31The Fearless Organization

Amy C. Edmondson, The Fearless Organization: Creating Psychological Safety in the Workplace for Learning, Innovation, and Growth, Wiley, 2018

Linked to 34 chapters: supports 22 · contradicts 5 · mechanism 19 · same claim 7 · pairing 6

Edmondson joined a study of medication errors in two hospitals in the mid-1990s expecting to show that the better a care team, the fewer mistakes it made. Her data said the opposite, at a level of significance she could not dismiss. The resolution she reached two years later — that good teams do not make more errors, they report more — became the subject of the next two decades of her work. Psychological safety is her name for a shared belief that a work group is safe for interpersonal risk: that asking a question, admitting a mistake or floating a half-formed idea will not be met with humiliation or reprisal. It is not niceness, not trust, not a personality trait and not a relaxation of standards, and it belongs to a group rather than a person, which is why it varies from one floor of a hospital to the next inside a single culture. Her case for why it now matters is a claim about the work: where a task is repetitive and knowable, silence costs little; where it is uncertain and interdependent, an unspoken question is lost information, and that loss leaves no trace except in the person who held back. The book sets two decades of studies against some twenty organisational cases — Volkswagen, Wells Fargo, a runway in the Canary Islands, Pixar, a South African platinum mine, a Japanese nuclear plant — and ends with what a leader does: set the stage, invite participation, respond productively.

The underpinning

The book opens with twin newborns delivered at twenty-seven weeks. A neonatal nurse practitioner Edmondson calls Christina Price has been taught that a high-risk premature baby should receive a surfactant for the lungs as soon as possible. The neonatologist has not ordered one. She steps forward to remind him and stops: the week before she had heard him berate another nurse in public for questioning an order. She tells herself the babies will probably be fine, that the doctor must have a reason, and lets him walk out on his rounds.

Edmondson reads the moment as an unconscious calculation. Christina has weighed an immediate, certain interpersonal cost against a delayed, uncertain clinical benefit and has done what people reliably do with that comparison: discounted the future. Underneath sits the older habit Goffman described, of managing the impression one makes by controlling what one reveals. Nobody, Edmondson writes, arrives at work wanting to look ignorant, incompetent or disruptive, and each risk carries an obvious avoidance rule. Do not want to look ignorant: do not ask questions. Do not want to look incompetent: do not admit mistakes. Do not want to be called disruptive: do not make suggestions. In one study of employee voice, eighty-five per cent of respondents could name an occasion when they had felt unable to raise a concern they believed mattered.

The accident that produced the concept was a piece of discordant data. Edmondson, a first-year doctoral student, had the good arrangement of measuring her independent variable while other people measured the dependent one: trained nurse investigators established error rates for each unit over six months, blind to her survey of team effectiveness. Both varied — there was a tenfold range in human errors per thousand patient days from the best unit to the worst, a wrong dose reported every three weeks on one ward and every other day on another — and the correlation between them was significant and pointed the wrong way. She then sent a research assistant to observe, blind to the error data, to the survey scores and to her new suspicion. He found that the units differed sharply in whether people felt able to talk about mistakes, and that this difference tracked the detected error rates almost perfectly.

She was not the first to use the term: Edgar Schein and Warren Bennis wrote in 1965 about the psychological safety people need to cope with the anxiety of organisational change, and William Kahn, in 1990, gave the concept its most useful gloss, the belief that you will be given the benefit of the doubt. Edmondson's contribution, in a dissertation on fifty-one teams in a Midwestern manufacturing company and the 1999 paper that came out of it, was to show that the thing lives at the level of the group, can be measured, and predicts both learning behaviour and rated performance. Where it differs between two units of one company, the leaders differ.

Much of the first chapter says what psychological safety is not, because the concept is easy to soften. It is not niceness: Edmondson has heard a dozen companies describe their own variant of "Company Nice", in which people are polite in the meeting, disagree in the corridor and never implement what was agreed. It is not extraversion, since climate acts on introverts and extraverts alike. And it is not trust, although the two overlap.

The last denial decides how the book is used. Psychological safety is not lowered standards. Edmondson draws it as one of two axes, the other being how demanding the work is, and the four quadrants are the argument in miniature. What worries her is not the comfortable quadrant but the one where standards are high and voice is absent, which she takes to be the common case.

The measure is seven survey items, three worded so that disagreement indicates safety, and variants used by other researchers yield reliabilities from about .73 to .94. Edmondson then narrows her claim: psychological safety takes off the brakes that stop people reaching what is possible, and is not the fuel.

