If you're still asking whether manufacturers are using AI, you're about a year behind the question that matters. A survey published in January says nearly all of them are, in some form. The more useful questions are what kind of AI they're using, where they're putting real money, and what's stopping them from getting it past the pilot stage.
The source, and what it does and doesn't cover
Rootstock Software published its 2026 State of Manufacturing Technology Survey on January 28, 2026. The research was fielded by Researchscape, an independent research firm, across 520 manufacturing leaders responsible for digital transformation at mid- to large-sized manufacturers in North America, Europe and Asia.
Three caveats before the numbers.
The sample is mid- to large-sized manufacturers globally. It isn't a survey of small fabrication shops or Okanagan manufacturers, so what follows describes the direction manufacturing is moving, not what your shop already looks like.
The data was published in late January. It's the most current manufacturing-specific data of its kind this year, not breaking news.
Rootstock sells ERP software with AI features, so it has a commercial interest in these numbers looking strong. Researchscape ran the fieldwork and the sample size and method are disclosed, but it's worth knowing who commissioned the study.
Adoption is settled. The type of AI is what's changing.
According to the survey, 94% of manufacturers now use AI in some form. At that level, adoption stops being an interesting number. Almost everyone answers yes.
The split that matters now is which kind. Predictive AI, the kind that forecasts demand or flags a piece of equipment before it fails, jumped 12 points this year to 48% of manufacturers. That was the largest gain of anything in the survey.
That's a meaningful change in what the software is being trusted to do. A tool that summarizes an email or drafts a quote is an assistant. A tool that tells you a spindle bearing is likely to fail next week, or that demand for a product line is about to drop, is making a call with consequences on your floor. Whether anyone acts on that call is a different question, and it comes up again below.
Where the budget went
The investment numbers are the most practical part of the survey, because money tends to follow what people expect to work.
AI investment among manufacturers shifted most sharply toward two areas:
- Supply chain planning, up 19 points to 35% of manufacturers.
- Process optimization, up 11 points to 36%.
Neither of those is glamorous. Neither one is a chatbot on the company website. Supply chain planning means fewer surprise material shortages and fewer rush orders at full price. Process optimization means catching waste on the line before it eats a shift's margin.
The survey doesn't say why the money moved there, so I won't claim to know. What can be said is that both areas share some useful traits. They deal with data most manufacturers already collect, they affect costs that show up on a monthly statement, and a change in either one is relatively easy to measure. That's a reasonable profile for a first serious investment, and very different from a general-purpose AI rollout with no defined target.
What that looks like at a smaller scale
A small manufacturer doesn't need a supply chain planning platform to recognize the underlying problem. If you've paid expedite fees because a component ran out, or held excess inventory because nobody trusted the reorder points, the problem is already costing you. The same goes for scrap and rework that nobody tracks closely enough to see a trend.
Those are the kinds of problems where a predictive tool can help, but only if the data behind it is reasonably clean, which leads to the part of the survey that gets less attention.
The barrier isn't the software
When asked what's holding back their AI efforts from scaling, manufacturers didn't point at the technology first. Thirty-three percent named a lack of the right talent as their biggest obstacle. Thirty-one percent pointed to poor collaboration across departments.
Not the tool. Not the vendor. The people running it and the structure around them.
That matches the way these projects tend to fail. A predictive maintenance alert doesn't help if nobody on the floor trusts it enough to act on it. A demand forecast doesn't help if operations and purchasing aren't working from the same numbers. And neither helps if operations and IT never agreed on what the tool was supposed to do in the first place.
Buying the license is the easy part. Building the team and the process around it is what decides whether it works.
For a smaller manufacturer, the talent barrier looks different but lands in the same place. You probably don't have a data team. You may have one person who's good with spreadsheets and already has a full-time job. Any AI plan that assumes a specialist will appear to run it is a plan built on hiring, not on technology.
Spending more in a year they expect to be rough
One of the more interesting findings is how cautious manufacturers were about the year ahead, and what they planned to do anyway.
