Insight
Adoption on the shop floor is mainstream now. Getting a real return from it is still rare, and the difference isn't the technology.
Insights / 51% of Manufacturers Use AI. 5.5% See the Returns.
Global manufacturing is under pressure from several directions at once: rising input and labour costs, tighter quality and traceability requirements, and a wave of Fourth Industrial Revolution technology that promises to resolve all three. Adoption has followed: over half of manufacturers now report using AI in some form, and nearly 80% report regular use of generative AI somewhere in the business. McKinsey's most recent data on engineering and manufacturing organisations puts the payoff in perspective, however: only 5.5% are seeing real financial returns from it, and fewer than three in ten product-development teams are using AI agents at all.
The pattern holds across McKinsey's broader research too: manufacturers embedding AI across a whole process report steadily higher gains than those running it in one isolated step. We believe buying the tool is the easy part. Redesigning the line, the maintenance schedule, or the inspection process around it is what actually shows up in the numbers, and it's the part most plants haven't done yet.
Audited results from sites that went past pilots, next to where most of the industry still sits. Figures shown as percentage change or adoption share.
| The upside available | ||
| Lead time reduction | 48% | |
| Cycle time reduction | 44% | |
| Product defect reduction | 41% | |
| Labour productivity gain | 40% | |
| Where most manufacturers are today | ||
| Manufacturers using AI in some form | 51% | |
| Product-development teams using AI agents | 27% | |
| Organisations seeing real ROI from AI | 5.5% | |
We believe the difference is rarely the model itself. Across the sites in the "upside" column above, the same handful of decisions recur.
Sites that redesign a process around AI, rather than inserting a model into the existing one, are the ones that show up in the upside column. Adding a tool to an unchanged line caps the gain at whatever that line already allowed.
Fewer than three in ten product-development teams use AI agents at all, even where generative AI is already in regular use elsewhere. Value comes from depth in a handful of processes, not breadth across dozens of pilots.
Predictive maintenance and quality-control models only compress cycle time and defects when they can reach the sensor, machine, and scheduling data those decisions depend on. Where that access stops at a single station, so does the gain.
A model's behaviour at commissioning and its behaviour a year into production are not the same question. Sites that name an owner for ongoing performance catch drift before it becomes a quality incident; sites that treat go-live as the finish line don't.
The sites furthest ahead didn't get there by adding a chatbot to the shop floor. They embedded AI-driven predictive maintenance, quality control, and workflow planning across the whole production system, and resourced it like a change to how the plant runs.
The number worth sitting with isn't the size of the opportunity. It's how few manufacturers have actually captured it, and how much of that gap comes down to the five decisions above rather than to which model was bought.
Sources: World Economic Forum & McKinsey, Global Lighthouse Network cohort reports, 2025; McKinsey & Company, State of AI 2025 survey of manufacturing and engineering organisations; National Association of Manufacturers, 2025 manufacturer survey.
Talk to us about where the gap actually is for you.
Get in touch →