client.
- Don’t put your own team on a usage meter. If you won’t sell clients a report that scores people on how much they use AI, don’t run one internally. Measure the process, not who opened the tool, or you’ll teach your own technicians the behavior Visier found in nearly half of workers.
If this trend continues: By the next MSP 501 survey, the AI revenue this year’s respondents forecast will either be booked or it won’t. The providers whose forecast held will be the ones who can show a rebuilt workflow of their own, with a measured before and after.