Skip to content
Practitioner10 min readUpdated September 2026

Culture Is A Lagging Indicator

Culture-change programmes fail because culture is an output, not an input. Westrum's typology, why psychological safety is produced by how incidents are handled, and how a leader changes culture by changing what is measured, funded and promoted.

At some point in most transformations, somebody concludes that the real problem is cultural. The processes have been redesigned, the teams reshaped, the tooling modernised, and delivery is still slow, so the remaining explanation must be the attitudes of the people. A culture programme follows: values, a communications campaign, workshops, a survey, posters in the stairwell, an offsite where everyone agrees the values are good.

Eighteen months later the survey scores have moved slightly and nothing else has. This is the standard result, and it is not caused by poor execution of the culture programme.

It is caused by the direction of causality. Culture is not a thing you install and then observe the effects of. It is the accumulated pattern of what people have learned about how this organisation actually behaves — what gets rewarded, what gets punished, what happens when you raise a problem, whose priorities win when two things conflict. That pattern is produced by structure, incentives and the observed behaviour of senior people. It is an output. Measuring it is useful; addressing it directly is not, because you are attempting to change a reading rather than the mechanism behind it.

The consequence for a leader is liberating rather than defeatist. If culture is a lagging indicator, the leading indicators are things you control: what you measure, what you fund, who you promote, and what you do in the first ten minutes after something goes wrong. Change those and culture follows within a few quarters, without a single workshop. Leave them unchanged and no amount of communication moves it, because people believe what they observe rather than what they are told.

Westrum's typology and what it is actually measuring

Ron Westrum's classification of organisational cultures — pathological, bureaucratic, generative — is the most useful frame available here, largely because it is defined by observable behaviour rather than by stated values.

PathologicalBureaucraticGenerative
OrientationPowerRulesPerformance
InformationHoardedIgnored unless it is your departmentActively sought
Bad newsShot the messengerLeads to justice-seekingLeads to enquiry
ResponsibilityShirkedNarrowShared
Bridging between unitsDiscouragedToleratedEncouraged
NoveltyCrushedCreates problemsImplemented
FailureScapegoatingBlame allocatedEnquiry into cause

What makes this typology valuable is that every row describes something you can watch happening. None of them asks what people believe. The question is what happens to information as it moves through the organisation — whether it flows, and what happens to the person carrying it.

That is not a coincidence. Westrum's underlying argument is about information flow, and it connects to the queueing argument running through this library. A pathological organisation is slow because information does not reach the people who need it, so decisions are made on incomplete data and corrected expensively later. A bureaucratic organisation is slow because information stops at departmental boundaries — the same boundary-as-queue mechanism described in organisational design above team topologies, applied to knowledge rather than work. A generative organisation is fast because problems surface early, when they are small.

So the culture question and the delivery question are not two separate agendas competing for executive attention. They are the same system observed through different instruments.

The diagnostic use of the table is straightforward. Take a real incident from the last quarter — a serious one, where something went badly wrong — and walk through the rows. How long did it take for the problem to reach someone who could act. What happened to the person who raised it. Was the review an enquiry or an allocation of responsibility. Your answers place you on the table more accurately than any survey, because a survey measures what people are willing to report about their perceptions and an incident measures what the organisation did.

Psychological safety is a residue, not a policy

Amy Edmondson's work on psychological safety is now widely cited and almost as widely misapplied. The misapplication is to treat it as a quality that can be established by declaration — stating that this is a safe space, running a session on vulnerability, putting it in the values.

Psychological safety is the accumulated residue of what people have watched happen to colleagues who spoke up. Nothing else contributes to it meaningfully. A person deciding whether to raise a concern in a meeting is not consulting the values statement; they are recalling what happened the last time someone did, and if that memory is bad then no amount of invitation will overcome it.

This means it is built and destroyed in specific, identifiable moments, and there are fewer of them than you might expect.

The first response to bad news. Not the considered response an hour later — the visible reaction in the first few seconds. Everyone in the room reads it and calibrates.

The incident review. Whether the output is a set of system changes or a set of process controls that constrain the person involved. The second is read, correctly, as blame with a procedural face.

What happens to the person who was at the controls. If they are quietly moved, excluded from the review, or subsequently passed over, the organisation learns within a week.

Whether a dissenting view survived. If someone disagreed with a decision and the decision went ahead unchanged, did the dissenter get a hearing and an explanation, or were they managed. This is the consultation obligation from decision rights and who breaks a tie, and failing it costs more than the decision.

Whether raising a risk early was rewarded or held against them. Many organisations formally encourage early risk-raising and informally treat it as a lack of confidence. People detect this immediately.

The asymmetry is important: safety is built slowly and destroyed quickly. One visibly punished messenger undoes a year of careful signalling, because a single negative instance is much stronger evidence than a sequence of neutral ones. The rational inference from one bad outcome is that the risk is real, whatever the base rate.

