- Step 1. Walk the floor before you read anything
- Step 2. Count the constraint yourself, twice
- Step 3. Read the gap between your count and the system
- Step 4. Establish the four numbers
- Step 5. Measure how long a decision takes
- Step 6. Sort twelve months of downtime by asset
- Step 7. Get gross margin by product family
- Step 8. Have the commercial conversation
- Step 9. Write the one-page diagnosis
You started ten days ago. You have been handed a monthly operating review deck, a capital request list, an org chart, and a quality scorecard, and you have sat through about fourteen introductory meetings. Everyone has been welcoming and slightly guarded. Several people have told you privately what the real problem is, and no two of them agreed. You have a strong suspicion that most of what you have been shown is accurate and none of it is the answer.
That is the normal condition. What follows is a procedure rather than a philosophy: nine checks, in order, that fit inside two weeks and produce a diagnosis you can act on. It is written for someone doing this for the first time, so nothing below assumes you already know where to look. If you want the full recovery sequence that this diagnosis feeds, it is in how to turn around a manufacturing plant.
One rule before you start. Every check below produces a reading, and every reading points somewhere. Do not act on any of them until all nine are done. A single reading in isolation will send you confidently in the wrong direction, which is the most expensive thing a new plant manager can do in month one.
Step 1. Walk the floor before you read anything
Walk the entire production floor on your first day, alone, without a guide and without a notebook. You are not inspecting. You are building a baseline impression that you will test against the data over the next two weeks, and it only works if you do it before anyone has framed the plant for you.
Three things to notice. Where material is sitting, and how much of it. Piles of work in process are a physical record of where the flow stops, and they will tell you more in ten minutes than a value stream map will tell you in a week. Which machines are running and which are idle, and whether the idle ones have people standing near them. And what the physical condition of the place says: whether tooling is stored or stacked, whether the aisles are clear, whether the maintenance shop looks like a workshop or a parts graveyard.
Do it again on a different shift before the week is out. Plants are frequently two different businesses depending on who is running them, and the difference between first shift and third shift is one of the more revealing numbers you will never see in a report.
Two additions if you have time on the first pass. Look at where people are standing rather than at the machines, because clustering tells you where the plant believes its problem is, and that belief is worth knowing even when it turns out to be wrong. Then find the whiteboards. Every plant has a set of them, and they will be either current, several weeks stale, or covered in something nobody uses any more. A stale board is not a housekeeping issue. It is a record of the approximate date on which the plant stopped believing that its own measurement changed anything, and that date is usually within a few months of when performance turned. Note it and test it against the numbers later in the fortnight.
Resist the urge to fix anything you see on this walk. You will see at least three things that are obviously wrong and could be corrected in an afternoon. Write them down and leave them. Correcting a visible problem on day one is satisfying, and it teaches the organization that your attention is captured by whatever is most visible, which is precisely the reflex you are here to remove.
Step 2. Count the constraint yourself, twice
Identify the constraint, then stand at it and count output for one full hour, twice, on different shifts. Not a sample, not an average, not a number somebody brings you. One hour, your own eyes, written down at the time. This single check does more diagnostic work than the rest of the list combined.
Finding the constraint is usually easy: it is the operation with material queued in front of it and starved operations behind it. If two candidates look equally plausible, count both. If nobody in the plant agrees on which operation is the constraint, that disagreement is itself a finding and you should write it down, because a plant that does not know its own constraint is not scheduling. It is reacting.
While you count, log the interruptions separately: minutes lost to changeover, to material arriving late, to a quality check, to an operator leaving the station, to a machine stopping. You are not building a case against anyone. You are building the distribution, because the shape of the interruptions tells you whether you are looking at a maintenance problem, a planning problem, or a decision problem.
A plant that cannot agree on which operation is its constraint is not scheduling. It is reacting, and every improvement it makes will land somewhere that does not matter.
Step 3. Read the gap between your count and the system
Compare your hourly count against what the system reports for the same hours, and treat the size of the gap as the primary diagnostic reading of the whole exercise. The absolute output number matters less than whether the plant’s own measurement is telling it the truth.
My working heuristic, and you should calibrate it to your own business, is that a gap inside a few percent is normal noise, and anything past roughly a tenth is a measurement problem rather than a variance. The reason for that threshold is practical: past that point, every improvement project in the building is being evaluated against a number that cannot detect whether it worked.
