Ask the person who signed for an AI tool how it's going, then ask the person who uses it every day. In a lot of manufacturing companies, you'll get two different answers. A survey released last week puts numbers on that split, and on a related problem: many manufacturers say AI is paying off without the tracking to show it.

The survey and its limits

Revalize, a software company that sells quoting, product lifecycle and CAD tools to manufacturers, released new research on September 29, 2026. Revalize commissioned the agency TEAM LEWIS to survey 500 business decision-makers at manufacturers with 100 or more employees. Fieldwork ran from July 7 to 17, 2026, with respondents recruited by OpinionRoute. The sample was 225 respondents in Germany, Austria and Switzerland, 50 in the U.K. and 225 in the U.S. The release notes that figures across those three segments aren't directly comparable.

There's more to flag than usual with this one.

The release also compares a first-quarter adoption figure with a mid-year one. It doesn't explain how the earlier baseline was produced, so I've left that comparison out.

Two answers from one company

The survey asked how far along each company's AI use really is. "Advanced adoption" meant AI built into multiple operations or fully embedded in how work gets done. C-suite leaders were more than twice as likely as senior managers to say their company had reached it.

The release gives the comparison but not the underlying percentages, so "more than twice as likely" is as precise as anyone can get from it.

Same companies. Same tools. A different answer depending on where you sit.

Neither group has to be lying for that to happen. An executive tends to see what got bought, what got announced and what showed up in a board update. A manager sees what actually gets used on a Tuesday afternoon, and what gets quietly worked around. Both views are honest. They're measuring different things.

In a smaller operation, that split can live inside one person. There's what you paid for, and there's what your crew actually opens. If you've ever been surprised to learn that a piece of software you pay for every month isn't really being used, you've seen this gap firsthand.

Credit without the math

Here's the core finding. Seventy-two percent of manufacturers in the survey say they can attribute business outcomes to AI. Only 49% say they measure AI return on investment rigorously.

Those come from two different survey questions, so the gap between them isn't a measured figure. But taken together they describe a situation many owners will recognize: a confident sense that the tool helped, sitting next to a much thinner record of how much.

Revalize's CEO says in the release that too many leaders are overattributing success to AI without the data to back it up. He also sells software built to address that, so take the comment as his pitch. The two numbers still stand on their own.

Saying a tool helped feels the same as knowing it helped, right up until someone asks for the figure.

I made a similar point about commercial contractors in AI Adoption in Construction Doubled. Most Contractors Still Can't Prove It Worked. That survey found most contractors couldn't yet point to a measured result. This one finds something slightly different: many manufacturers do claim a result, but about half aren't measuring it rigorously.

The napkin math problem

Half the manufacturers surveyed, 50%, said tracking of AI outcomes at their company is inconsistent or informal. Nearly a quarter said they're only somewhat confident AI is having an impact at all.

Eighteen years in construction taught me that a margin nobody tracked while the job was running is a margin somebody guessed afterward. The job might have made money. It might not have. Without the numbers collected along the way, nobody can say, and the next bid gets priced on the guess.

AI tools are following the same pattern. Informal tracking is a notebook, a gut feeling, a figure somebody remembers from a meeting. It can work for a while. It can't tell you whether to spend more next year, which tool to cut, or whether the time savings were real or just felt real during a busy month.

A tool nobody scores tends to get renewed out of habit, or cancelled on a hunch. Neither is a decision.

What "rigorous" doesn't have to mean

Measuring AI return doesn't require a data team. For most smaller businesses it means deciding, before a tool goes live, which number it's supposed to move: quote turnaround time, hours spent on documentation, rework rate, response time to customers. Then you write down where that number is today and check it again in sixty or ninety days. That's most of the discipline. The hard part is doing it before the purchase instead of trying to reconstruct it afterward.

A simple before-and-after

Here's what that can look like in practice, using a hypothetical. A 40-person manufacturer adds an AI tool to help its inside sales team draft quotes for custom parts. Before turning it on, the sales manager pulls the last two months of quote requests and notes how long each took from request to sent, and what share turned into orders. Call it an average of three working days, with a certain win rate.

Ninety days later, the manager runs the same numbers. Maybe turnaround dropped to one day and the win rate held. That's a result worth paying for, and it's easy to defend. Maybe turnaround dropped but more quotes needed revision after a customer caught an error. That's useful too, because it points to where the review step needs tightening. Or maybe nothing much changed, which is the most useful answer of all, because it stops the renewal.

None of those outcomes required a dashboard or an analyst. They required writing down a number before the tool arrived. That one step is what separates the manufacturers who can attribute outcomes to AI and back it up from the ones who can only say they believe it helped.

Ask the people who use it

The executive-versus-manager gap in this survey suggests one more habit. When you check the numbers, ask the people who use the tool daily what changed for them. They'll tell you whether the time savings are real, whether they've found workarounds, and whether the tool is being used the way it was intended. Their answer and the owner's answer won't always match. When they don't, the gap is usually where the real story is.

Pressure bought the budget, not the proof

One more pair of findings explains why so much AI spending runs ahead of measurement.

The share of manufacturers citing tariff and compliance costs fell from 50% to 39% compared with a year earlier. The release doesn't spell out the exact base for that figure, but it reads as the share of respondents reporting that pressure. The pressure eased.

Yet 44% said they're still accelerating AI adoption specifically to offset tariff and cost-driven pressure.

That tells you what's driving at least some of the spending: not a result someone confirmed, but a squeeze someone wants out of. Buying a tool to fight a cost problem is a reasonable instinct. The weak spot shows up when the pressure eases and nobody can say whether the tool was ever what helped. With only 49% measuring ROI rigorously, for about half of these companies that question has no ready answer.

A different manufacturing survey earlier this year, which I covered in Manufacturing AI Use Hit 94%. Where the Budget Went Is the Real Story, found manufacturers raising software spending even while expecting a difficult year. This survey adds the other half of that picture: the spending is happening whether or not the measurement is.

Near universal isn't the same as working

Sixty percent of the manufacturers surveyed said AI is integrated across multiple operations or fully embedded in their workflows. Only 8% said they're still in initial exploration. By these respondents' own account, adoption in this part of manufacturing is close to settled.

And still, only 49% measure AI ROI rigorously.

The sample caveats apply here more than anywhere. These are larger manufacturers in the U.S. and Europe, surveyed through a vendor with a reason to want this gap to look important. I wouldn't treat any of these percentages as a description of a small Canadian shop.

The direction still carries over. Once everyone has the tool, having it stops being an advantage. The businesses that pull ahead are the ones that can show what it did, decide what to expand and what to drop, and defend those choices to a partner, a lender or themselves. The rest are paying for something they can describe but not evaluate.

There's also a practical cost to not measuring that gets overlooked. When a business can't say what its existing AI tools have done, every new tool proposal becomes a matter of opinion. The enthusiastic person in the office wants to try something, the skeptical person doesn't, and neither can point to evidence from the last purchase. Measurement isn't just about proving value to someone else. It's what lets a business learn from its own decisions instead of relitigating them each time.

What to do with this

If you run a smaller manufacturing or trades operation, the practical lesson is less about AI than about bookkeeping for decisions. Before any tool gets more budget, someone should be able to say what it was supposed to change and whether it did. And the person answering should include whoever uses it every day, not only whoever approved it.

Setting up what to measure, before the money gets spent, is part of the AI Discovery and Readiness Assessment. It's a paid engagement with a defined scope: a discovery call, structured input from the people closest to the work, and a written report that puts your answer on paper from the shop floor up.

Explore the assessment →