Change a culture fast

Someone ran an offsite. There were breakout groups, a facilitator, sticky notes in four colours. Out of it came five values, and the third one was Ownership, and it went on a poster near the kitchen and into the onboarding deck.

Nine months later a service went down on a Saturday and the same three people fixed it while forty others didn't notice. Nobody looked at the poster. Nobody has ever looked at the poster.

The exercise wasn't stupid. The people in that room were serious and the words they picked were the right words. It just aimed at a layer of the organisation that doesn't take instructions, and everything downstream of that mistake was going to be slow no matter how much energy went into it.

Two claims here, and the second one is the useful one. Culture change is genuinely possible on the timescale of quarters rather than years. And the reason it usually isn't is that most attempts make three specific choices, each of which adds about a year on its own.

Nobody has ever measured the 70%

You've heard that seventy percent of change programmes fail. It gets quoted in the room where the decision is made, usually by someone sensible, usually to argue for a longer timeline and lower expectations.

It's made up.

Mark Hughes went looking for the source in 2011 and traced it back through the five places everyone cites: Hammer and Champy, Beer and Nohria, Kotter, Senturia and Keller, Aiken. Not one of them has data behind it. Each either states the number flat or cites someone else who states it flat. The earliest version, from Hammer and Champy in 1993, actually describes itself in the text as "our unscientific estimate." Beer and Nohria then put it in the Harvard Business Review as "the brutal fact" with no reference at all, and thirty years of slide decks have been quoting the brutal fact ever since.

This isn't pedantry about citations. That number does real damage, because it's used in advance, as permission. It buys a three-year timeline and lowers what anyone expects to see in the first six months, and a three-year culture programme has a specific failure mode that has nothing to do with culture: it outlives its sponsor. The VP who started it leaves, the new one has their own ideas, and the whole thing quietly becomes something people used to talk about.

Speed isn't a nice-to-have here. It's most of the design.

The arrow runs the other way

Schein's model of culture has three layers. Artifacts on top (what you can see, the posters, the rituals, the seating), espoused values in the middle (what we say we believe), and underlying assumptions at the bottom (what everyone actually takes for granted, mostly without noticing).

Almost every culture programme starts at the middle layer and pushes down. Get the values right, communicate them well, and behaviour will follow.

Behaviour doesn't follow. It leads.

The best-documented case is NUMMI, and it's worth knowing in detail because it's the one that settles the argument. GM's Fremont plant was the worst in the company: absenteeism regularly over 20%, quality at the bottom of the system, a workforce with a reputation nobody wanted. It closed in 1982. Toyota reopened it as a joint venture two years later and hired back the same people. About 85% of the workforce were the same UAW members, including the former strike leaders. The troublemakers were still there.

Absenteeism went to a steady 2%. Quality went from GM's worst to GM's best in a single year.

Nobody's mind was changed first. John Shook, who was there, puts it in the sentence I'd hang on the wall instead of the values poster: it's easier to act your way to a new way of thinking than to think your way to a new way of acting.

What actually moved, and the part people copy wrong

The mechanism at NUMMI was the andon cord. Everybody knows that part. The design details underneath it are where the transferable lesson lives, and they're the bit that gets left out.

Three properties, and you want all three.

It changed the work, not the messaging about the work. Nobody was asked to care more about quality. The job itself changed so that stopping the line was part of doing the job properly.

It was an obligation, not a permission. This distinction is doing enormous work and it's the one I see missed most often. "You're empowered to raise concerns" leaves the choice with the individual, which means the first person to use it is visibly choosing to be difficult, and pays for it socially. Make it an obligation and there's nothing to choose. You're not the person who stopped the line. You're the person who did their job.

And the response was guaranteed. Pull the cord and a team leader arrives inside your job cycle. Every time.

That last one does more work than anything else in this guide, so here it is on its own.

The first person who does the new thing is your entire communication strategy.

Whatever happens to them is the message, and it overrides anything you say at an all-hands by roughly an order of magnitude. If the first engineer who says "I don't think we should ship this" gets thanked in public and the date moves, you have told everyone something true. If they get a slightly cool response and the date doesn't move, you've also told everyone something true, and it's the opposite thing, and no poster is going to fix it.

You need a quarter of the room

Here's the finding that changes the tactics most, and it's recent enough that it hasn't reached most management writing.

Centola and colleagues ran controlled experiments on how social conventions flip, published in Science in 2018. Groups of twenty settled on a norm, then a committed minority pushed for a different one. Below about 25% of the group, the minority failed. Completely. Not slowly, not partially. Then at roughly 25% the thing flipped abruptly and the majority adopted the new norm.

Two consequences worth sitting with.

You don't need to persuade everyone, and trying to is what makes this slow. You need about a quarter, genuinely committed, which is a much smaller and more achievable target than "bring the organisation along."

And your dashboard is going to lie to you. Progress toward a threshold looks like no progress at all right up until it isn't. Run a sentiment survey at month four and it'll come back flat, and somebody senior will conclude the thing isn't working, and they'll be reading the data correctly and drawing exactly the wrong conclusion from it.

Seed clusters, don't broadcast

The second thing the research overturns is how you spread it.

Information spreads well through weak ties. One person tells one other person, it's true or it isn't, done. That's why the all-hands works fine for announcing a reorg.

Behaviour doesn't work like that. Centola's work on complex contagion shows that adopting a behaviour usually needs social reinforcement: seeing several people you know and trust do it, more than once. A single contact isn't enough, which means the network structure that spreads news fastest is close to the worst one for spreading a practice. Behaviour needs dense, clustered, overlapping ties.

