The Checklist Manifesto: How to Get Things Right

Miles: Wright Field, Ohio, the autumn of nineteen thirty five. The Army is picking its next long range bomber, and one entry is so far out in front that people treat the contract as settled before anybody flies.

Miles: It rolls, it lifts, it climbs a few hundred feet, and it comes down. Two of the five men aboard don't survive it. A lock fitted to hold the control surfaces still on the ground had never been taken off.

Miles: The verdict is that the aircraft is more than one person can fly. And then a few pilots do the unexpected thing. They don't ask for more training, and they don't ask for a simpler aeroplane. They write a card.

Miles: Today: The Checklist Manifesto, by Atul Gawande. In two sentences: what costs us most has stopped being failures of knowledge and become failures of applying knowledge we already have, in jobs with too many steps. And the cheapest repair anybody's found is a list somebody reads out loud.

Miles: Sixty Books. Presented by Stoar Consulting and Coaching.

Miles: That's book thirty four. Welcome in. Sixty Books, I'm Miles, and the terms don't move: one management title per episode, sixty of the finest of the past century, each one asked the same pair of questions. What's in it for a person who leads, or gets led by, a mind wired to a different diagram. And what's left standing now AI sits in the middle of the working day. Nora holds the research, and every figure here got checked before a word went down.

Miles: Nora, we finished yesterday on the failure mode of speed. It isn't that the job gets dropped. It's that step four does.

Nora: And this book says step four has a character to it. It's never the hard part. It's the part everybody in the room has done a thousand times, which is why nobody watches it happen.

Miles: So what gets you is the routine item.

Nora: Missed by an expert, in a job with far more steps in it than anyone can hold. And his answer to that is a piece of paper you read aloud. There's a word on the cover of this book, and the word is manifesto.

Miles: For a list.

Nora: Yeah. For a list. He knows how that sounds, which is why half the book goes on convincing you the problem is bad enough to need one.

Miles: Tell the aircraft story properly, because he opens the book with it.

Nora: The thirtieth of October, nineteen thirty five. Boeing's Model two nine nine, outside Dayton. Major Ployer Hill has the controls, and he runs the Army's flying branch, so this isn't an inexperienced person having a poor morning. The cause was a gust lock. On the ground, wind batters an aeroplane's control surfaces, so they're pinned, and somebody frees them before flight. Nobody freed them. Hill died the same day, Boeing's own chief test pilot three weeks later.

Miles: And the contract went to the competitor.

Nora: It did. Then a loophole got found and thirteen Boeing machines were taken for evaluation, which is where the book picks up. The pilots flying them decided the aircraft wasn't past human ability. It was past human memory.

Miles: So what they wrote.

Nora: Fitted on an index card. Takeoff, flight, landing, taxi. Not one item was something those men couldn't have recited in their sleep. Instruments, door, windows, controls unlocked. That type went on to be built nearly thirteen thousand times.

Miles: The author, then.

Nora: Atul Gawande, born nineteen sixty five in Brooklyn to two doctors who'd come from India, raised in Athens, Ohio. Stanford, biology and political science, then a Rhodes Scholarship to Balliol College, Oxford, where he read philosophy, politics and economics. Harvard for medicine afterwards, and a New Yorker staff job from ninety eight, while still in surgical training.

Miles: And the books.

Nora: Complications in two thousand two, Better in two thousand seven, a MacArthur Fellowship in two thousand six. This one lands in December of two thousand nine, and Being Mortal follows in twenty fourteen and is the one most people have read. Since then, a health systems research centre, a healthcare venture set up by three very large American firms, and three years running the global health bureau at the United States development agency.

Miles: And there's one thing I want on the table before the evidence.

Nora: State it now, because it matters later. He led the World Health Organization programme that produced the surgical checklist, and he's a named author on the study that tested it. So this is somebody writing about work he was part of, and the book says so. Nothing improper in that. It's a fact the reader should be holding.

Miles: Now the argument, in his order.

Nora: Two kinds of failure, taken from a philosophy paper by Gorovitz and MacIntyre on the limits of what an expert can know. Errors of ignorance, where it goes wrong because nobody alive knows the answer. Errors of ineptitude, where the answer is known and somebody didn't apply it. For most of history the first dominated. His claim is that across a lot of fields it's now the second.

Miles: And the reason.

Nora: Volume. More is known than a professional can carry, and the work has more steps than a person can hold. His number comes from a nineteen ninety five study in one teaching hospital's intensive care unit, where trained observers recorded activity over four months. An average of a hundred and seventy eight separate actions per patient per day. One unit, one hospital, so it describes a place rather than handing you a rate.

Miles: So the failure isn't knowledge.