The paper trail

Mark Costa, chief executive of Eastman Chemical, told a Harvard classroom in 2018 that his greatest fear in the job was that people were not telling him the truth. Edmondson uses him to make the chapter's point: psychological safety is not a perk granted to make employees happy but, in work that depends on integrating what different people know, a condition of getting the knowledge at all. She and her colleagues read more than a hundred studies and sorted them into five groups.

The first documents how much silence there is. Milliken, Morrison and Hewlin interviewed forty employees across five industries, and every one could describe an occasion on which they had not spoken; the reasons given most often were fear of being labelled negatively and fear of damaging a working relationship, and what surprised the researchers was that people withheld suggestions for improvement as readily as bad news. Edmondson and Jim Detert found the same in more than two hundred and thirty interviews at a multinational technology company, and extracted a set of taken-for-granted rules: do not criticise something the boss may have built, do not speak without solid data, do not speak up when the boss's boss is present. They tested the third with vignettes given to managers elsewhere, who were significantly more likely to offer an important correction when the boss's boss was absent. A manufacturing technician explained that he had kept an idea for speeding up production to himself because he had children in college, then admitted that nobody at that plant had ever been fired.

The caution persists because the terms of the trade are lopsided.

Nobody has ever been fired for silence. Against a benefit that is deferred, diffuse and doubtful, self-protection wins on every term, which is why exhorting people to be braver does not change the outcome.

The second and largest group concerns learning. Hannes Leroy and colleagues surveyed nurses in fifty-four departments across four Belgian hospitals and separated two things usually confused: errors reported and errors made. Departments with higher psychological safety reported more; where safety was high and the department genuinely prioritised patient safety, fewer were made. Edmondson, with Ingrid Nembhard and Anita Tucker, studied over a hundred quality-improvement teams in neonatal units at twenty-three hospitals and split their activity into learning what — reading the literature, catching up alone — and learning how, which means asking, suggesting and thinking together. Psychological safety predicted the second and had no relationship at all with the first, which is the cleanest demonstration in the book that the mechanism runs through interpersonal risk and nothing else. Enno Siemsen and colleagues added the finding that matters most for an idea not yet sure of itself: confident people speak up anywhere, and a safe workplace is what lets a person share something they are only forty per cent sure of.

The third group links safety to performance. Markus Baer and Michael Frese studied forty-seven mid-size German firms, all pursuing process innovation, and found the innovation raised return on assets only where psychological safety was present. Google, in a five-year internal study code-named Project Aristotle, examined a hundred and eighty of its own teams for the composition effects the literature predicted — educational background, personality mix, whether people socialised outside work — and found nothing. Turning to norms, the researchers came upon the published work on psychological safety, and, as the study's lead researcher Julia Rozovsky put it, "it's the underpinning of the other four." The other four were clear goals, dependable colleagues, meaningful work and a belief that the work has impact, and without the first they did not operate.

The fifth group Edmondson treats as the most interesting, because in it psychological safety does not cause an outcome but changes the sign of somebody else's relationship. Bret Bradley and colleagues found that conflict improved the performance of project teams where safety was high and damaged it where safety was low; a study of a hundred and ninety-five master's teams at a French university found that expertise-diverse teams did well with psychological safety and badly without. The long-standing inconsistency in the literature on demographic diversity, Edmondson suggests, is what a missing moderator looks like.

Avoidable failure

The cases that open the second part are organisations that had the talent and lost anyway. Martin Winterkorn took over Volkswagen in 2007 with a precise ambition: triple United States sales within ten years and pass Toyota and General Motors. Clean diesel was the instrument, and diesel could not meet American limits on nitrogen oxides. The engineering manager Wolfgang Hatz said as much at the time: the Californian regulator was not realistic, and the impossible could not be done. What the engineers did instead was write software that could tell from the two rotating wheels that a car was on a test bed, run the emissions equipment there, and switch it off on the road, where the cars emitted up to forty times the permitted level. About eleven million vehicles carried the device and the sales target was met four years early, until engineers at West Virginia University compared laboratory with road performance in 2013. Winterkorn resigned saying he was not aware of any wrongdoing on his part. Edmondson resists the single villain, although Winterkorn is a good candidate: executives described a pulse that rose on the way to his office, and he was the protégé of Ferdinand Piëch, who had been convinced that terrorising subordinates was how profitable design got made. But the failure did not need a villain. It needed only a leader who assumed that no news meant good news.

