The Intelligent Factory: The Six Decisions
Six decisions that run every factory.
Strip away the org charts and the technology, and running a process plant comes down to six decisions, made over and over, every shift, by people working with incomplete information. Intelligence doesn't replace these decisions. It improves their quality and consistency.
The Framework
The decisions that shape every shift.
The framework works across mills, plants, and other process operations. A paper mill calls the plan its grade wheel and the rate its machine speed; a specialty chemicals plant calls them the campaign plan and the feed rate. The decision logic is the same: leaders must decide how to run the operation under real operating and market conditions. Each decision is explored in depth below, and each returns throughout Leading the Intelligent Factory.
One sets the plan. Two and three set the run. Four and five handle what goes wrong. Six makes it stick.
Plan it, run it, react to it, learn from it.
Decision 1
What should we run, and when?
Before anyone asks how fast, someone decided what to make, in what order, and how much to keep on the shelf. That choice sets the week’s losses.
The decision
Every operation chooses its product mix and its sequence: which grades, recipes or products to make, how long each campaign runs, what order they run in, and what gets made to stock rather than to order. In a paper mill it is the grade wheel, with basis weight, color and furnish changes grouped so each transition costs as little as possible. In food it is the line schedule and its allergen and clean-in-place changeovers. In specialty chemicals it is the campaign plan and the cleanouts between products.
What makes it hard
Every transition costs something: off-spec tons, washouts, cleanouts, a break on the first reel after a change. Long campaigns cut transitions but build inventory and stretch lead times. Short campaigns serve customers but spend capacity on changeovers. Make-to-stock smooths the schedule but ties up cash; make-to-order protects cash but turns the schedule into a reaction to the order book.
Most plants schedule from habit and the order book, and never measure what each transition actually cost. The sequence gets inherited, not designed.
Where intelligence helps
Measuring the real cost of each transition from the plant’s own history, product pair by product pair, turns sequencing from habit into evidence. Add margin by product and the lead times customers will accept, and the right mix and order are usually clear once the numbers are on the table.
Key Questions
Before you set the schedule
- Which products make money at today’s prices, and which only keep a customer?
- What does each transition really cost us in off-spec, time and energy, by product pair?
- Which products should we make to stock, and which only to order?
- How long should each campaign run, given inventory and lead time?
- Is our sequence designed, or inherited from habit?
- What lead time are we promising, and what does it cost to keep it?
In practice: Envoy cost studies measure transition losses by product pair from your own history, and Smart Centerlines holds the targets for each product once the schedule is set. See Smart Centerlines.
Decision 2
How fast should we run?
At the front door it's an economics question. Everywhere upstream it's a coordination question. Same decision, different driver.
The decision
Every factory has to decide how hard to push the operation. This is not simply a speed question: it is a rate-setting decision that balances throughput, yield, quality, reliability, energy, safety, equipment health, constraints, customer demand, margin, inventory position, and profit per hour. In a paper mill, it's deciding how fast to run the machine on a given grade based on moisture profile, sheet break risk, dryer limitations, recent instability, and order demand.
Two questions, not one
Where you stand in the mill changes what this decision even is. At the paper machine it is an economics question, and the market sets it. When everything you make can be sold at attractive margins, tons beat efficiency. When the market is soft, the same machine should be dialed back and run cleaner, because cost per ton is now the measure. Same machine, same crew, opposite right answer.
Everywhere upstream, nothing sets its own rate. The digester's rate comes from the machine, from the tank levels in between, from white liquor availability, from how fast the evaporators keep up. Rate propagates backward from the front door.
What makes it hard
The trap is treating how fast can we run as though it were how fast should we run. Capability is a constraint question; rate is an economics question, and most plants only ever answer the first. Operating rate is a commercial decision as much as a production decision. When the market will absorb everything the plant can make at attractive margins, the right move may be to push toward demonstrated capability while managing quality and reliability risk. When the market is constrained, running flat out creates excess inventory, wastes energy, and stresses equipment. The intelligent decision may be to optimize yield, energy, quality, grade mix, or cash contribution instead of maximum throughput. Intelligence improves the decision by connecting current operating conditions with historical performance, known constraints, quality risk, maintenance risk, and economic impact.
Key Questions
Before you push the rate
- Is the market asking us to maximize output, or are we under demand, margin, or inventory constraints?
- What is the current constraint: process, equipment, utilities, materials, downstream capacity, or market demand?
- What happens to quality, yield, energy, and downtime risk as rate increases?
- Is the operation stable enough to push? Are we carrying unresolved maintenance or reliability risk?
- Would running slightly slower produce more good output, better yield, or better profit per hour?
- Are different shifts, grades, or lines running at different rates for good reasons?
In practice: Envoy Intelligence computes the rate against live constraints and inventories, and keeps it in view as conditions move.
Decision 3
Where should everything else be set?
There are many ways to hit the same number. They are not equally good, and most plants settle that question by argument rather than by evidence.
The decision
The rate is settled. Now where do the other hundred variables go? On a recovery boiler, solids multiplied by flow gives the firing rate, but many combinations of flow, solids, and air distribution produce that same firing rate, with different steam generation, different reduction efficiency, and different plugging risk. In a digester, many combinations of temperature and time reach the same H-factor, with different yield, different rejects, and different steam. On a paper machine, many combinations of refining, steam, and speed hold the same quality spec at very different cost.
Why operators argue
Not because one recipe works and another doesn't. Because they all work. Each one hits the number. The argument is about which path costs least, and most plants have never had the evidence to settle it, so it gets settled by seniority, by habit, or by whoever is most stubborn on that crew. Everyone in the control room is right, and the plant still pays for the difference.
