Manufacturers are putting AI on the factory floor. The decisions that move the business are still made the old way.
Predictive maintenance, vision inspection, robotics: AI is treated as a plant project. Meanwhile the capital and planning calls that actually move a manufacturing business get a generic answer or no AI at all.
Request a Strategic Briefing →Ask most manufacturers where AI lives and they will point at the shop floor: predictive maintenance, computer-vision quality control, robotics, a plant and IT capex line. That is real work, and it is worth doing. The trouble is that the decisions that actually move a manufacturing business, capital allocation across plants, planning under volatility, make-versus-buy, the footprint and capacity bets, still get made the old way or handed to a commodity AI that answers with confidence and no knowledge of your business. Pilots multiply on the floor. Scaled value does not. That is High-Speed Waste in capex form. Gildoni installs the Havruta Methodology (formerly the Think Partner Methodology) into how operations leadership reasons with AI, at the altitude where the planning and capital judgement is actually made.
AI is everywhere on the floor and barely anywhere in the planning
The industry's instinct is to scope AI to the plant. It is the most visible, most measurable place to start, and the vendor market is built to sell there. So that is where the pilots land.
The honest difficulty is that a shop-floor pilot is easy to launch and hard to scale into something the whole business feels. The technology is maturing fast, but the value keeps stalling at the edge of the plant, while the decisions one altitude up carry on without it.
The World Economic Forum, tracking its Global Lighthouse Network of leading production sites, reports that AI and generative AI now enable up to half of the top use cases the best manufacturers deploy. The capability is no longer the bottleneck. The reasoning around it is.
The numbers show the same story from three angles.
of manufacturers run AI and machine learning at the facility or network level, while roughly the same share are still only piloting it. Adoption is real, but it is stuck at the edge.
of supply-chain organisations apply AI incrementally or scale it gradually into existing processes, while only 17% pursue a real redesign of how decisions get made.
of the top use cases at the world's leading production sites are now AI or generative-AI enabled. The frontier has the capability. Most of the industry has the gap.
The pilots are not the problem. The altitude they stop at is.
Global supply chains, once optimised for cost and scale, are now redefined by proximity, risk, and resilience.
Why the value stops at the edge of the plant
If the capability is there, why does the value stall? Three reasons.
AI gets scoped as a plant and IT project, so it inherits a plant and IT remit. A predictive-maintenance model can shave downtime on one line. It was never meant to weigh in on whether to add a line, move it, or close it. The remit it was given is the ceiling on the value it can return.
The decisions that actually move a manufacturing business live above the floor. Capital allocation across plants, planning and supply judgement under volatility, make-versus-buy, the footprint and capacity bet. These are bets under uncertainty, made by people, and they are where AI is least present and most needed.
And where AI does reach those calls, it tends to arrive as a vending machine. Put a capacity question to a commodity model and it returns a fluent, confident answer that knows nothing about your plants, your contracts or your demand signal. Pilots multiply because they are cheap to start. Scaled value does not follow, because nobody installed the reasoning that turns an AI answer into a defensible decision.
It is worth being honest about the older truth underneath this. Capital discipline and planning judgement under uncertainty are old skills. AI has not replaced them. It has stress-tested them, and found most manufacturers deploying at production speed on the floor while still deciding at committee speed about the business.
Scoping AI to the shop floor is the easy part
Here is what the plant-first instinct leaves out. Every vendor and every roadmap now says the same thing: deploy AI on the line, measure the downtime, scale the pilot. Almost none of them say how an operations leadership team actually reasons through the capital and planning decisions that sit above the line, the ones where the real money moves.
That is the real gap, and it is not a tooling gap. A factory full of working pilots can sit directly beneath a footprint decision still made on a spreadsheet and a hunch. At the altitude where the bets are largest, that is the Mirror Principle at its most expensive: if the reasoning going into a capacity bet was generic, the bet coming out is generic, however much AI ran on the floor below it. A predictive model can tell you a bearing is about to fail. It cannot make the thinking behind a make-versus-buy call any good. That is a different discipline.
AI pilots cluster low, on the shop floor, which is where most manufacturers stop. The decisions that move the business, capital allocation and planning judgement under volatility, sit at the operations-leadership altitude above. The red is the distance between where AI is piloted and where the value actually lives: automated on the floor, ungoverned in the planning.
