Smart Manufacturing Trends: Four Years, One Stuck Number

Estimated reading time: 12 minutes

I have read four editions of the Rockwell State of Smart Manufacturing Report back to back, and I keep coming back to the same number. In 2024, manufacturers said only 44% of the data they collected got used well. Two years later, in a report branded the “execution era”, that figure sits at 43%. Almost nothing else stayed still. Generative AI arrived and then got rebranded as agentic AI. Cyber risk climbed the obstacle list. AI moved from pilot lines to production at scale. The one number that should have moved if any of that mattered, the share of data actually changing a decision, barely shifted.

That is the story most coverage of this report misses. Everyone covers the new headline. Almost nobody asks why the same constraint keeps showing up underneath it, year after year, regardless of which technology is fashionable. I think that question is more useful to a COO than this year’s headline will ever be, and this piece is my case for why.

The Reality

You do not get the luxury of treating each year’s report as a fresh start. The pressures arrive on top of each other, not in sequence. One quarter it is energy costs. The next it is a cyber incident, or a key supplier missing a delivery window, while last quarter’s problem is still only half solved.

For three years running, inflation and slow growth held the top spot on the external obstacle list, a single villain everyone could point to. By 2026, that one big threat had splintered into a cluster. Supply costs, raw material swings, energy bills, cyber risk, and a shortage of staff now sit within a few points of each other. I read that shift as more troubling than the survey language suggests. A single dominant threat at least gives an organisation something to rally around. A flat field of five roughly equal threats does not. It just means firefighting in five directions with the same headcount as last year, and nobody outside the building can tell you which fire to prioritise.

Budgets back this up. Technology spend climbed from 23% of operating budgets in 2023 to 30% in 2024, then settled back to 28% by 2026. My read on that plateau is not retreat. It is the bill changing shape: money shifting from proving an idea works in one pilot cell to running it at scale, which costs differently and less visibly. Licences, integration, and training across more lines rarely make as dramatic a budget line as the original pilot did. If your finance team is asking why technology spend has flatlined despite three more years of AI headlines, that is almost certainly why.

Insight: the report changes its headline every year. The constraint underneath doesn’t.

Here is the bit most readers skip past. Each edition sells a new lead story. 2024 was generative AI. 2025 was cyber risk’s jump up the threat list. 2026 is the “execution era,” AI graduating from pilot to production at scale. Each story is real on its own terms. None is the deepest one in the data, and I think four years of headline-chasing has left most leaders better informed and no closer to fixing what actually holds their plant back.

The deepest pattern is this. Manufacturers collect more data every year. The share they use well has stuck near the mid-forties since anyone started measuring it carefully. More sensors. More dashboards. Roughly the same small share of it ever reaching a decision. If a number survives three rounds of “this is the year everything changes” without moving, that is not coincidence. That is evidence the industry has been solving the wrong layer of the problem.

Skills shortages tell a near-identical story from a different angle. Manufacturers named a lack of skilled staff their top competitive weak point in 2023, and named it again in almost the same words two editions later in 2025. Two survey cycles, two leadership cohorts answering independently, the same diagnosis. That is a structural weak point the industry has named twice and acted on once, if that.

So where does the bottleneck actually sit?

My hypothesis, and I want to flag it honestly as a hypothesis rather than something the data proves outright, is that these two gaps persist because money and attention keep going to tooling, not to who is allowed to act on what the tooling produces. You can license an AI platform in a quarter. You cannot install decision rights and a habit of writing down the reasoning behind a call anywhere near that fast. Those things are cultural, not contractual. The test is simple: two plants on the same software with different decision owners will get different results. If you have watched the opposite happen, that is worth knowing too, and it probably means your bottleneck really was the tool.

One constant has held without exception across all four editions, and it is the one I would put in front of any board still nervous about automation: more automation has never meant fewer people needed. It has meant different people, doing different work. Nearly nine in ten manufacturers planned to maintain or grow headcount even while scaling automation in 2023. If your workforce plan still assumes headcount falls as AI scales, four years of data say you are planning against the evidence.

The contrarian read

The consensus reading of this report series, the one most vendor webinars push, is that manufacturing is on an AI adoption curve and the job of leadership is to climb it faster than competitors. I think that framing is mostly marketing, and it has done leaders a disservice for four straight years. The curve framing implies the constraint is adoption speed. The data says the constraint is something closer to organisational plumbing: who owns a decision, how fast they can act on what the data tells them, and whether the people closest to the work trust the number enough to act on it without checking it twice. Climbing the adoption curve faster while that plumbing stays broken just means you generate more dashboards nobody reads, faster. That is not progress. It is expensive busywork with a better demo.

