By Bryan J. Kaus
Putting out fires is not improvement of the process.
— W. Edwards Deming, Out of the Crisis
In 2004, BP’s Texas City refinery recorded the lowest injury rate in its history, roughly a third of the refining sector average. Management, by the Chemical Safety Board’s later account, believed the plant had turned the corner. On March 23, 2005, the isomerization unit overfilled during startup, a blowdown drum vented hydrocarbon to the atmosphere, and the vapor cloud found an ignition source. Fifteen people died, most of them contractors in trailers parked too close to the unit. One hundred eighty were injured.
The injury-rate chart was real. The people who built it had done real work. It measured slips, cuts, strains, and vehicle incidents, and it measured them accurately. What it did not measure was whether a 1950s-era process unit with a history of overfills and a disconnected relief system was any safer than it had been the year before. The board found that the focus on personal safety had let the company believe conditions were improving while process safety deteriorated. Fixed costs had been cut 25 percent across the company between 1998 and 2000. The refinery’s central training staff had gone from thirty people in 1997 to eight in 2004. Four process safety coordinator positions sat unfilled on the day of the explosion.
I spent the better part of two decades in and around refineries, and I have sat in rooms where an injury-rate chart earned a round of applause. The applause was not wrong. The number deserved it. The error was in what the room concluded from the number.
I have been thinking about that error since finishing Malcolm Gladwell’s The American Way of Killing, which came out last month. The book is about why the United States has so much lethal violence, and it is worth reading for the stories alone, whatever you make of the conclusions. I expect plenty of readers will argue with them. That argument is not mine to settle here. What I took from the book was a distinction that travels well beyond its subject.
Gladwell opens with a young gunshot victim in Washington, D.C., and the trauma team that saves him. The surgeons are extraordinary. American emergency medicine has become so good at keeping shooting victims alive that the survival rate itself has changed the statistics. That is a genuine achievement, and the patient and everyone who loves him would call it a miracle. It is also a separate question from whether fewer people are being shot. The physician manages the consequence. Whether the condition that produced it has changed requires different evidence and the work of institutions well beyond the hospital.
That is the distinction: managing the consequences of a problem is not the same as changing the conditions that produce it. The trouble is that the first can be done so well that it removes the pressure to attempt the second.
The people who keep everything working
Anita Tucker, then a doctoral student at Harvard Business School, spent 197 hours shadowing twenty-two nurses in eight hospitals. She and Amy Edmondson catalogued 120 instances of what they called operational failures: a missing medication, a broken piece of equipment, a linen delivery that did not arrive. In 92 percent of those cases, the nurse solved the problem on the spot and moved on. A unit short of linen borrowed from the unit next door. The patient was cared for. The nurse felt the satisfaction of a problem handled. And the reason the linen had not arrived went unexamined, which meant it would not arrive again next week.
Tucker and Edmondson named the two behaviors. First-order problem solving gets the work done. Second-order problem solving investigates why the work needed rescuing. The nurses they watched were skilled at the first and almost never did the second, not because they were careless but because the job gave them no time, no authority, and no mechanism to carry what they knew to anyone who could act on it. Tucker was explicit that the nurses were not to blame. The hospital had built a system that converted their competence into concealment.
Anyone who has run an operation will recognize the pattern. Customer service promises a delivery date against inventory data nobody fully trusts. The plant finds out on Tuesday. A scheduler tears up the week’s sequence, a supervisor finds someone willing to stay late, and the truck leaves on time. The customer gets what was promised. Every person involved did something worth recognizing. The process that made the promise has not improved by one hour, and next Tuesday looks the same.
Over time, experience in a business like that becomes partly the knowledge of how to work around the business. People learn which spreadsheet to believe, whose phone call releases an order, which commitments can only be kept by personal negotiation. That knowledge is valuable. It is also a dependency, and it stays invisible right up until the person who holds it leaves, or the volume rises past what one person’s judgment can absorb.
Employee resourcefulness is an asset. Leadership’s job is to know what it is being spent on. Judgment applied to a genuinely unusual situation strengthens a company. Judgment consumed every week by the same preventable failure is a tax on the people who have it, and a company paying that tax has less capacity for growth than its results suggest.
What an improving number can miss
Texas City is the severe case, but the mechanism is ordinary. A measure improves. The improvement is real. The organization draws a conclusion wider than the measure can support.
Strong margins can reflect better execution, a favorable market, deferred spending, or all three. On-time delivery can reflect a reliable process or a heroic one. A smaller headcount can reflect simpler work or the same work moved onto the people who remain. A rising maintenance budget can mean the plant is being protected or that it is breaking more often. In BP’s own account, maintenance and turnaround spending at Texas City had risen 40 percent over the five years before the explosion. Money was going in. It was going in to deal with what had already broken.
Before I treat a good result as evidence that a business has become stronger, I want to know what changed in its ability to produce that result again. Which constraint was removed? Which cost went away permanently rather than moving to a different line, a different quarter, or a different person’s evening? What did the result consume that will not be there next time?
Those questions separate an improvement that reduced a burden from one that relocated it.