Nora: It's that the knowledge won't fit in a head. Then he sorts problems three ways, which is the most portable idea in the book. Simple, where a recipe gets you there. Complicated, where you need expertise and coordination but the thing is repeatable. Complex, where every instance differs and no recipe is ever finished. A list works in the complicated middle, and here's where he gets misread. It doesn't solve complexity. Complexity is what your judgment is for, and every routine item you spend judgment on is judgment you no longer have.

Miles: Right.

Miles: And then he leaves medicine.

Nora: For a building site, and it's the best chapter in the book. Nobody's been a master builder for a century, because no one person can hold what a modern tower needs. So that trade keeps two lists. A build schedule, every task and the order it happens in. And a submittal schedule, the conversations that have to happen before certain steps and who has to be in them.

Miles: So one list is tasks and the other is talking.

Nora: And it's the part that travels. The second list never says what the answer is. It says these four people have to have spoken before that wall goes up. Two books ago we watched a department rescue itself with index cards on a wall. Same technology, fifty years later.

Miles: And the design rules.

Nora: From a man at Boeing who spent two decades writing them. Pick your pause points, the moments where the list gets run. Then the kind. A do confirm list, where the team works from memory and then stops and verifies. Or a read do, where you read the item and then perform it. Five to nine items, about what a person holds at once. Sixty to ninety seconds, one page, exact wording. And only killer items, the steps most dangerous to miss and most often missed. Not everything you do. The things that end you.

Miles: And a book about checklists closes with.

Nora: A checklist for making checklists. In an appendix.

Miles: Of course it does.

Nora: And a last chapter saying the resistance isn't laziness, it's a picture of what a capable person looks like. We reward whoever improvises out of a mess over whoever quietly never got into it.

Miles: Now the part I've been waiting three days for. This is the one book on our shelf whose central instruction got tested against real outcomes.

Nora: It is, and one thing first. All of this is a report about published studies and about a book. Nobody here can speak to anybody's treatment, and none of it is a suggestion about anyone's care.

Miles: Understood. What does the book stand on.

Nora: Two results. Two thousand six, the New England Journal of Medicine, Pronovost and colleagues. A collaborative cohort study in intensive care units across Michigan, a hundred and three units reporting, three hundred and seventy five thousand catheter days. The median rate of one bloodstream infection went from two point seven per thousand catheter days at baseline to zero within three months. The mean went from seven point seven to one point four.

Miles: And the design.

Nora: Is the sentence that got dropped along the way. No randomization, no control group, volunteer units, before and after.

Miles: And the surgery one.

Nora: Two thousand nine, same journal, Haynes and colleagues, Gawande the last named author. Eight hospitals in eight cities, spanning rich and poor health systems, Toronto and London through to Manila and Ifakara. Three thousand seven hundred and thirty three consecutive patients before a nineteen item checklist, three thousand nine hundred and fifty five after. Death within thirty days, one point five percent before, zero point eight after. Complications, eleven percent to seven.

Miles: In absolute terms.

Nora: Seven fewer deaths per thousand operations, on those figures in those hospitals. And the design again: prospective, consecutive, before and after. No control group, no randomization, sites that had volunteered, everyone knowing they were measured.

Miles: All right. Twenty fourteen.

Nora: Twenty fourteen, same journal. Ontario had pushed surgical checklists across the province, hospitals reporting their compliance. Urbach and colleagues went to the administrative records and compared three months before each hospital's adoption against three months after. A hundred and one hospitals, a hundred and nine thousand operations before and a hundred and six thousand after. Adjusted risk of dying in hospital or within thirty days: zero point seven one percent before, zero point six five after. Odds ratio zero point nine one, interval zero point eight zero to one point zero three. Not significant. Complications, three point eight six against three point eight two. Also not.

Miles: So it doesn't work.

Nora: No. That's the reading I'd argue hardest against, and it's the one that number gets used for. What it measured is a mandate, at scale, in ordinary hospitals, over three months, out of records kept for billing.

Miles: Then say what changed between the settings.

Nora: Four things. In Michigan and in the eight hospitals the checklist arrived inside a programme: training, local leaders, people watching, the unit's own figures fed back month after month. In Ontario it arrived as a requirement to file a compliance rate. And that point has evidence behind it. Dixon-Woods and colleagues, Milbank Quarterly, twenty eleven, with Michigan's own programme leaders among the authors, rebuilt the theory of why it had worked and listed six mechanisms, none of them the checklist alone. Second, everybody in the earlier work had volunteered, and a province doesn't volunteer. Third, the baseline: those eight hospitals started at one and a half percent mortality, Ontario at zero point seven one, and you can't remove what isn't there. Fourth, three months is a thin window for a change in a rare event, and what got counted was a box, not whether the thing in the box happened.

Miles: And the journal said something at the time.