What the Havruta Methodology installs at operations-leadership level
The Havruta Methodology is that discipline. It changes the default behaviour of the machine from agreeing with you to reasoning with you, which is exactly what a capital or planning bet needs.
The Flip
The Flip puts the machine on the other side of the question. Instead of confirming the capacity case, it argues against it: where is this demand forecast soft, what is the footprint plan not pricing in, what would have to be true for this bet to be wrong. The leadership team gets challenged before the market does the challenging.
Ground Truth
Ground Truth keeps the reasoning anchored in the real business, its actual plants, contracts, lead times and demand signal, rather than the generic planning language an AI produces by default. A capital decision built on a plausible average is worse than no AI at all.
Decision Velocity
And Decision Velocity lets the team decide at the speed the volatility moves, compressing the path from question to defensible plan without handing the judgement to the machine.
The fuller account of how all of this works is on the methodology page.
What this is not
This is not factory automation and it is not a piece of plant software. It is not predictive-maintenance tooling, an industrial IoT platform, a vision-inspection system, or an MES. It is not AI training or general AI literacy. The tooling and the platforms are a separate market. This is the thinking underneath the decisions they never reach.
It changes how operations leadership reasons about the decisions it already owns: the capital allocation, the capacity bet, the planning call under volatility, the make-versus-buy.
Where an operations leadership team starts
The methodology is installed along a ladder, and a leadership team enters at the rung that fits.
Most begin with the Eye-Opener Workshop, a half-day in which the team sees the shift on its own real work.
A leadership group embeds the practice through the Havruta programme, taking the discipline across the team.
A single high-stakes question, a capacity bet, a footprint call, a make-versus-buy decision, can be worked through Advisory Havruta.
How a COO and operations leadership reason with AI
For the operations leaders specifically, the role page takes the same discipline to the role that owns the planning and the plants day to day. A Strategic Briefing is how to decide where to begin.
Go to the COO pageLeadership questions about AI in manufacturing
Is AI in manufacturing a shop-floor project or a leadership decision?
It is both, but most manufacturers only treat it as the first. Predictive maintenance, vision inspection and robotics are genuine shop-floor work and worth doing. The decisions that actually move the business, though, capital allocation across plants, planning under volatility, the capacity bet, the make-versus-buy, live above the floor. Scoping AI only to the plant leaves those calls to be made the old way or handed to a generic answer. The value you are missing sits at the leadership altitude, not on the line.
Why do manufacturing AI pilots multiply but rarely scale?
Because a floor pilot is cheap to launch and was never scoped to reach the decisions that carry the real money. Gartner's data shows most supply-chain organisations applying AI incrementally rather than redesigning how decisions get made. A pilot can cut downtime on one line, but it cannot improve a footprint bet sitting one altitude up. Scaled value follows only when the reasoning at the planning and capital level changes too, and that is a discipline, not another tool.
Isn't predictive maintenance the highest-value place to use AI?
It is high value and clearly worth doing, but it is bounded. Predictive maintenance optimises an asset you have already decided to run. It cannot tell you whether to add capacity, move a line, or close a plant. Those capital and planning calls move far more money than downtime ever will, and they are exactly where AI is least present today. The biggest return is not a better model on the floor. It is better reasoning where the bets are largest.
Is this factory automation or a plant software platform?
No. It is not predictive-maintenance tooling, an industrial IoT platform, a vision system or an MES, and it does not touch your plant stack. Those address the floor. This addresses the thinking above it: how an operations leadership team reasons through a capital or planning decision so the answer is genuinely theirs, anchored in their real plants and demand signal, and stress-tested before the market tests it for them.
How should operations leadership use AI for capital and planning decisions?
Start by treating AI as a thinking partner for the decision, not a vending machine for an answer. Make the model argue against the capacity case rather than confirm it. Anchor it in your real plants, contracts and demand signal rather than generic planning language. Then decide at the speed the volatility moves, without handing the judgement to the machine. The floor pilots tell you what is happening on the line. This is how you reason about what to do with the business.
Where should we start?
With a Strategic Briefing, or with the Eye-Opener Workshop, where an operations leadership team sees the difference between instructing AI and reasoning with it on its own real work. From there the path depends on whether you are embedding the practice across the leadership team or working a single high-stakes question, such as a capacity bet, a footprint call, or a make-versus-buy decision.