The related check is frequency. Ask what the shortest interval at which anyone can see line output actually is. If a supervisor makes resource decisions hourly but receives his number daily, he is making eleven decisions blind for every one he makes informed, and the hour lost at nine in the morning is not visible until the following day, by which point it has been absorbed into the schedule and quietly rebaselined as normal capacity. That single mismatch between measurement frequency and decision frequency is the most common structural defect I find in underperforming plants, and it is cheap to fix.
Step 4. Establish the four numbers
Four numbers give you the plant’s actual position, and you want all four in hand before you say anything public about strategy. Cash and the weekly burn. Output per labor hour on the constraint. Gross margin by product family. On-time delivery measured from the customer’s originally requested date.
Cash and burn tell you how much calendar you have, and the calendar governs everything else. A plant with eighteen months of runway and a plant with eleven weeks require different first moves even when the underlying operational problem is identical, and new managers routinely design the eighteen-month plan for the eleven-week situation.
Output per labor hour on the constraint gives you real capacity, as opposed to nameplate capacity, which is a number derived from a machine specification and a set of assumptions nobody has revisited. Gross margin by family rather than in aggregate matters because the aggregate hides both ends of the distribution, and the whole point of the exercise is to find the ends.
On-time delivery is the one that starts arguments. In most systems the promise date gets reset after the first miss, so the reported number measures performance against a date the plant chose rather than a date the customer asked for. Ask for the original request date. You will frequently discover that the delivery performance sales has been apologizing for is better than the delivery performance you actually have.
A note on where these numbers come from. Get all four from the systems of record rather than from a deck, and get them for the last eight quarters rather than for the current year, because a single year hides the direction of travel and direction is most of what you are trying to establish. A plant sitting at a given gross margin that was five points higher two years ago is a different business from one that has climbed to the same place, and the deck you were handed will usually show you only the current number.
Then check one thing about each: who produces it, and whether that person also has a stake in what it says. This is not an integrity question, it is a design question. Numbers assembled by the function whose performance they describe drift over time in a predictable direction, and the drift is neither deliberate nor visible to the people producing it. Where you find that arrangement, note it, and plan to move the production of that number somewhere neutral during your first quarter.
Step 5. Measure how long a decision takes
For one week, log every decision that leaves the floor: what was asked, who it went to, and when an answer came back. This is the single most useful week of data collection available to a new plant manager and almost nobody does it.
You are looking for two readings. The median return time, which tells you how much friction sits between a problem and its resolution. And the distribution of destinations, which is almost never even. In most plants two or three people have accumulated approval rights over a decade of small reorganizations, and they have become, without anyone deciding it, the plant’s actual constraint. They are usually the most conscientious people in the building, which is exactly why the queue formed at their desk.
Read the second-order effect as well. When escalation reliably takes more than a week, supervisors stop escalating and start working around, and the plant loses its early warning system. The signature of that condition is reporting that gets quieter and more positive while the underlying numbers get worse. If your predecessor’s last two monthly reviews looked calm, check this carefully.
Two practical points on running the log. Do not announce it as an audit. Ask two or three supervisors to keep a simple list for a week, framed as you wanting to understand where their time goes, and keep it in their handwriting rather than in a system. The moment it becomes a formal exercise the data changes, because people begin escalating things they would ordinarily absorb, which is exactly the behavior you are trying to measure the absence of.
And log the decisions that never left the floor as well. Ask each supervisor at the end of the week what he handled himself that he was not strictly authorized to handle. That second list is the more interesting one. It is a direct measure of the gap between the authority you have granted on paper and the authority the plant has quietly assumed in order to keep running, and closing that gap in writing is one of the cheapest improvements available to you in month one.
Step 6. Sort twelve months of downtime by asset
Pull twelve months of unplanned downtime, sort it by asset, and look at the shape of the distribution rather than the total. The total tells you how much you lost. The shape tells you whether you have an equipment problem or something else wearing equipment’s clothes.
A broad distribution across many aging assets is a genuine equipment and capital problem, and no operating change will substitute for a maintenance strategy and money. A concentrated distribution, where two or three assets produce most of the loss, is usually not an equipment problem at all. Check whether those assets have appeared on the capital request list before. If they have been requested and deferred across multiple cycles, your cause is a capital allocation process that cannot tell a constraint asset from a convenience asset, and that process lives in an office.
