The most efficient line can make the factory worse
A production line can hit every local target and still leave the factory performing worse. Higher utilization may create excess work in progress, a faster machine may overwhelm the next process, and a larger batch may improve unit cost while extending lead times and tying up inventory. The problem is not that the metric is wrong. It is that the boundary around the metric is too narrow.
Manufacturers have become highly skilled at improving individual assets and departments, but the modern factory behaves as a network of dependencies. Production, maintenance, quality, logistics and supply chain decisions continually reshape one another, often faster than conventional planning cycles can respond. Optimization only creates value when it improves the performance of the whole operating system rather than one isolated part of it.
Local success can create system failure
The most obvious example is throughput. A team rewarded for maximizing the output of one machine will naturally keep it running, even when downstream capacity is constrained or the materials it consumes are needed for a more urgent order. The local result looks efficient, while queues lengthen and the wider schedule becomes harder to recover.
Tony Smith, Solutions Engineer at Simul8, says simulation often creates an “aha moment” because it makes those trade-offs visible to the people responsible for different parts of the process. “People are departmentally focused, or their own KPIs are what matter for them,” he explains. “When you bring them together in a room to create a simulation, they start seeing that by maximizing their throughput in one area, they put strain on other parts of the business. They may be starving upstream processes or overloading downstream processes.”
That perspective changes the purpose of modeling. Historical dashboards can show that a bottleneck existed, but they cannot reliably explain what will happen when demand, staffing, machine availability or material supply changes tomorrow. A simulation allows the manufacturer to test alternative decisions before committing the factory to them.
“Simulation is not to be used to analyze what happened in the past,” Smith says. “It is to predict what is going to happen under certain conditions in the future.” That can mean deciding whether to repair a failed machine immediately, slow an upstream process, reroute work or switch the schedule to another product while a material shortage is resolved.
The value lies less in finding one perfect answer than in exposing consequences that are otherwise distributed across several functions. A proposed automation investment may meet its own specification and still fail to increase finished output because the surrounding process cannot absorb the additional capacity. Spare capacity can also be valuable when it protects the operation from normal variability, even if a line-level dashboard labels it underutilized.
One event can look like several problems
The same fragmentation appears in software. Manufacturers frequently buy a separate application for OEE, another for scrap, another for quality and another for maintenance. Each system may identify a legitimate performance issue, but none necessarily understands that the same production event sits behind all of them.
Leonor Marques, Architecture & Advocacy Director at Critical Manufacturing, argues that this is where manufacturing context becomes decisive. “Imagine that you invest in a tool to manage OEE and another tool to manage material scrap or scrap reasons,” she says. “The same event on the production shop floor can affect OEE and scrap yield. If you do not know the same event is behind it, you will look at two different problems, find two different root causes and try to solve them separately instead of taking the situation as a whole.”
A modern MES can provide the missing relationship between the process, product, machine state and order being executed. Temperature, vibration or cycle-time data has limited meaning until the operation knows what was being produced, which configuration was active and how the event affected quality or schedule performance. Without that context, an optimization engine may recommend the right action for the wrong situation.
Marques is particularly cautious about insight without execution. “The point is not to take insights. It is to act on them,” she says. “If you take the wrong insights, it is worse than having no insights, because you will take wrong decisions based on wrong data.” Her argument is not that every decision should be automated, but that the architecture must preserve enough context for a recommendation to be trusted and applied at operating speed.
This also explains why local optimization becomes more dangerous in multisite organizations. Two plants may produce the same product using different systems, data structures and process definitions. Comparing their OEE or yield without understanding those differences can create false confidence and encourage leaders to transfer a practice that does not fit the receiving site.
The factory extends into the supply chain
Even a perfectly coordinated plant can be undermined by decisions outside its walls. A production schedule depends on materials, transport, supplier responsiveness and customer priorities. Improving one line without considering those constraints can increase instability rather than remove it.
Will Dutton, Director Supply Chain Solutions at UiPath, describes supply chain management as “a global optimization problem.” He argues that manufacturers have spent too much effort building comprehensive visibility layers without first deciding which behaviors need to change. “If we are not pulling any levers or changing any behavior, what is it doing?” he asks. “Transparency is valuable, but organizations should be clearer about how these projects drive behavioral changes and what levers they are pulling.”
One UiPath customer illustrates the difference between planning and response. When a customer requests a different paint shade or specification, the manufacturer must determine whether the production line can be reconfigured in time, what the change will cost and which constraints it will break. Simulation can assess the supply chain in seconds, after which agents can communicate the request to suppliers and bring their responses back into the decision.
That execution layer matters because many operational delays occur in handoffs rather than machines. Employees extract information from emails, re-enter it into transactional systems, request approval and then contact suppliers or customers. Dutton calls this “people as middleware.” Automating those exchanges can reduce latency, but only if the planning logic reflects the wider business objective rather than accelerating a poor local decision.
Maggie Slowik, Global Industry Director for Manufacturing at IFS, sees the same issue in the isolated systems surrounding production. Manufacturers may have separate demand planning, scheduling, warehouse and transport tools, each optimized for its own purpose. The organization then struggles to connect decisions across the product and operational lifecycle.
“Data exists everywhere, in different carriers, TMS portals, systems and spreadsheets,” Slowik says. “If you had better control of this data, you could start orchestrating some of the decisions you are making. For me, it is the fact that the data is isolated that puts people into strategic paralysis.”
Better optimization begins with a wider question
Technology can connect the factory, model alternatives and automate execution, but it cannot determine the objective by itself. Manufacturers still must decide whether the priority is throughput, margin, delivery performance, working capital, resilience or a combination that changes with market conditions.
That choice should shape the hierarchy of metrics. OEE, cycle time and utilization remain useful, but they must sit beneath factory-level outcomes rather than become ends in themselves. A temporary reduction in utilization may be the right decision when it protects a critical delivery, prevents excess inventory or allows maintenance before a longer failure occurs.
The people closest to the process are essential to defining those trade-offs. Slowik rejects the idea that generic AI can optimize an industrial operation without domain expertise. “You need to be there with the people,” she says. “You are not going to talk only to the CIO or the chief digital transformation officer. You need to be with the people doing the job, who know where the pain is and where the bottlenecks are.”
Simulation, MES, enterprise platforms and automation each contribute a different part of the answer. Simulation tests how a change propagates. MES provides production context. Enterprise systems connect the plant to inventory, finance and customer commitments, while automation carries the decision across organizational and system boundaries.
The most advanced factory will not be the one that drives every machine toward its local maximum. It will be the one that recognizes when a local improvement transfers cost, delay or risk elsewhere, then tests and executes a better system-wide response. Efficiency is not how hard one line works. It is how effectively the entire operation turns demand, materials, labor and capacity into the outcome the business actually needs.