Wells Fargo ran the same script with different instruments. Its Community Banking division, six thousand branches serving roughly one household in three, grew by cross-selling, and by 2015 was averaging 6.11 products per customer against an industry average of 2.71. The target had outrun what customers could afford and was demanded anyway, on pain of dismissal, and branch employees found it easier to open accounts nobody had asked for than to report what they were learning in the field. The bank settled with three regulators for $185 million in September 2016. Nokia held twenty-three per cent of the world's mobile phone market in the late 1990s and by 2012 had lost three-quarters of its value; a study built on seventy-six interviews with its managers and engineers concluded that it lost the smartphone to a fearful emotional climate, to temperamental leaders and middle managers afraid to carry the truth upward, rather than to poor vision. The fourth case is the regulator that missed the crisis: David Beim, commissioned to examine the New York Fed's own supervision practice, found officers who described themselves as intimidated and passive, meetings that sought consensus rather than resolution, and, within three weeks, signs that supervisors dependent on the banks for information had become deferential to the institutions they regulated.

What Edmondson draws from the four, with Paul Verdin, is a proposal about strategy: treat it as a hypothesis rather than a plan, tested by action, so that the front line's encounters with customers become data senior executives want. Under that reading, Wells Fargo's missed targets were results rather than insubordination, and pressing harder on execution after a shortfall makes things worse when the shortfall is telling you the assumption was wrong.

Dangerous silence

On 1 February 2003 the Space Shuttle Columbia broke up on re-entry and its seven crew died. Two weeks earlier a NASA engineer named Rodney Rocha had watched the launch footage repeatedly, convinced that insulating foam had struck the left wing and unable to tell from grainy video whether it had done damage. Resolving it meant satellite photographs, which meant senior managers asking the Department of Defense; his own manager thought it unnecessary. Rocha wrote to fellow engineers that they were not to send messages much above their own rung of the ladder, and when the foam strike came up in the mission management meeting he sat on the periphery and said nothing. Asked afterwards by a television interviewer why, he said: "I just couldn't do it. I'm too low down", raising his hand above his head to show where the meeting's chair sat. Edmondson asks the reader to take the verb seriously. He did not say he chose not to. He said he could not, which is an accurate description of an experience common in hierarchies and invisible to the people at the top of them, who rarely notice the silencing effect of their own presence.

Twenty-six years earlier two Boeing 747s had collided in fog on the runway at Los Rodeos on Tenerife and five hundred and eighty-three people died, still the worst accident in civil aviation. The KLM captain, Jacob Veldhuyzen van Zanten, was the airline's chief flight instructor, held the power to extend pilots' licences, and had tested his own first officer on the 747 two months before.

Telling the first officer he should have shown more backbone is an ethical argument, not a method. What changed the outcome for later flights was institutional: cockpit training was rebuilt around crew decision-making, pilots were trained to assert a concern and captains to receive one, and the practice became the crew resource management training every commercial pilot now undergoes.

The hospital case is Betsy Lehman, a health columnist at the Boston Globe, who died at the Dana-Farber Cancer Institute on 3 December 1994 after a prescription ordered the entire four-day chemotherapy dose for each of the four days. It was written by a clinical research fellow, copied into the record by a nurse, and filled by three pharmacists, none of whom queried it in an institution where a third of patients might be on a trial and unusual doses were ordinary. Lehman spent three weeks vomiting in a way her doctors called the worst they had seen and also normal for the treatment. The error was found three months later by a routine data check rather than by anyone asking a question, and Edmondson's interest is in why nobody did.

A culture of silence has a second half, which is a culture that does not listen. Before the Challenger launch in 1986 the engineer Roger Boisjoly did raise the cold-weather risk to the O-rings; his data were incomplete and his argument vague, and the group could have resolved the ambiguity with simple analysis had it listened. Edmondson calls the pattern a Cassandra culture, and her long example is Fukushima, where a 2009 meeting on Daiichi's readiness for natural disaster kept tsunamis off the agenda, and where the seismologist Yukinobu Okamura, pointing to the Jōgan tsunami of 869 and returning with models showing the seawall would fail above magnitude 8.4, was told that a legendary wave measured by no modern instrument was a poor basis for standards. The 2012 parliamentary investigation concluded that the accident was manmade and its direct causes foreseeable, and its chairman wrote that the mindset behind it was made of reflexive obedience, reluctance to question authority and insularity.

