The Category Trap That Keeps Factories Stuck

new technology. The gap is between insight and action, then between action and control, then between control and a shaped outcome that holds under pressure. That is what this piece is about. We keep mistaking category adoption for value creation. We keep believing that if we buy what others are buying, the result will be what others are claiming. We keep confusing visibility with authority and measurement with control. “What would have to be true for this outcome to keep repeating.” The chart feels encouraging because it is full of the unglamorous work. Automation, analytics, scheduling, quality, execution. The backbone. Most expert’s social media commentary often makes that case plainly enough. They are right that every factory has the same human story, and he is right that leaders are tired of asking good people to compensate for broken systems. He is right to call out the gap between stated need and follow through on training and adoption. The trap is believing that an investment in a category closes the gap by itself. A category does not decide. A category does not act. A category does not absorb risk when reality deviates from plan. A category does not hold the line on Monday morning when the material is late, the machine is hot, the schedule is wrong, and quality is in question. Control does. An agent must be able to shape an outcome, otherwise it is not an agent. If a system cannot shape an outcome, it is not the thing we think we bought. It is an observer with a budget.

The chart that sells comfort and hides the bill

The category chart is comforting because it allows a clean conversation. We are investing in the right areas. We are not asleep. We are not stuck in the past. We are building the factory of the future. It is the kind of artifact that makes a board feel like it is doing diligence. It also hides the mechanism that determines whether the money becomes earnings, cash, and resilience, or becomes another layer of screens the plant learns to ignore. The hidden mechanism is simple and cruel. Value appears when an insight changes a decision in time to change an action, and the action is permitted, executed, and verified inside a loop that keeps working even when conditions degrade. If any link fails, the investment becomes a cost with a story attached. This is why so many plants end up with a modern stack and an old operating feel. The tools multiply. The meetings do not shrink. The reconciliations persist. The escalation load stays high. The operation becomes more visible and no more controllable. The board hears “AI and analytics,” then sees the plant still operating like a series of negotiated exceptions. The CFO hears “automation and scheduling,” then watches inventory rise because

the plan is still fictional. The COO hears “quality management,” then sees customer complaints persist because containment is still manual and late. Nobody is lying. That is what makes it expensive. Each purchase is defensible. Each vendor demo is plausible. Each dashboard is true enough to be dangerous. The bill shows up in the only place that matters, which is the time between signal and action and the cost of decisions that arrive after the window closes. That window is the difference between a stable plant and a heroic plant. It is the difference between a schedule that holds and a schedule that is rewritten by noon. It is the difference between quality that is controlled and quality that is documented.

When the backbone becomes a pile of parts

Automation. Analytics. Scheduling. Quality. Execution. Sensors. Cloud. AI. Vision. IoT. GenAI. Edge. Each category can matter. Each category can also become a new source of disagreement. If you want to see the trap in the wild, look for the moment when competent people argue about which number is real. Look for the moment when a supervisor carries two printouts that do not match and asks which system is right. Look for the moment when a planning team says the schedule is feasible and the floor says it is not. Look for the moment when quality says the lot is suspect and production says the orders must ship. Look for the moment when maintenance says the machine can run if we baby it, and the system says the asset health score is fine. That argument is not a people problem. It is a control problem. A modern stack often increases the count of places where truth can diverge. Each new tool brings its own semantics, its own data model, its own definitions, its own timestamps, its own failure modes. Even when integration exists, meaning often does not. The plant then becomes a courtroom where data is presented, challenged, and interpreted, rather than a cockpit where evidence triggers action. The cost is not only time. It is trust. Trust is the first requirement for permission. Permission is the first requirement for action. Action is the first requirement for control. When the truth is contested, permission retreats upward. Leaders stop delegating. They start requiring more reviews. They create gates. They call them governance. The gates slow decisions and invite workarounds. Workarounds create more exceptions. More exceptions create more gates. It looks responsible and it behaves like sludge. This is the part that is hardest for executives to admit. The organization is not slow because it lacks intelligence. It is slow because it lacks a reliable way to grant permission to act on intelligence.

If the system cannot be trusted, it cannot be allowed to shape outcomes. If it cannot be allowed, it cannot create value at scale. It can still produce beautiful reports. That is not what the CFO is paying for.