Incentives reward what they measure, not what you meant

Goodhart's Law is usually quoted as a warning about metrics. It is better understood as a description of how culture is actually set.

People respond to the incentive as written, not the intention behind it. This is not cynicism; it is a reasonable response to the fact that the written version is the one that determines outcomes. If you measure utilisation, you get started work. If you measure velocity, you get inflated estimates — the mechanism is laid out in velocity is not a performance metric. If you measure defects found in testing, you get fewer defects logged. If you reward heroic recovery, you get an organisation that is excellent at recovery and never invests in prevention, because prevention is invisible and recovery is celebrated.

That last one is the most common and least recognised. Firefighting is rewarded structurally even where nobody intends it: the person who worked through the weekend to fix the outage is visible, thanked, and remembered at promotion time. The person whose careful work meant there was no outage is not visible at all, because absence of incident generates no event. Over a few years this produces excellent incident response and chronic underinvestment in prevention, and it reads as a cultural preference for action.

The corrective is not to stop thanking people who fix things. It is to make prevention visible — to name, in the same forums and with the same seriousness, the work that stopped something from happening, and to ask in every incident review what would have had to be true for this not to occur and then to fund that.

The general rule for this layer: before setting any measure, describe the cheapest way to make the number good. If the cheapest way is not the behaviour you want, the measure will produce the cheap way, every time, and the people producing it will be responding rationally to what you asked for.

The three levers that actually move it

If culture follows structure, then a leader has three instruments that matter and a great many that do not. The three are what gets measured, what gets funded, and who gets promoted.

What is measured. Not for its own sake — because measurement directs attention, and attention is the scarce resource. An organisation that reports cycle time and flow efficiency weekly will have conversations about queues. An organisation that reports utilisation will have conversations about who is busy. Neither requires anyone to be told what to value; the measure does the work. This is the cheapest lever and the fastest to move.

What is funded. Budget is the most honest statement of priority an organisation makes, and everyone reads it accurately. A stated commitment to quality alongside no funded capacity for technical work is not a mixed message; it is a clear message that the words are decorative. If you want to signal that something matters, fund it in a way that survives the next cost pressure, because how a commitment behaves under pressure is what defines it.

Who is promoted. The strongest signal available and the slowest to act. Every promotion is a published statement about what this organisation values, read closely by everybody at the level below. If the people who advance delivered visibly against odds, you get an organisation optimised for visible delivery against odds, whatever the values framework says. If they are the ones whose teams work well and who leave systems better than they found them, that is what people pursue. The lever is under-used because it is slow and individually contestable, but it is the one nobody disbelieves.

Two further levers are frequently mistaken for the main event. Stories and language matter, but only when true: a leader repeating an accurate account of something that actually happened — a team that stopped work to fix a problem and was supported, a decision reversed when new information arrived — transmits real information about how the organisation behaves. An aspirational story about behaviour nobody has observed transmits that the leadership is out of touch. And your own visible behaviour is weighted heavily, because it is one of the few direct observations most people have of senior leadership. How often you overrode a team, how you reacted the last time a date slipped, whether you read the pre-read — these are watched carefully and generalised freely.

What to do with a culture survey

Given all this, the reasonable question is whether to bother measuring culture at all. Yes, but as an instrument rather than a target.

Treat the survey as a lagging indicator of structural decisions made six to twelve months earlier. When a score moves, look backward for the structural cause rather than forward to a communications response. A drop in a team's sense of autonomy usually traces to a specific approval that was reintroduced or a decision that was overturned, and finding that event is worth more than any number of follow-up workshops.

Resist the urge to set targets on survey scores. The moment a manager is assessed on their team's engagement score, the score stops measuring engagement and starts measuring their ability to influence a survey. Goodhart applies with unusual force here, because the instrument is self-reported and respondents know who is being assessed.

And read the free-text comments rather than the numbers. The numbers give you the temperature; the comments give you the mechanism, because they name specific events, and that is where the structural causes are.

Treat what you find as information rather than as complaint. The argument in resistance is information applies directly: a team reporting that they do not feel able to raise issues is usually reporting an accurate observation about a specific incident, not a general mood.

What to do on Monday

Take the last significant incident or failure in your area and read the write-up. Count how many of its actions change a system — a default, a control, an interface, an automated check — and how many add a process step, a review, or an expectation on individuals to be more careful. If the second number is larger, you have found the single most informative artefact about your culture, and it was produced by your own organisation without any prompting.

Then change one measure in your weekly review. Remove the one that rewards being busy and add one that rewards finishing. Say nothing else about culture for a quarter and watch what the conversations become. The measure will do more work than a communications plan, and unlike the communications plan, nobody can disbelieve it.