The chapter closes in a newer medium. Susan Fowler's account of being propositioned by her manager at Uber on her first day, reporting it with screenshots, and being told it was a first offence by a high performer appeared on her own website in February 2017; Travis Kalanick resigned four months later.

The fearless workplace

Against those Edmondson sets organisations that built candour deliberately, and insists they are not alike. Pixar institutionalised it as the Braintrust, created in 1999 to rescue Toy Story 2: a standing group of directors and storytellers convened every few months to watch a film in progress and tell its director what is not working. The rules are that the feedback is about the film and not the person, that the comments are suggestions rather than mandates because the director remains responsible, and that criticism comes from people who have been in the chair themselves. Ed Catmull's framing statement — that all their films start out bad — is what makes the rest possible, because a director who expects an early version to be poor can hear that it is.

Ray Dalio built something more severe at Bridgewater, where the founding principle is that no one has the right to hold a critical opinion without voicing it; Edmondson marks the distance between his position and hers, since he frames silence as an ethical failure and she does not. At Google X a team spent two years trying to make fuel from seawater, established that production costs would not come down far enough, cancelled the project, and every member received a bonus. Astro Teller's formulation is that he is not pro failure but pro learning, and the economics are that killing an unpromising moonshot early frees the scientists and the money.

Two cases are about physical safety and show the work of building voice where none existed. Cynthia Carroll became chief executive of Anglo American in 2007, having found that the company had lost nearly two hundred workers in five years, and was told by an executive that zero harm would not happen in their lifetime. She shut Rustenburg, the world's leading platinum supplier, generating about eight million dollars a day, and said she would not restart it until she had heard from every worker. Because the workforce spoke several languages and literacy was low, the company gathered thousands at a time in a stadium, hired a theatre group to act out exchanges between a worker and a supervisor, then broke them into groups of forty or fifty, which said nothing, because nothing in the history of the mine or of the country suggested they should. What unlocked it was a change of question: instead of asking about safety, managers asked what was needed to create a workplace of care and respect, and used the lekgotla, the South African village assembly in which everyone speaks in turn, uninterrupted, until the matter is settled. One group asked for hot water to wash and make tea, and got it. Fatalities fell from forty-four in 2006 to seventeen in 2011, and never reached zero.

The other is Naohiro Masuda, superintendent of the Fukushima Daini plant five miles down the coast from Daiichi on the same afternoon in March 2011. Three of his four reactors had lost their cooling systems, nobody knew what was broken, and finding out meant sending people outside after an earthquake and a tsunami. Masuda did not issue orders. He wrote on a whiteboard the magnitude and frequency of the aftershocks and a chart showing the danger falling over time, having not been sure his team would go if he asked. At ten that evening he asked for four groups of ten to survey the four reactors, and no one refused. All forty returned by two in the morning with the news that there was still power behind Reactor 1.

Making it happen

The instruments come from Julie Morath, who arrived at Children's Hospital and Clinics in Minneapolis in late 1999 as chief operating officer with the goal of one hundred per cent patient safety, at a time when clinicians mostly believed that when something went wrong somebody was to blame. Edmondson divides what Morath did into three, and the division is the book's practical residue.

Setting the stage is mostly framing the work. Morath toured the hospital explaining that tertiary care is a complex system prone to breakdown, presented the research on adverse events, and changed the vocabulary — a study rather than an investigation, an accident or a failure rather than an error. A frame is a set of assumptions laid over reality that we do not notice because we take ourselves to be seeing what is there, and the default frame, in which an accident means somebody was careless, makes concealment rational. Framing is not done once. Had NASA's leaders kept emphasising how uncertain and interdependent the work was, Edmondson observes, the invitation to bring a tentative worry would have been visible to the engineers who had one.

Part of framing is being explicit about what failure means here, which differs by the kind of work.

Inviting participation is the second activity, and Morath's version of it was a question. Having explained the complexity and heard nothing back, she resisted explaining it again and asked instead: was everything as safe as you would like it to have been this week with your patients. The wording is doing work. It is concrete — this week, your patients — aspirational rather than accusatory, and unanswerable with a yes or no. People began to describe incidents they had seen and caused. She then built structures that made speaking easier than staying quiet: a steering committee whose members were each invited with a personal reason why their view was wanted, a blameless reporting system, and eighteen focus groups, in which a contribution is explicitly requested and silence becomes the awkward option.