Solution lens: how to read next year’s report without chasing the headline

Example

A mid-sized component plant read the 2024 report and saw generative AI ranked as the top new priority. It ran a pilot summarising maintenance logs with it. Six months in, nothing on the floor had changed. The operations director eventually worked out why: the pilot had no owner. The AI-generated summaries went into a shared folder nobody was accountable for reading, a familiar pattern dressed up in newer technology.

She picked one decision instead: whether a flagged machine ran another shift before inspection. She named the shift supervisor as owner, no committee, no sign-off chain. The same maintenance data now fed straight into that single recurring call. Within a quarter, unplanned downtime on that line had dropped measurably. The tool had not changed. The decision rights had. The software was identical before and after. The only variable that moved was who was allowed to act on it.

Where leaders keep getting this wrong, in my experience reading four years of this data

Chasing each year’s headline stat is the obvious mistake, and the easiest to spot in hindsight. Assuming more data collected automatically means more decisions improve is a quieter one, and four years of evidence says that assumption is false. Treating the skills gap as solved once one cohort gets trained is another, when the data treats it as a permanent constraint, not a project with an end date. Benchmarking against the survey’s largest respondents is a mistake I see constantly: more than half report revenues above one billion dollars, and your eighty million dollar plant is not playing the same game. Waiting for next year’s report to validate action wastes a year you did not need to lose. And buying a new platform before naming the decision it should change is, in my view, the single most expensive habit in the industry: the launch happens, and the question of what should change afterwards never quite gets asked out loud.

What I would want to see instead is straightforward. A named owner who can tell you, without checking, the last decision they made with the new tool and what changed because of it. Data effectiveness treated as a live, weekly talking point rather than a once-a-year survey answer that gets filed and forgotten. Skills planning sitting inside the operating budget, not parked in a separate HR project. Reskilling on the same roadmap as new technology, not bolted on once someone notices nobody can run it. Cyber, supply chain, and workforce risk reviewed together in one room, not three meetings with three owners who never compare notes. And last year’s headline trend either earning its place on merit, or quietly retired, rather than sitting in limbo because nobody wanted to be the one who killed it.

Weekly signals

Track these for thirty days before deciding whether next year’s headline trend is worth chasing any further than a conversation.

• Time-to-decision for the one repeat call you named in the solution lens above.
• Share of flagged issues that get an owner’s response within the agreed time-box.
• Escalations logged after the time-box ends, and who they actually reach.
• Scrap or rework tied directly to the decision you are tracking.
• Number of people on shift who can explain why a call was made, not just what was decided.
• Data effectiveness, asked plainly in your own next team meeting: of everything we collected this week, how much did anyone actually use?

Close

Trend reports earn their keep by telling you what the rest of the industry is reacting to this year. They are far less good at telling you what has quietly stayed broken underneath, year after year, while the headlines moved on without it. Four editions of this one suggest the industry has spent four years getting better at reacting and no better at fixing the layer beneath the reaction.

So here is the one worth sitting with, and I would genuinely rather hear your answer than give you mine: which number in your own operation has stayed stuck for four years while everything around it kept changing, and what would it actually take to finally move it?

• Digital Building Blocks of a Sustainability Agenda
• Latest Insights
• Our predictions for 2024 priorities for manufactures…
• Decision latency in manufacturing: the hidden cost of slow decisions in 2026
• Every manufacturing plant needs a custom AI system

---

What do we mean by “Smart Manufacturing Trends”?

Frequently Asked Questions

---

Sources

• Rockwell State of Smart Manufacturing Report 2023 (8th Annual, Rockwell Automation, Publication INFO-BR027B-EN-P, March 2023)
• Rockwell State of Smart Manufacturing Report 2024 (9th Annual, Rockwell Automation, Publication INFO-BR027C-EN-P, March 2024)
• Rockwell 2025 State of Smart Manufacturing Report.pdf (10th Annual, Rockwell Automation, Publication INFO-BR027D-EN-P, June 2025)
• Rockwell State of Smart Manufacturing Report 2026 (11th Annual, Rockwell Automation, Publication INFO-BR027E-EN-P, May 2026)