A small problem on the shop floor is easy to hide. Put a plaster over the missed changeover, the recurring jam, the report that never quite matches what happened on the line, and move on to the next fire. The trouble with plasters is that they do not fix cracks, they cover them, and in a metal forming operation a covered crack has a habit of widening into a stoppage nobody saw coming.
Across the sector, that habit is getting harder to sustain. Product ranges keep growing, and every new variant adds another changeover and another way for a schedule to slip. Labour, energy and raw material costs keep climbing while competitive pressure makes them almost impossible to pass on.
Customer demand no longer arrives in predictable batches, and lead times keep shrinking rather than stretching. Put more inventory in and availability still falls. Put more hours in and output per person still declines. The instinct is to work harder. The organisations pulling ahead are making better decisions, not more of them.
That was the question at the centre of a webinar CBM ran with Kaizen Institute on 8 June: not whether data and AI can help metal forming operations, but where they genuinely add value and where they simply add cost. Presenting was Christopher Conlon O’Neill, from Kaizen Institute’s UK team, and his argument set the tone for everything that followed. The technology is rarely the problem. What surrounds it usually is.
When technology meets an unstable process
Manufacturers have spent the past decade investing heavily in ERP systems, MES platforms and, increasingly, AI tools, and the results have too often disappointed. The reason, O’Neill argued, is rarely the technology itself.
The first failure pattern is starting with the tool rather than the problem: a business buys an analytics platform before it has agreed what operational question the tool is meant to answer, and the conversation drifts from outcomes to features. The second is data nobody can trust: incomplete production records, inconsistent downtime coding, different versions of the truth sitting in different systems. O’Neill described one client with six separate systems, each holding a different slice of the same operation, with no link between them.
A business we had been working with recently had six different fragmented systems, all carrying separate pieces of data, but there was no real link between them. We had to go piece by piece and build a tool that could bring those systems together, just to get the data quality to a decent standard.
The third failure is no ownership: dashboards get built and reports get generated, but nobody is accountable for acting on what they show, so the tool gets used once and then quietly ignored. The fourth is treating AI as a black box. If operators cannot see how a recommendation was reached, they stop trusting it, and a system nobody trusts is a system nobody uses. The fifth is a tool that never touches the shift meeting or the daily routine where decisions actually get made, so behaviour never changes even though the technology works exactly as designed.
Apply technology to a process that is still unstable and none of this improves. It simply moves the waste from the shop floor onto a screen, faster and at a larger scale than before.
Stabilise first, then digitise
At the heart of KAIZEN™ is the principle of continuous improvement: creating lasting performance through small, structured changes made every day. Rather than relying on occasional large-scale transformation projects, organisations achieve sustainable results by systematically eliminating waste, standardising processes and empowering the people closest to the work to solve problems. This disciplined approach creates stable, repeatable processes and, importantly, generates the consistent, reliable data that digital technologies and AI depend upon to deliver meaningful business value.
O’Neill set out four areas where this translates into measurable performance. Quality, through more consistent, well-controlled work that reduces scrap and rework. Cost, through the systematic elimination of waste such as waiting time, excess inventory, unnecessary movement and overproduction that consume resources without adding value. Delivery, through fewer bottlenecks, improved planning and more predictable scheduling. Finally, People, by giving those closest to the process the tools, skills and authority to identify problems and implement improvements, because they are usually the first to recognise where change is needed.
Building this foundation is what makes it possible to introduce data analytics and AI successfully. “We don’t start with AI,” O’Neill told the session. “We start with stabilising the operation.” The logic runs one way: stable, standardised processes naturally generate consistent, trustworthy data. Reliable data allows organisations to identify genuine patterns rather than noise, enabling AI to produce accurate, actionable insights. Skip the first two steps, and even the most sophisticated technology has little of value to learn from.
Where AI actually earns its place
With stable processes and reliable data in place, AI can move beyond reporting problems to actively improving operational performance. O’Neill highlighted four areas where manufacturers are already seeing measurable returns. Demand forecasting and production planning is usually where manufacturers start analysing historical orders, seasonality and customer patterns to answer the two questions that actually run a plant, what to produce and when, with fewer late schedule changes and better material availability as a result.
The second is production throughput. Bottlenecks and changeover losses are common across metal forming, and the fix is not always to run the changeover faster. Sometimes the greater opportunity lies in reducing the number of changeovers altogether by sequencing production more effectively, rather than simply completing each changeover faster using SMED. Either way, the target is the same: more output from the equipment already on the floor.
Third is predictive maintenance. Most plants still run to a fixed schedule or wait for something to fail. Analysing machine conditions against a history of past breakdowns lets a team intervene before equipment fails rather than after it does. Fourth is quality: tracing which process conditions tend to precede a defect, so the correction happens at the source rather than at the inspection bench.
None of this sit in isolation, and that point matters more than any single use case. Improve visibility of demand and inventory, and production scheduling gets sharper as a result. Improve routing and dispatch planning, and the saving shows up as fewer missed delivery windows and a lower transport bill. Treat demand, production, delivery and the customer relationship as connected links in one value chain, and the gains compound. Treat them as separate projects, and each one delivers a fraction of what it could.
Teams that trust the data
None of this holds without the people reading the numbers. O’Neill was direct about it: even the most advanced platform delivers almost nothing if the team using it lacks the confidence to read it, question it and act on it.
Four capabilities matter most. Data literacy comes first, the ability to look at a dashboard and diagnose what it is actually saying, rather than simply noting that a number has moved. Kaizen Institute has a phrase for the discipline this builds: no data, no discussion, meaning a problem gets assessed against what actually happened on the line, not against whoever argues loudest in the meeting.
The second capability is genuine problem solving: seeing that performance has dropped is one thing, working out why is another, and the two get confused more often than they should. The third is operational understanding, knowing how material, information and decisions move through the plant, from customer order to shipped product, without which the same number can point a team toward the wrong fix entirely. The fourth is a habit of continuous improvement: reviewing performance, closing a problem, and doing the same again the next day rather than treating any single fix as finished business.
Put together, these four are what separate manufacturers that pull ahead from the ones working just as hard for less. As O’Neill put it, the businesses that outperform their competitors are not always the ones putting in more effort. More often, they are the ones making better decisions, faster, and doing so more consistently than the plant down the road.
Start where the evidence is
For a plant deciding where to begin, O’Neill’s advice avoided any mention of a platform. Start by making the data that already exists visible. Most metal forming businesses have plenty of it, scattered across systems that do not talk to each other, and the first task is pulling the numbers that matter into a single place rather than adding another report to an inbox nobody reads.
From there, use that visibility to find the costliest recurring problem, the highest downtime, the changeover eating the most capacity, and put a figure on what solving it is worth. That figure prioritises where the effort goes and gives a business case something firmer to stand on than instinct. Only then pilot a solution, on one line or one product family, measure it against a baseline, and scale only what the results support.
“Technology cannot fix broken processes,” O’Neill said near the end of the session. “It can only amplify them.”
That is the line worth carrying back to your own operation. Before investing in the next AI platform or analytics tool, manufacturers should ask a simpler question: are the processes generating the data stable enough to support better decisions? Technology can accelerate improvement, but only when it is built on reliable processes, trustworthy data and a culture of continuous improvement. Without that foundation, digital transformation risks becoming an expensive way of automating existing problems.
This is the same logic Kaizen Institute has applied on shop floors for decades, just extended to a newer resource: data. Stabilize the process, and the numbers become trustworthy enough to act on — which is exactly where our AI Solutions for Manufacturing practice picks up, turning that stability into predictive maintenance, sharper demand forecasting, and quality control that catches problems at the source.