By Paraic O’Lochliann, VP of eMaint, a Fluke Brand

Ask anyone who has moved an AI program from pilot to live environment, and the story tends to follow a similar shape. The technology does what it was selected to do – sensors feed clean data into models that generate useful predictions under controlled conditions. Full production introduces everything the pilot excluded: incomplete asset histories, SCADA feeds not built for interoperability, shift patterns that make consistent alert follow-through genuinely difficult. The operational infrastructure that AI depends on to create value turns out to be more variable, and more consequential, than the procurement decision accounted for.

That’s where the real test begins. Fluke Corporation’s research, conducted by Censuswide among more than 600 senior decision-makers and maintenance professionals in the US, UK and Germany, suggests expectations are already resetting. As organizations move from experimentation to execution, confidence in near-term Industry 5.0 timelines has fallen sharply. The pattern reflects a maturing understanding of what industrial AI actually runs on.

Manufacturers are funding the plumbing, not the novelty

What manufacturers are choosing to fund is the clearest signal in Fluke’s research. Two-thirds of organizations now direct 16–30% of their maintenance budgets to new technologies – but the bigger indicator is the four categories absorbing most of that spend: cybersecurity, data management, Generative AI and Industrial AI. Exploratory AI, which led the previous investment cycle, is no longer where the budget concentrates.

Each of those four categories sits close to operational risk. Cybersecurity defends operational continuity. Data management protects the quality of the decisions AI is asked to make, and Industrial AI protects uptime directly. None of them are bets on novelty; they’re investments in the machinery of running a plant well.

The pattern extends beyond manufacturing. Wharton’s 2025 AI adoption research describes the wider market as moving into “accountable acceleration,” with 72% of leaders formally measuring GenAI ROI through metrics such as productivity gains and incremental profit.

For manufacturers, that accountability is unforgiving. If AI can’t be trusted in the conditions where downtime, safety and throughput are on the line, the business case weakens quickly.

AI is reading records the humans were finishing in their heads

Industrial data has always been built around the people reading it. A maintenance log that says “checked, OK” works because the technician who wrote it knew what was checked and what OK meant. A vibration reading carries weight because the engineer remembers it ran high last winter too. Decades of operations have run successfully on records like these, with the missing context supplied silently by whoever was looking at them.

AI doesn’t supply that context. A model fed the same maintenance log sees two characters. It can’t tell whether “OK” meant healthy, deferred, or pending a part. It can’t tell whether last winter’s high reading was an outlier or a pattern. The data isn’t wrong, exactly. It’s thinner than it looked, and the thinning only becomes visible when something tries to use it without the human in the loop.

Cisco’s 2026 industrial AI research found that 61% of organizations are already using AI in live industrial operations, while only 20% report mature deployments at scale. Most of the distance between those numbers is the work of rebuilding decades of context into records the AI can actually use.

The constraint sits inside the organization

For leadership teams, the sharper question is whether the organization is mature enough for AI to change decisions consistently. Maturity shows up in the routines around the technology – how exceptions get reviewed, how knowledge travels between sites.

Fluke’s research shows why this can’t be handled as a systems issue alone. Around 78% of the reported barriers to progress are workforce and capability-related – not budget, not technology, not vendor support. The constraint sits inside the organization, in what people know and how teams work.

The manufacturers that gain the most from industrial AI will be those with enough discipline beneath the technology to make insight repeatable. Adoption can show momentum. Digital maturity decides whether that momentum becomes measurable performance.

GenAI writes the report. Industrial AI runs the plant.

Fluke’s data shows manufacturers investing in both Generative AI and Industrial AI, but those categories shouldn’t be treated as interchangeable. The distinction often gets blurred because most AI vendors have a commercial reason to keep it blurred – they sell into both categories and benefit from the conflation. Manufacturers end up making the call themselves, often after the procurement decision has already been made.

Generative AI can make reporting faster, improve search across manuals, draft work instructions or reduce admin around work orders. Useful gains, but they sit at a different distance from operational consequence. 

Industrial AI operates closer to the asset. It informs decisions about which asset gets attention next and when a shutdown is justified, which means the surrounding process has to be tighter. A vague answer in a report is one kind of problem. A weak signal in a production environment is another.

The distinction changes the productivity argument. In manufacturing, loss often comes from variability: a technician waiting for the right procedure, a planner working from partial history, a maintenance task performed differently across shifts, a critical judgment held by one experienced person. AI can reduce that friction only when the information it surfaces is trusted and tied to the work.

The practical test is simple: does AI shorten the distance between a signal and a sound decision?

The most valuable AI is the failure that didn’t happen

For industrial AI, proof often shows up quietly. A stoppage that doesn’t happen. A fault identified early enough to plan around. A technician reaching the right procedure before the repair window closes. A maintenance team acting from the same asset context across sites.

Connected reliability is the bridge between what operations need now and where Industry 5.0 is supposed to lead. The Fluke survey finds that nearly half of respondents plan to advance reliability initiatives in the next 12 months. The investment is the work that makes the rest of the AI agenda possible at all.

The point is prevention, not visibility for its own sake. Dashboards can show that risk exists, but the value comes when asset data and field action are connected closely enough to change the outcome. In that environment, AI has something useful to work with: a clearer record of what has happened, and the ability to act on what’s changing now.

That’s where the AI conversation is heading. Leaders are asking about exposure, continuity and defensibility because those are the terms that live operations understand. The manufacturers that move furthest will be the ones mature enough to make AI useful where the consequences are felt first – on the floor.