Nora: It ran an editorial beside the Ontario paper, by Lucian Leape, called The Checklist Conundrum, and he gave three reasons of his own. Compliance was reported as almost universal and he doubted that reflected use. Most hospitals took the list unmodified instead of building it with their own teams. And three months was probably far too short.

Miles: So where does the record sit.

Nora: In better condition than the argument suggests. Norway, twenty fifteen, Annals of Surgery. Haugen and colleagues ran a real randomized trial, stepped wedge, two hospitals, five surgical departments taking the checklist in random order. Two thousand two hundred and twelve control procedures against two thousand two hundred and sixty three. Complications fell from nineteen point nine percent to eleven point five, an absolute drop of eight point four points. Mortality went from one point nine percent to zero point two in one hospital, and across both, one point six to one point zero, it wasn't significant.

Miles: And the pooled picture.

Nora: Abbott and colleagues, twenty eighteen, British Journal of Anaesthesia, eleven studies and four hundred and fifty three thousand patients. Lower mortality, odds ratio zero point seven five, fewer complications, zero point seven three, very high disagreement between studies, all observational. And their own closing sentence, which belongs beside the number: this could simply reflect wider quality of care in hospitals where checklist use is routine.

Miles: Give me the tier statement.

Nora: And this book earns a different one from the last three. It's journalism: a practitioner writing for general readers, reporting studies rather than running one, which puts it where every book on this shelf sits, outside peer reviewed research. What's rare is that its central instruction got measured against hard outcomes, repeatedly, including once in a randomized trial. What the record won't carry is the short version its fame produced, that introducing a checklist reduces deaths. It carries something narrower. A checklist built with the team that runs it, inside a programme that trains and measures and reports back, has gone with better outcomes in several settings and was shown once in a trial. A checklist ordered from above and counted as a filing requirement has not. The distance between those two sentences is the lesson, and it isn't a verdict on anybody.

Miles: Our half, and you've been sitting on this since the cold open.

Nora: Because there's a claim here I'd defend harder than anything I've said in a fortnight. A checklist is an external working memory. It's the highest value adjustment in most workplaces, and nobody files it as one, because it turns up dressed as a safety practice.

Miles: Take the first half.

Nora: What's being externalised has been measured. Alderson, Kasper, Hudec and Patros, twenty thirteen, in the journal Neuropsychology. Thirty eight pooled studies of working memory in adults with ADHD against adults without, on laboratory tasks. Moderate sized differences in both domains they examined, carrying into adulthood rather than being a childhood matter. Group averages. Not a test for anything, and not a description of any particular person.

Miles: And the second population.

Nora: Gives us something sharper. Landsiedel, Williams and Abbot-Smith, twenty seventeen, Journal of Autism and Developmental Disorders. They split prospective memory, remembering to do a thing later, into two kinds. Time based, where you catch the moment yourself. Event based, where something in the world prompts you.

Miles: And the two come apart.

Nora: Badly. Time based: six studies, a hundred and eighteen autistic participants against a hundred and eighteen comparison, effect size minus zero point eight seven, interval minus one point one four to minus zero point six zero. Large, and consistent. Event based: eleven studies, three hundred and eleven against two hundred and eighty seven, minus zero point four one. Small, and the studies argue with each other.

Miles: Put that in plain words.

Nora: Catching the moment yourself is where the difference lives. Being prompted from outside is where most of it goes. And that conversion is what a checklist performs. It takes a job that depended on you noticing and makes it one where a piece of paper notices.

Miles: Huh.

Miles: And does externalising actually work, or is that just tidy.

Nora: Measured, and outside any clinical population. Sam Gilbert, University College London, twenty fifteen, Quarterly Journal of Experimental Psychology. Four online experiments, eleven hundred and ninety six participants, on a task where you hold intentions briefly and may set a reminder. Setting them improved performance, and people set them adaptively, more as the load rose and more when distraction was likely. The same laboratory, five years on, found offloading earlier made almost no difference to those people's later unaided performance. It changed what they chose next time, not what they could do.

Miles: And the part about not having to ask.

Nora: Is why I'd put this book in more hands than any of the last three. An adjustment normally requires somebody to explain how their head works to a person who controls their pay. A checklist requires nobody to say anything. It's issued to a whole team as a quality practice, the one who needed it gets it, the one who didn't gets it too, and nobody's identified. Same structure as the cap on open work we described two days ago, coming through a different door.

Miles: Cautions.

Nora: Two. First, I've done the thing I always do, which is set two literatures beside one another and draw a line between them. Nobody has run a trial of workplace checklists with neurodivergent employees as the outcome. That's an argument, and I'd rather label it than let it pass for a finding.

Miles: And the other one.