So the all-hands is the wrong instrument. Not useless, but it reaches everyone exactly once, which is the one thing a complex contagion can't use.

What works instead is unglamorous. Pick two or three whole teams rather than a scattering of enthusiastic individuals across twelve of them. Get the behaviour to saturation inside those teams, where everyone sees everyone else doing it daily. Then connect outward.

The connection part is where it usually gets bungled. The instinct is to name one enthusiast per team, a well-connected person who'll carry it across on your behalf. That's a weak tie, and structurally it's the wrong thing entirely. What you want is what Centola calls a wide bridge: several people with existing relationships spanning the boundary, showing up with the same behaviour at roughly the same time. Two engineers who already pair with the next team will beat one appointed evangelist every time.

Six moves

Concrete version. This is what a leader actually does, and it's less about communication than most people expect.

Pick one behaviour, not a value. "Be more collaborative" isn't a behaviour, it's a mood. "Every design doc has a named reviewer within one working day" is a behaviour, because I can tell you at a glance whether it happened. If you can't count it, you can't change it and you certainly can't tell whether you have.

Change the work so the behaviour is the default path, not the brave path. If doing the right thing takes courage, you've designed it wrong, and you'll get it from the four people who have courage spare and from nobody else.

Guarantee the response, in public, before anyone has tried it. Say exactly what happens when someone does the new thing, then make sure it happens the first three times regardless of how inconvenient the third one turns out to be. The third one is the real test. It always arrives on a bad week, in front of a customer, attached to a date somebody senior already committed to, and how you handle it is what everyone will remember about the entire initiative.

Do it yourself when it costs you something. The cheap version (leader performs the easy instance of the new behaviour) reads as theatre and every person watching can tell. What lands is a leader eating a real cost in public, once. Slipping your own date because somebody raised a concern is worth more than a year of encouragement.

Change what happens to the old behaviour. Culture is the worst behaviour a leader walks past. Especially if the person doing it is good.

Then remove one thing that contradicts you. There's always at least one, and it's usually a metric or a piece of process. Asking for careful work while ranking people on ticket throughput isn't a mixed message, it's a clear message that your careful-work message is decoration. Killing the contradiction is worth more than adding anything.

The part you have to be willing to lose

Some people will leave. The ones who go first are frequently good, and they're good at the culture you're removing, which is a genuinely uncomfortable thing to sit with at the time.

That's the change working. It is not the change failing, and you'll need to have decided that in advance, because you will not be able to reason your way there calmly in the week it happens.

The harder version is the one everyone's waiting for. If your highest performer is your biggest violator, that's the test, and the whole organisation already knows it's the test, and has known for longer than you have. Whatever you do there is the actual policy. Everything else is commentary on it.

Worth knowing what the old culture is already costing, since that number rarely makes it into the conversation. Sull's team at MIT ran 1.4 million Glassdoor reviews through a classifier and found that toxic culture predicts attrition 10.4 times more powerfully than compensation does. Compensation ranked sixteenth. You're paying for this already, on a line item that says something else.

How to tell it's working before it looks like it's working

Given the threshold effect, sentiment lags badly, and if sentiment is your only instrument you will cancel this at month four. So measure behaviour.

Count the artifact, not the feeling. If the behaviour is a named reviewer within a day, count the docs. If it's raising concerns before the ship date, count the concerns raised, and be honest that a number going up is the good outcome here even though it feels like a problem.

Watch the seeded teams for saturation rather than average adoption. Three teams at 90% is a completely different position from twelve teams at 22%, even though the second one looks better on a chart and involves four times the people.

The signal that matters most is the first defection toward the new norm from somebody who was never in the seed group and has no reason to be helping you. That's the threshold starting to bite, and it usually shows up well before anything on the survey moves.

A rough timeline, on the understanding that yours will differ. One quarter to move behaviour inside the seeded teams. Two to three quarters for the norm to flip more broadly. About a year before someone describes it back to you as just how things work here, unprompted, having forgotten it was ever a project.

When it isn't a culture problem

Most things called culture problems aren't, and starting here saves an enormous amount of wasted effort.

People aren't writing tests because the suite takes forty minutes. That's a systems problem in a culture costume, and no amount of behaviour change survives contact with a forty-minute feedback loop. Go fix the loop. Most common misdiagnosis I see, and you can usually settle it in an afternoon by asking four people why they skipped it and listening to the answer instead of the one you expected.

Or the behaviour is a rational response to an incentive you have no intention of changing. People hide bad news because the last three who delivered it are no longer here and everybody can name them. Asking for candour on top of that is asking people to be brave on your behalf for nothing, and the ones who try will prove the old rule right in front of an audience.

No control over the work? Then it's influence, not culture change. Still worth doing. Just plan it as a slower, different thing rather than being surprised when the fast version fails to show up.

And if your teams are fully distributed and asynchronous, the mechanics get harder, because reinforcement is what complex contagion runs on and reinforcement is a lot cheaper when people share a room. Not impossible. It does mean the dense clusters have to be built deliberately, in recurring small-group time, rather than assumed to exist.

Where to start

Pick the single behaviour you'd most like to be true, phrased so that a stranger could count it.

Find the two teams where it's closest to happening already, and change their work so it's the default rather than an act of nerve. Say out loud what will happen to the first person who does it, then make that happen even when it's expensive.

Then hold your nerve through the quiet part, which is longer than feels reasonable and comes right before the thing you were waiting for.