The false certainty that category spending creates value

The prevailing belief is reasonable. It goes like this. Industrial work is complex. Data is fragmented. The solution is to modernize the stack. If we invest across the backbone, we will reduce heroics, reduce surprises, and improve performance. The evidence seems visible. Companies with modern tools look more advanced. Vendors show case studies. Conferences celebrate the category leaders. Consultants map the maturity ladder. Peer benchmarks prove we are behind. Then the bills arrive. The bills arrive as integration cost. The bills arrive as training cost. The bills arrive as process redesign cost. The bills arrive as governance cost. The bills arrive as exceptions that still require human judgment, except now the human judgment happens under a higher cognitive load. Most of all, the bills arrive as unchanged decision latency. The signal still travels up the chain, across functions, into meetings, into approvals, and back down. The plant still learns that the fastest way to get something done is to bypass the system and call someone who can authorize it. Category spending does not fix this because category spending does not specify the decision. It does not specify who can act. It does not specify the evidence threshold for action. It does not specify verification. It does not specify how the system learns when it is wrong. It buys parts. It does not buy a loop. This is why I treat the category chart as a useful mirror and a dangerous map. It can remind us where the operational backbone lives. It cannot tell us what to do Monday morning to shape an outcome, deliver the value we want, and repeat it. That question has to be answered directly. What decision do we need to make relative to the action we need to take, enabling the control that shapes an outcome that delivers the value we want. Everything else is decoration.

The hidden mechanism. Why insight does not convert into control

Factories do not fail because they lack ideas. They fail because they cannot convert evidence into action inside the window where action matters.

That conversion has a structure. It is not philosophical. It is mechanical. A signal appears. Someone notices. A decision is required. Permission must be present. An action must occur. The result must be measured. The system must learn. The loop must hold under variation. Many modernization programs stop early. They build sensing and reporting. They build awareness. They do not build decision objects, permission pathways, and verified actions. So the factory becomes better at noticing problems and no better at preventing them. That is a cruel kind of progress. It raises anxiety and does not raise control. A control loop is not a dashboard. A control loop includes actuation. It includes authority. It includes constraints that prevent dangerous action. It includes auditability that makes delegation safe. This is where language matters. A factory can have “quality management” and still ship defects. A factory can have “advanced scheduling” and still miss delivery. A factory can have “AI and machine learning” and still suffer downtime. A factory can have “automation hardware” and still struggle because the chaos is automated, not removed. The categories are not lies. They are incomplete nouns. They imply a verb and then hide it. The verb is shape. An agent must be able to shape an outcome. If the tool cannot shape an outcome, the tool is not wrong. The buying logic is wrong.

Where Jeff is right, and where the encouragement becomes a trap

Jeff’s post lands because he does not insult anyone. He describes competent people working around broken systems. He praises investment in the unglamorous backbone. He warns that human capital is behind, and that training and adoption standards lag the stated need. That encouragement is valid if it pushes leaders toward disciplined work. It becomes a trap if it reinforces a familiar habit, which is believing that buying the backbone is the same as fixing the operating system. The backbone is not software. The backbone is decision rights and control loops. You can invest heavily in quality software and still operate with unclear disposition authority. You can invest heavily in scheduling tools and still have no agreed constraint model. You can invest heavily in analytics and still have no agreed definitions or thresholds that trigger action.

You can invest heavily in automation and still have an unstable process window that turns automation into a brittle machine for producing scrap. Jeff also calls out training and adoption. That matters, but it is not the deepest gap. The deepest gap is that we keep building systems that ask people to compensate for the architecture. We then call the compensation “adoption.” We call it “change management.” We call it “support.” In practice, it is asking humans to bridge what the system did not close. The hardest truth in modern manufacturing is that the most expensive labor in the plant is often the labor spent reconciling the enterprise’s own contradictions. Modern stacks can multiply those contradictions if the organization does not decide what it believes, what it permits, and what it will do when the evidence crosses a line.

The second order effect nobody budgets for. More tools, more gates

Once a plant has multiple systems, every decision becomes vulnerable to a new kind of delay. The delay is not lack of data. The delay is contest over what data counts. Each new system becomes another witness. Each witness tells a slightly different story. Leaders then create a procedure to reconcile the stories. Procedures feel safe. Procedures also become permanent. That permanence is what turns diligence into drag. Drag shows up as governance rituals. It shows up as “alignment.” It shows up as “approval cycles.” It shows up as “risk reviews.” It shows up as “data quality remediation.” It shows up as exception meetings that become daily. None of this is evil. It is the organization defending itself from acting on evidence it does not trust. The defense costs money. It also costs speed. Speed is not a vanity metric. Speed is the only way to keep the loop inside the boundary where action still changes outcomes. When the loop is slow, the plant becomes reactive. Reactivity is expensive. It looks like overtime, expediting, rework, scrap, premium freight, and missed customer promises. This is why the CFO flips the page and nothing moves. The organization is spending money to become more informed, then spending more money to defend itself from acting on what it learns. The net result is more knowledge and the same control. That is the category trap.