Underneath the invitation sit two behaviours. The first Edmondson calls situational humility, borrowing Schein's "here-and-now humility": not modesty, and not the opposite of confidence, but the recognition that you do not have all the answers about this situation, which in a changing environment is accurate. Anne Mulcahy was known at Xerox as the master of "I don't know", and found that admitting it raised rather than lowered her colleagues' confidence in her. In the neonatal study, what distinguished the units whose improvement work succeeded was a leader who was approachable, acknowledged fallibility, and actively asked for input. The second behaviour is inquiry, harder than it sounds because of naive realism, the sense of simply seeing what is there, which stops us wondering what anyone else is seeing. A question qualifies, on Edmondson's rules, if you do not know the answer, it cannot be answered yes or no, and it is framed narrowly enough to be answerable. Bob Pittman's habit at MTV was to ask what the dissenter said, and to send the room back to find out when told that everyone agreed.

Responding productively is the third activity and the one that decides whether the first two survive. Its first element is appreciation, offered before evaluation: had the neonatologist thanked Christina for raising the surfactant he could then have explained why he judged it unnecessary, and the thanks would have bought him a nurse who speaks up next time. Good process and good outcome come apart, so rewarding only outcomes punishes the people whose judgement was sound and whose luck was not. The second element is destigmatising failure, and here Edmondson has a number she has collected from audiences for years. Asked what proportion of the failures in their organisations are genuinely blameworthy, managers say one to four per cent. Asked what proportion are treated as blameworthy, they say seventy to ninety.

Children's Hospital's answer to a complex failure was the focused event analysis, a meeting of everyone whose work touched the event, examining it from each of their angles to find the contributing factors rather than the culprit. For an intelligent failure the response can go further: the chief scientific officer at Eli Lilly held parties for high-quality experiments that did not work, which frees the scientists in time and, because people come to parties, spreads the knowledge that stops the same failure twice.

The third element keeps the rest honest. Psychological safety is reinforced, not damaged, by a fair response to a genuine violation, and dismissal can be a productive response to a blameworthy act. Edmondson works through the case of the Google engineer fired in 2017 over a widely circulated memo about the company's diversity policy, without taking a side, in order to separate two situations. Where a clear policy was broken, a tough sanction is fair and tells everyone the policy is real. Where the boundary was unclear, the sanction teaches something else — that unpopular opinions are not tolerated here, and that one mistake ends you — and both lessons reduce the willingness to speak that the organisation needs.

Since 2018

The term has travelled further than almost any construct in organisational research, and further than its evidence in places. Edmondson was named the most influential management thinker on the Thinkers50 ranking in 2021 and again in 2023, and psychological safety became a standard item in engineering and healthcare team surveys, in aviation and regulatory reform, and in the measurement practices around software delivery. Along the way it acquired the meaning she spent a chapter denying, and much of her later writing has gone into pushing it back: safety is not comfort, and not an alternative to accountability but the thing that makes accountability survivable.

The core empirical claims have held up in the shape she stated them. That psychological safety varies at the level of the group rather than the company has been replicated repeatedly. That it matters more as work becomes uncertain and interdependent, and barely at all for independent routine tasks, is what later meta-analytic work finds — a real test, since it is a prediction that could have failed. The moderator result, that diversity and disagreement and distance help or hurt depending on it, has accumulated support.

What is contested is the inferential machinery. Much of the literature is cross-sectional survey work in which safety and outcomes are measured at the same time and often from the same respondents, so the causal arrow is argued rather than demonstrated, and estimates of the effect on performance shrink as designs tighten. Error reporting, the outcome in the study that started everything, is measured by a process the climate itself affects, which makes the original finding compelling as a story and awkward as evidence. Project Aristotle, the result that carried the idea into general circulation, was an internal analysis reported in a newspaper and on a company blog; it has never been published where anyone could check the method. Edmondson is careful about that and less careful readers have not been.

Two of her own cases moved on. Volkswagen's settlements passed thirty billion dollars and several executives went to prison, and the Federal Reserve capped Wells Fargo's assets in 2018, a restriction that stood for years. A case the book could not include has since become the standard one: the Boeing 737 MAX, where two crashes in 2018 and 2019 and a door panel departing an aircraft in 2024 produced regulatory reviews reporting that employees feared retaliation for raising safety concerns and did not trust the channels for doing so. Edmondson's next book, in 2023, went back to the failure typology that appears here as a table and made it the whole subject, on the grounds that the hard part is not persuading people that failure is informative but teaching them to tell which kind they have.