What makes it hardest
Turndown. Back-end processes are tuned, centerlined, and remembered at full rate. Every operator knows how the digester runs flat out. Ask the same area to run at eighty percent, which is precisely what Decision 1 does when the market softens, and there is no recipe. The operating point that was optimal at full rate is not optimal at reduced rate, and almost no plant has worked out what replaces it.
The trap
Local optimization is not plant optimization. The best operating point for the evaporators alone is not the best operating point for the recovery island. One area can look better while the total operation gets worse.
Key Questions
Finding the operating point
- For this target, how many combinations of settings would reach it?
- Which of those combinations costs least, and how do we know?
- Do all crews run this the same way? If not, which one is right?
- What changes about the answer when we are not at full rate?
- Are we optimizing one asset, line, or department at the expense of the total process?
- What does this operating point cost us per ton, and what would the best one cost?
In practice: this is the decision Envoy Smart Centerlines answers, with cost-optimal targets and limits for every grade, rate, and run condition.
Decision 4
When should we intervene?
Intervene at the right time, for the right reason, with the right level of response.
The decision
Every factory has to decide when a signal, trend, or abnormal condition has crossed the threshold from watch to action. This is not only a control-room decision: it can be an operator adjustment, an engineering investigation, a maintenance inspection, a quality hold, a rate change, or a leadership escalation. In a mill, it's deciding when moisture variation, vibration, sheet instability, or a steam imbalance deserves action before it becomes a break or a quality issue.
What makes it hard
The timing tradeoff cuts both ways. Late intervention lets small abnormal conditions become downtime, quality loss, equipment damage, or customer impact. Over-intervention creates alarm fatigue, unnecessary work, operator frustration, and loss of trust. Intelligence improves this decision by distinguishing normal variation from meaningful change and guiding the right level of response. Early warning is only valuable when it leads to timely, proportional, and trusted action.
Levels of Response
From watching to learning
Intervention is not one thing. It has degrees:
- Monitor: the signal is visible, but no action is required yet
- Investigate: the pattern is unusual enough to check context and likely causes
- Adjust: a controlled operating change is appropriate
- Escalate: the issue requires engineering, maintenance, quality, or leadership involvement
- Contain: the risk is high enough to protect safety, quality, equipment, or production continuity
- Verify: confirm whether the intervention worked
- Learn: update the workflow, threshold, runbook, or operating practice
In practice: Smart Centerlines defines what out-of-limits means; Envoy Intelligence groups deviations into events and makes the call.
Decision 5
What should we fix first?
The loudest problem is not always the most important problem.
The decision
Every factory has more problems, risks, and opportunities than resources to address them. And “fix” means more than repairing equipment. It can mean fixing an asset, a process condition, a quality driver, a reliability pattern, a workflow, an information gap, or a management-system gap: the signal no one owns, the alert no one closes, or the recurring issue that never becomes standard work. In a mill, it's choosing between a dryer limitation, a recurring sheet break pattern, a critical pump vibration, a bad actor valve, and a quality variation on a high-margin grade.
What makes it hard
Priority is not the same as urgency. Something can feel urgent because it is recent, visible, loud, or politically important, but the right priority depends on risk, value, constraint impact, customer impact, recurrence, and actionability. Safety, regulatory, and environmental obligations are never just another economic tradeoff; existing compliance systems remain authoritative. Once those are respected, intelligence helps leaders focus scarce resources on the problems that matter most. A signal without prioritization becomes backlog.
What to Weigh
Before it goes on the list
Seven tests for what deserves attention now:
- Risk: what could happen if we do nothing?
- Value: how much performance, margin, quality, energy, or production is at stake?
- Constraint: does this affect the current or future bottleneck?
- Customer: could this affect quality, delivery, or customer commitments?
- Urgency: how soon does action need to happen?
- Actionability: can we do something useful with the people, parts, and window we have?
- Recurrence: will fixing this remove a repeating loss, or only treat today's symptom?
In practice: Process Reliability Reporting delivers findings ranked by impact before the shift starts.
Decision 6
What can we learn from yesterday?
Reporting tells you what happened. Learning changes what happens next.
The decision
Every factory has to decide what to remember, change, standardize, or improve based on recent events, and “yesterday” means the last shift, the last grade change, the last upset, the last maintenance job, or the last month of recurring losses. In a mill, it's learning from a sheet break, a moisture variation, or a steam imbalance so the next similar event is diagnosed faster and handled more consistently.
What makes it hard
Looking backward is not the same as learning. Plants already have production reports, downtime reports, shift logs, maintenance histories, and dashboards, but those artifacts only become learning when they change what people do next. Too much knowledge stays trapped in shift notes, maintenance comments, engineering spreadsheets, and individual memory, and recurring problems get solved from scratch again and again. A factory has not learned from an event until the learning changes a workflow, runbook, threshold, standard, model, or training material.
What Learning Takes
From event to standard
Seven steps that turn experience into advantage:
- Capture: what happened?
- Contextualize: what was the operating situation?
- Explain: why did it happen, or what are the likely causes?
- Decide: what should we do differently?
- Update: which workflow, runbook, threshold, standard, or training material should change?
- Share: who needs to know before the issue repeats?
- Verify: did the change improve future performance?
In practice: Envoy Intelligence runbooks turn the investigation into something repeatable, so the second time costs less than the first.
Go Deeper
How well does your operation make these six decisions?
The six decisions look different at every stage of operational maturity, from unstable firefighting to a plant that learns. Score each decision against the five levels of The Maturity Model and the answer isn't a number, it's a sentence: repeatable on quality, unstable on reliability, so start there. The full story is in Leading the Intelligent Factory.