The decision object the enterprise is missing

If you want to diagnose whether a plant will turn category spending into value, you do not start by asking which vendors are chosen. You start by asking whether the enterprise has a repeatable unit of decision. Most firms have a repeatable unit of reporting. They have KPIs. They have dashboards. They have weekly reviews. They have monthly operating calls. They have scorecards with colors. Reporting is not a decision object. A decision object is the smallest bundle of context, evidence, authority, and action that allows a decision to be executed without a meeting. It is not a slide. It is not a dashboard. It is something that can move permission down the chain because it carries the evidence and the constraint. Without that object, every exception becomes a meeting. Every meeting becomes a gate. Every gate increases decision latency. Every increase in latency increases the need for heroics. Heroics then become culture, and culture becomes a substitute for control. This is why some factories feel like they run on adrenaline. They do. Adrenaline is a poor operating system. When the decision object exists, the conversation changes. Evidence becomes the trigger, not the argument. Authority becomes explicit, not negotiated. Action becomes auditable, not improvised. Verification becomes normal, not political. That is when the categories start to matter, because they now feed a loop.

Two paragraphs you can read aloud in an executive meeting

When you look at your modernization program, can you name the five decisions that most often decide whether a day goes smoothly or becomes a fire drill? Can you point to the moment in the day when each decision is made, by whom, and with what evidence? When the evidence is conflicting, who has the right to declare a source of truth and act, and what is the cost of waiting for consensus? If the same decision is escalated three times a week, is it truly a complex decision, or is the enterprise refusing to grant permission where the work happens? When a model flags risk, what happens next, in the record, on the floor? Does anyone have permission to intervene without asking for approval, and is that permission safe because it is auditable and bounded? If the intervention fails, does the system learn in a way that changes future action, or do we merely document the failure and call it governance? If you removed every dashboard tomorrow, would the plant lose control, or would it lose only visibility, which is a very different thing? If those questions produce silence, the category spending is not a plan for value. It is a plan for more observation.

The credibility tax. A prediction that would be embarrassing if wrong

Here is a prediction that can be tested without slogans. Across the next 24 months, many manufacturers that increase spending across the category stack will not see a material decline in daily exception escalations unless they also rewrite decision rights and build auditable decision objects that push permission to the point of action. If the exception load falls sharply while decision rights remain implicit and contested, I am wrong, and the field should say so. The reason I believe this is not because people are resistant. It is because systems that cannot be trusted cannot be permitted to act, and systems that cannot act cannot shape outcomes. The organization then keeps the same human loop and adds more evidence to argue about inside it.

The conversion problem. Why the same investment creates value in one plant and waste in another

It is fair to say that some manufacturers do convert these categories into value. They do it with discipline that is boring in presentation and decisive in consequence. They treat scheduling as a constraint problem, not a wish. They maintain routings, cycle times, and changeovers as audited facts, not tribal stories. They treat quality as containment, not documentation. They treat maintenance as intervention policy, not hero repair. They treat analytics as decision triggers with thresholds, not as a debate stage. They treat automation as stable process enforcement, not as a shiny robot. In those environments, the categories work because the control loops already exist. The tools serve the loops. They do not replace them. This is the counterexample that matters, because it shows the trap is not technology. The trap is believing technology substitutes for operating architecture. A plant with strong operating discipline can add tools and gain speed. A plant without that discipline can add tools and gain friction. The discipline is not a slogan. It is a set of decisions about what evidence counts, who may act, and how learning is captured. It is permission, bounded by auditability.

Permission is the missing layer that the stack charts never show

The charts list technologies. They do not list permission. Permission is the architecture of control. It determines whether insight stays in a report or becomes an action that changes the line. Permission is also where risk lives. Executives hesitate to push permission down because they fear mistakes. They fear a bad stop. They fear a bad hold. They fear a bad setpoint change. They fear a bad expedite decision that damages a customer relationship.

That fear is rational. The answer is not to keep permission centralized. The answer is to make permission safe. Permission becomes safe when it is bounded and auditable. Bounded means the action is constrained by policy, thresholds, and guardrails. Auditable means the evidence and the action are recorded in a way that allows learning and accountability. This is why category spending without auditability often produces either paralysis or chaos. If permission stays centralized, the system becomes a slow reporting machine. If permission is pushed down without auditability, the system becomes a risk machine. Either way, control does not improve. The practical implication is not a new tool. It is a new question for every tool. Where does permission live, and what makes it safe. An agent must be able to shape an outcome. The enterprise should treat that sentence as a procurement filter.