Nora: Comes straight out of the last segment. The moment a list is imposed and scored, it stops being a memory aid and becomes a compliance record. The box gets ticked, the step doesn't happen, and the person it was meant to help is carrying paperwork on top of the job. Ontario is the warning, and it applies in an office as it applied in a hospital. Build it with the people who'll run it, keep it to the items that actually bite, and never let the score become the point.

Miles: Fair enough.

Miles: A minute, in the frame this show keeps. The Checklist Manifesto says most of what goes wrong now goes wrong because a known step got skipped by a competent person carrying too many steps, and a short list read out loud is the cheapest repair anybody's found. For a person leading a brain wired to another plan, or running one, here's Monday. Write the list with the team that's going to run it, because the one place a mandate got measured on its own, it found nothing. Keep it to five to nine items, and only the ones that have actually bitten you. Put the pause points where the handoffs are, and let it say who has to have spoken, not only what has to have been done. And notice what you've quietly built: an adjustment that reaches everybody who needs it, costs nobody a disclosure, and never asks anyone to explain their own memory to their manager.

Miles: Now bring the machines in.

Nora: The standing cautions about this whole class of tool live at the end of the episode now, past the music, in words nobody moves, so I'll point you there and spend the minutes on what this book alone gives us.

Miles: Go on.

Nora: His two kinds of list are the cleanest way I've found to talk about these things. A read do use is the tool performing the step. A do confirm use is you performing it and the tool checking afterwards. Nearly all the enthusiasm points at the first. Nearly all the value sits in the second, and the second is the one almost nobody's building.

Miles: And writing the list itself.

Nora: Is free now, and that's the trap. Ask one of these for a checklist covering any process you like and you'll have a plausible thirty item document inside a minute. Thirty is four times what he says a person will run. And the items are generic, because it's never seen your failures. Killer items come from what's gone wrong in your building, and there's no route to those through a model that wasn't there.

Miles: So it writes the wrong list.

Nora: It writes a reasonable one, which is worse, because a reasonable list gets adopted. The work the book asks for is the subtraction. Which of these thirty has ever hurt us, and what are the five that survive.

Miles: And the checking use.

Nora: Has a known failure with a literature behind it, called automation bias, meaning leaning too hard on a system that's advising you. Goddard, Roudsari and Wyatt reviewed it in twenty eleven for the Journal of the American Medical Informatics Association. Thirteen thousand eight hundred and twenty one papers screened, seventy four included, from several fields. Workload, time pressure and task complexity make it worse. Training makes it better, and so does holding the user accountable for the outcome rather than for having checked.

Miles: And somebody went back at it.

Nora: Lyell and Coiera, twenty seventeen, same journal, eight hundred and ninety screened and forty included. The received view was that this needs a distracted person juggling several things. They found it in single task work whenever verifying the answer was hard, and they note that few of those studies used a control condition, so take the direction as established and the size as not. Which gives you the rule. The easier a thing is to check, the safer it is to let a machine check it, and a great deal of what these tools get handed is expensive to verify.

Miles: Then the last word.

Nora: His, in a sentence he'd recognise. A ticked box has never once been the same thing as a completed step, and the value of a list is that a person said the words aloud and somebody else heard them. These tools will hold your list, print it and tick it. Saying it out loud is still yours.

Miles: That's The Checklist Manifesto.

Miles: Next time: Accelerate, by Nicole Forsgren, Jez Humble and Gene Kim. Gawande put a list into surgery and then went and measured what happened to the patients afterwards, which is more than almost any management idea has ever had done to it. Tomorrow, at long last, somebody does that to software.

Nora: Four measures, a great deal of statistics, and a sampling problem I'll want a minute for.

Miles: That's good work, Nora. Thank you.

Nora: Go read it.

Miles: That's the episode.

Miles: Sixty Books is made by Stoar Consulting and Coaching. Logan Williamson is a Licensed Professional Counselor in Texas and Louisiana. What that was, was a book taken apart for people who run teams and people who are on them. It wasn't coaching. It wasn't counseling. And at no point was any of it a statement about you.

Nora: The citations sit on the episode page with the transcript beside them. Number thirty five tomorrow.

Miles: Two things for the record. Neither voice belongs to a person. Miles doesn't exist, Nora doesn't exist, neither of them is licensed to practise anything, and nothing in the past twenty minutes was addressed to whoever's listening. Every script is drafted with AI in the loop, then edited and approved by Logan Williamson, a Licensed Professional Counselor in Texas and Louisiana, at Stoar Consulting and Coaching.

Nora: And one note about the AI segment. The evidence that these tools help neurodivergent people at work is early. There are controlled trials now, but they measure things like conversation skill rather than how a job actually goes, and for work itself most of what exists is individual providers reporting success. So take it as promising and not proven. The tools also get rebuilt every few months, so whatever you set up today is built on something that moves.