Most factories still can't answer "how did we do yesterday?" — and AI won't fix that

Walk into a North American manufacturing plant on a Tuesday morning and you will find a ritual you would also have found in 2006: a supervisor at a whiteboard, a stack of shift-handover sheets, and yesterday's production numbers being written in dry-erase marker while the day shift stands in a semicircle with coffee.

The board says: Target 4,200. Actual 3,780. Reason: "material issue."

That is the measurement layer underneath a $200-million plant. It is also the layer we are now being told to put artificial intelligence on top of.

The state of measurement on the floor

The gap between what a leadership team thinks is measured and what is actually captured is the first thing anyone doing plant-floor work learns. It is not a technology gap. It is a discipline gap that persists because it is quietly expensive but rarely urgent.

Four questions most plants cannot answer within 60 seconds:

1. What was our OEE by line, by shift, yesterday?
2. What were the top 3 downtime causes this week, ranked by minutes lost?
3. What is our first-pass yield trend over the last 8 weeks?
4. What did we spend on emergency maintenance last month versus planned?

If those answers require someone to "pull a report," the KPI layer is not built. It is being assembled on demand, from scratch, every time someone asks.

Every predictive-maintenance model, every AI scheduling optimizer, every generative-AI copilot for operators reasons over the data your KPI layer produces. Manual, retroactive, gap-riddled data does not just limit AI. It makes AI dangerous — because it produces confident answers built on incomplete inputs.

What North America is actually doing

The reshoring wave is real. The physical capital is moving. The digital-maturity gap underneath it is bigger than most of the reporting admits.

The tailwind is unambiguous. The US Inflation Reduction Act and CHIPS & Science Act have anchored more than $500 billion in committed manufacturing and clean-energy investment since 2022. Canada's Strategic Innovation Fund and Net-Zero Accelerator have committed billions more to advanced manufacturing, battery, and cleantech capacity. The Reshoring Initiative recorded another strong year in 2024 for announced US manufacturing job returns.

The gap underneath it is also unambiguous. Deloitte's US Manufacturing Outlook has, for several years running, flagged workforce, digital-maturity, and data-integration as the top barriers to competitive scale — ahead of capital access. The National Association of Manufacturers' quarterly survey consistently shows the majority of small and mid-sized US manufacturers rate their own digital maturity as "beginning" or "developing," not "connected." Statistics Canada and BDC data show adoption of advanced manufacturing technologies in Canada is heavily concentrated in the largest 10 per cent of firms.

The practical translation: new plants are going up, older plants are being retooled, and AI initiatives are being announced on top of a base where most of the shop floor is not yet instrumented. The physical investment is racing ahead of the measurement investment.

Where poor KPI foundations show up — three different desks

Where Industry 4.0 actually stands, with AI on top

Industry 4.0 has been a decade-long conversation. The generative-AI overlay is new. What works and what does not is finally visible enough to talk about honestly.

The pilot-purgatory problem is real and quantified. McKinsey's global Industry 4.0 research has, for multiple years, found roughly 70 per cent of digital manufacturing transformations stall at the pilot stage. BCG's manufacturing surveys consistently find fewer than 30 per cent of AI and analytics pilots reach full production scale. The most common failure mode is not the technology. It is that the pilot proved a point on a clean, curated subset of data that does not exist across the rest of the plant.

The pattern that is working is remarkably consistent — and it is stubbornly sequential:

The pattern that is not working is buying an AI platform first and expecting it to sort out the mess underneath. It cannot. It will confidently generate outputs, but those outputs will inherit every gap and inconsistency in the data that trained it — and the plant will spend the next two years arguing about whether the model is wrong or the data is wrong. Usually it is both.

How KPI-ready is your operation?

A brief self-assessment. Nothing tracked, nothing saved.

What to watch next

The technology to fix the KPI layer is not the bottleneck. It has not been for a decade. The bottleneck is organisational: who owns each number, how it is defined, and whether the plant reviews it before it needs it — not after. In a country pouring hundreds of billions into new manufacturing capacity, the plants that will win the next five years are the ones doing the boring foundational work while everyone else is buying the AI platform.

Industry 4.0 without KPI clarity is a spreadsheet with better colours. AI without KPI clarity is confident hallucination at plant scale. Neither is the future anyone was sold — and neither will be, until the layer underneath is built.