The investor problem inside the plant. Why the CFO page stays cold

Boards and markets do not reward intent. They reward repeatability. Repeatability is evidence that a firm can control outcomes under variation. That evidence reduces uncertainty. Reduced uncertainty changes the discount applied to future cash flows. This is why the valuation page matters. Category spending is easy to communicate. Repeatability is hard. The CFO cannot underwrite “we bought a scheduling tool.” The CFO can underwrite “we reduced lead time variability because we reduced decision latency from signal to replan, and the plant now holds schedule adherence under real constraints.” The CFO can und erwrite “we reduced customer escapes because containment now occurs at the moment evidence crosses a line, not after a meeting.” The difference is not the category. The difference is the loop. This is why the CFO page freezes the room. The operating story is full of activity and the valuation story is waiting for a different kind of proof. Proof is not a dashboard screenshot. Proof is a repeated sequence where the firm sees, decides, acts, verifies, and learns faster than the problems can compound.

The mold you have to break is older than the software

The enterprise you are running was designed for human decision speed. Layers exist because information had to travel. Gates exist because accountability had to be preserved. Procedures exist because memory was scarce and mistakes were expensive. Those features made sense when the world moved slower. Now visibility is cheaper. Compute is cheaper. Connectivity is everywhere. The scarce resource is not data. The scarce resource is permission that can move at speed without raising risk. This is why many modernization programs fail in a predictable way. They try to pour modern visibility into an old permission structure. The result is glare. Everyone sees more and nobody can move. You cannot modernize that with another dashboard. You have to reorder the relationship between evidence and authority. You have to decide what the organization will permit when the signal crosses a line. You have to capture that permission in a decision object that can be executed and audited. This is the only way category spending stops being theater and becomes value.

The last mile is not technical. It is moral and economic

At some point, every plant faces the same decision. Will we keep asking people to compensate for broken loops, or will we build loops that stop requiring compensation. The compensation looks like dedication. It looks like pride. It looks like ownership. It is also a tax. The tax is paid in turnover, fatigue, errors, and the slow drift of standards. Drift is a quiet killer. It does not arrive as a crisis. It arrives as a gradual acceptance that the schedule is always wrong, that quality always requires sorting, that maintenance always lives in reactive mode, that expediting is normal, that exceptions are the job. That acceptance is not culture. It is surrender. Breaking the mold means refusing surrender and refusing theater at the same time. It means insisting that every investment be tied to a decision that changes an action inside the window where action shapes outcomes. It means insisting that permission be explicit, bounded, and auditable. It means insisting that verification and learning be built into the loop, not added later as documentation. A modern factory does not need more categories. It needs fewer excuses.

What the chart should force us to say, and what it should never be allowed to say

The chart should force a sharper conversation than “we are investing where others invest.”

It should force a conversation about which decisions decide the day, who owns them, what evidence triggers action, what permission exists at the point of work, and how the enterprise learns when the intervention fails. It should force a conversation about whether our tools shape outcomes or merely explain them. The chart should never be allowed to imply that category buying equals value. Because if you accept that implication, you will spend heavily and still live in heroics. You will become more visible and no more controllable. You will carry a modern stack into the next operating review and watch the CFO page stay cold. An agent must be able to shape an outcome. That sentence is not philosophy. It is a budget filter. It is also a warning. If your modernization program cannot answer, in plain language, what decision it changes and what action it permits, it is not building control. It is building a more expensive way to watch yourself stay stuck. The room goes still for a reason.

References

This essay draws on the Deloitte 2025 Smart Manufacturing and Operations Survey as presented in the chart and on LinkedIn post commentary as a live example of how category benchmarks shape executive belief, then pulls the mechanism from operations and organizational economics that explain why category adoption often fails to convert into repeatable control, including Deming’s Out of the Crisis in 1986 on systems and variation, Goldratt’s The Goal in 1984 on constraints and the difference between local improvement and flow, Herbert Simon’s Administrative Behavior in 1947 and March and Simon’s Organizations in 1958 on bounded rationality and why procedure expands when consequence is feared, Jensen and Meckling’s 1976 work on incentives and delegated authority, Dixit and Pindyck’s Investment Under Uncertainty in 1994 on why timing and option value govern capital decisions, Judea Pearl’s The Book of Why in 2018 on the difference between association and intervention when the aim is to change outcomes, and the user’s own authorship standards for WSJ grade longform, scene integrity, falsifiability, and references discipline, which govern the prose and constraints used here.