Why the same problem keeps coming back
There are honest reasons organizations get stuck managing consequences. The immediate failure has a customer, a deadline, and a name attached. Prevention competes with all of that while offering a benefit shaped like something that never happens. The scheduler who saves the order is visible on Tuesday afternoon. The person who fixes order promising prevents fifty rescues nobody will ever see.
The recurrence also eats the capacity needed to end it. Managers spend their days clearing exceptions. The most experienced people become indispensable to keeping work moving and therefore unavailable to study why it keeps stopping. The people who understand the failure best are the ones with the least time to investigate it, and a proposed fix lands as one more demand on a stretched floor.
Under those conditions, telling people to take ownership accomplishes very little. A supervisor can see a recurring obstacle clearly and still have no control over the commercial policy, the system setting, or the staffing decision that produces it. The problem lives between functions, and each function, measured on its own, appears to be doing its job.
Growth and acquisitions sharpen all of this. The informal arrangements that hold a business together at one scale do not automatically survive the next one. A company can add customers, plants, and revenue faster than it builds the ability to coordinate them. The resulting strain gets described as a shortage of good people, when the good people are already spending most of their week holding the seams together by hand.
Southwest Airlines learned a version of this in December 2022, when a winter storm triggered cancellations that cascaded through a crew-scheduling system unable to keep up, grounding more than 16,000 flights over ten days. The storm was weather. The recovery capacity was something the airline had built, funded, and deferred over many years. The two questions have different answers, and only one of them is in management’s hands.
Giving the organization a chance to learn
Toyota’s production system contains a deliberate answer to first-order problem solving. Under the principle it calls jidoka, a machine stops itself when it detects an abnormality, and any operator can pull a cord and stop the line. The stated purpose is to keep a defect from moving downstream and to make the abnormality visible to someone who can address its cause.
What interests me is the willingness to let a problem interrupt the work. The quieter alternative, catching the defect at final inspection and fixing it there, preserves the appearance of flow while making the weakness harder to study. Toyota chose the interruption because an interruption produces information and a repair produces only a repaired part.
Applying that requires judgment. Not every irregularity justifies stopping an operation, and a reporting channel on its own changes nothing. Somebody has to investigate, somebody has to decide what to change, and somebody has to check afterward whether the change worked. For the plant that keeps rescuing Tuesday’s order, the first move might be pulling twenty expedited orders from the last quarter and tracing each one back to the moment it became an expedite. The answer could be inventory data, lead-time assumptions, equipment availability, or a sales incentive that rewards commitments the operation cannot keep. Usually it is several of these interacting.
The fix follows the diagnosis, not the other way around. New software does nothing if the commercial incentive still produces the same promises. More reporting consumes time if nobody can act on what it shows. And the test comes later: fewer rescues, less rework, the same service delivered with less intervention, and no new problem created somewhere else by solving this one.
One warning. Better visibility usually makes the business look worse before it looks better. Work that was being absorbed silently starts reaching people who can see it. That is progress, if the organization treats it as information rather than as evidence that something has gone wrong.
The judgment behind prevention
The argument has a limit. Demand changes, equipment fails, and circumstances arise that no economical design could have prevented. The ability to recover well is a real capability, and a company without it will not survive long enough to prevent anything.
Some exceptions are rare and cheap and should simply be handled when they occur. Others deserve investment despite years without incident, because the consequence if they ever arrive is unrecoverable. Texas City had gone decades. The judgment turns on exposure, severity, how much is actually known, how reliable the existing safeguards are, and what a proposed change can realistically accomplish.
What that judgment needs is an honest accounting of the recurring cost. Repair bills, overtime, and customer concessions are the visible part. The larger part is the skilled time absorbed by rework, the management attention pulled into familiar crises, and the growth that did not happen because the people who would have led it were busy saving Tuesday.
Prevention costs money. So does recurrence. The leader’s job is to see both sides of that ledger with equal clarity, and most organizations are far better at counting the first.
The Point Taken
What stayed with me from Gladwell’s book was not a conclusion about guns. It was a reason to ask more carefully what an improved outcome actually proves. A successful intervention can demonstrate extraordinary skill at handling a consequence. Changing the condition that produces the consequence is a separate commitment, usually spread across different people, different decisions, and different budgets, and it is the commitment that gets deferred when the intervention keeps working.
In business, that commitment starts with curiosity about the work behind the result. Recognize the people who delivered under difficult conditions. Then use what they know to understand why the difficulty keeps returning. Their resourcefulness is worth more to the company once the company learns from it.
A strong quarter deserves attention. So does the condition the business is in when it finishes one. Did it gain capability, free up time, and reduce its reliance on repeated rescue? Or did it deliver by drawing down reserves of effort and goodwill that will be harder to draw on next time?
For anyone who has led a team through a recurring problem, the telling moment comes after the pressure passes. The order has shipped. The customer has been reassured. Everyone is relieved. Does the organization make time to understand what happened, or does the next urgent demand carry that opportunity away?
The fire is out. What did we learn about why it keeps starting?
© 2026 23.5 Strategies; The Point Taken™