Thirty-one percent expected customer demand to decrease, against 19% who expected it to grow. Thirty-nine percent expected tariffs to raise the cost of raw materials or components. That's a gloomy outlook by any standard.
Despite that, 61% planned to increase spending on enterprise software over the following 12 months.
That isn't as contradictory as it sounds. When demand is uncertain and input costs are moving, the businesses that come out ahead are usually the ones that see a problem sooner and react faster. Shorter time between a signal showing up and someone acting on it is worth more in a volatile year than in a calm one.
There's a risk in that logic too. Pressure is a good reason to look for an advantage and a poor reason to skip the step where you decide what the advantage is supposed to be. Software bought in a hurry to answer a cost squeeze often ends up as another subscription nobody can evaluate when the squeeze eases.
Why consolidation keeps coming up
The highest-rated answer in the entire study wasn't about AI directly. Forty-nine percent of manufacturers named simplifying infrastructure and standardizing platforms as the outcome they want most from their ERP.
That connects straight back to predictive AI. Forecasting tools pull from inventory, orders and production data at the same time. When those live in separate systems that don't sync, the model is working from a partial picture, and the forecast is only as good as the gaps allow. If your systems don't talk to each other, your AI is guessing.
Most small and mid-sized manufacturers know this problem well. Quotes live in one place, the production schedule in a whiteboard or spreadsheet, inventory in the accounting system, and quality records in a binder. Each of those works on its own. Together they make it very hard for any tool, AI or otherwise, to see the whole operation.
Everyone thinks they're keeping up
One more number from the same survey: 73% of manufacturers believe they're on par with or ahead of their peers on AI.
That figure can't be literally true for everyone who said it, but it tells you something about the mood. AI use in manufacturing has moved past a handful of early movers. Most manufacturers now consider it normal, and a large share believe they're doing fine. Some of them are. Others are counting a few licenses as progress without anything measurable to show for it.
For a smaller shop, the honest starting point is often less about AI than about getting quotes, schedules, inventory and quality records into one place, or at least into systems that can share data. That work is unglamorous and rarely gets a vendor demo. It's also the step that decides whether any future forecasting or optimization tool has something reliable to work from.
Questions before any predictive tool
Predictive AI was the fastest-growing category in this survey, and it's the kind most likely to be pitched to manufacturers over the next year. Before taking a demo seriously, a few questions are worth answering internally.
- What decision would this forecast or alert change, and who makes that decision today?
- How many years of reasonably clean history do we have for the thing being predicted, whether that's orders, machine downtime or scrap?
- If the tool flags a problem at two in the afternoon, who acts on it, and do they have the authority to stop a job or change an order?
- How will we know in six months whether the predictions were any good?
The third question is the one most often skipped, and it's where the survey's collaboration barrier shows up in practice. A maintenance prediction that lands in an inbox nobody owns produces nothing. A demand forecast that purchasing doesn't trust produces a second set of spreadsheets. The tool can be accurate and still fail if nobody on the floor is set up to use it.
Answering those questions honestly usually takes an afternoon and a few uncomfortable conversations between departments. It's still far cheaper than discovering the answers after the contract is signed.
What this means for a smaller manufacturer
A global survey of mid- to large manufacturers doesn't tell you what to do in a 25-person shop in the Okanagan. I'd treat the percentages as a picture of where the larger part of the industry is heading, not a benchmark to measure yourself against.
A few things carry over at any size, though.
- The money is moving toward specific operational problems with measurable costs, not toward AI in general.
- The hard part is the people and the cross-department process, not the purchase.
- Disconnected systems limit what any predictive tool can do, so data housekeeping comes before forecasting.
None of that requires a large budget to act on. It requires knowing which operational problem is actually costing you the most, and whether your data and your team are ready to support a tool aimed at it. If you want more background on where AI has already paid off in a manufacturing setting, this piece on engineering documentation walks through one example, and the manufacturing page covers the broader picture.
Whether your operation is ready to hand AI a decision with real consequences, or is still at the experiment stage, is what the paid AI Discovery and Readiness Assessment works out. It looks at your team, workflows and systems before it looks at any software, and it ends with a scored report on where the blockers actually are.