Then take the ratio of planned to unplanned maintenance hours. When unplanned dominates, maintenance has been converted into a reaction queue, and the mechanism is nearly always that planned windows keep getting surrendered to the schedule. That is a repeated decision, not an equipment fact, and it is reversible in a quarter.
Step 7. Get gross margin by product family
Ask finance for gross margin by product family and expect the answer to be less useful than it sounds, because standard costing allocates overhead evenly and therefore reports every product as roughly as profitable as its neighbor. Your job in this step is to find out how the allocation is done, not to accept its output.
Ask one question: which costs are attached to products by volume, and which are attached by anything else. In most plants the honest answer is that essentially everything is attached by volume, which means changeover time, tooling upkeep, forecast error, documentation burden, and the engineering and customer service attention a difficult product consumes are all spread evenly across the line. The effect is that your best products are subsidizing your worst and the P&L cannot show it.
You are not running a full costing project in your first two weeks. You are establishing whether a product mix problem is plausible, and the quick test is to count the product families in the bottom decile of volume and then count the share of the constraint’s changeovers they consume. When a small share of volume is eating a large share of your scarcest hours, you have found something worth a proper analysis later. The wider version of this pattern is cause seven in why manufacturing plants underperform.
Standard costing allocates overhead by volume, so it will report every product as roughly as profitable as its neighbor. That is an accounting convention, not a finding.
Step 8. Have the commercial conversation
Before the two weeks close, sit down with whoever owns the commercial side, because everything up to this point has been an inside-the-building diagnosis and the problem may not originate inside the building. Three questions, one meeting, two hours.
What did we quote over the last four quarters and what share did we win. Where we lost, what did the customer say the reason was, and press past the first answer, because in industrial businesses the stated reason is price roughly every time and the actual reason is often lead time or a delivery failure that nobody logged. And which customers have quietly reduced their share of wallet without ever leaving, since that decline never appears as a lost order and never triggers a review.
What you are testing is direction of causation. A plant winning the work it quotes and failing to deliver has an execution problem and the recovery sequence applies. A plant that has stopped winning has a position problem, and running an execution recovery on it produces a very efficient operation with nothing to make. Two hours in week two prevents two quarters of the wrong work.
Step 9. Write the one-page diagnosis
Write one page, before day fifteen, that states what you found, what you believe the primary cause is, and what you are going to do first. One page, not a deck, and written by you rather than assembled by anyone else. The discipline of the single page is the point: a cause you cannot state in a paragraph is a cause you have not isolated.
Three sections. What the numbers say, including the readings that contradicted the reporting. What you believe the primary cause is, stated as one sentence with the evidence that supports it. And the first three things you are going to do, with dates and with names attached. Nothing else, and no strategy statement.
Then give it to your boss and to your leadership team on the same day. Two weeks of visible counting followed by a clear written position does more for your standing in a plant than any amount of listening, because it demonstrates the thing the organization is actually assessing, which is whether you will look at something and then decide.
Two things not to put in the page. Do not include a strategy statement, and do not include a target. Both are decisions that belong to the end of month one at the earliest, and putting them in a document written on day fourteen commits you to positions formed before you finished the analysis. The page is a diagnosis and a set of first moves. That is enough.
And keep a copy of the raw readings, dated, in your own files. Six months in you will be told by several sincere people that the plant has always run this way, or that the hour you counted in week one was unrepresentative, or that a particular gap did not exist before you arrived. Your own hand counts, taken before anybody had an interest in the outcome, are the only version of the baseline that cannot be revised later. That single habit has settled more arguments for me than any reporting system I have ever installed.
About the Stagnation Assassin
Todd Hagopian is a Fortune 500 transformation executive who has generated $3B in shareholder value across Berkshire Hathaway, Illinois Tool Works, Whirlpool, and JBT Marel, where he serves as Global VP of Product Strategy and Marketing. Known as The Stagnation Assassin, he is the author of two published books: The Unfair Advantage: Weaponizing the Hypomanic Toolbox and Stagnation Assassin: The Anti-Consultant Manifesto. His blog is published in 15+ languages and read by operators worldwide. Bring him to your stage via his speaking page or connect with him on LinkedIn.
Putting a new operating leader into a plant this year? This procedure works as a session for the people who will run it. If you want your operations leadership taken through the two-week diagnostic before they are dropped into one, book it for your team.
See also: The 4 Beliefs Your Industry Has Never Tested, Why Your 30-60-90 Plan Should Be Shorter