Shoot the Messenger, Kill the Transformation
Why Ron Westrum’s Typology of Information Flow Predicts Whether Your Organisation Will Learn From AI or Merely React to It
Every organisation claims to value transparency. Every leader says their door is open. And in almost every organisation, the people closest to the work know things that the people making decisions do not, and have learned that saying so carries a cost.
Ron Westrum, a sociologist at Eastern Michigan University, spent decades studying what happens when organisations encounter information they did not expect. His career began not with management but with the sociology of science; his early work examined how scientific communities respond to anomalous observations. In a 1982 paper, “Social Intelligence about Hidden Events,” he documented a pattern that would prove foundational: when a person perceives a phenomenon that their community deems implausible, the social context will cause the observer to doubt their own perceptions, leading to denial or misidentification of what they have seen. Biologists refused to believe the platypus existed. Physicians denied that child abuse was systematic. The pattern was consistent: the community’s prior beliefs determined which observations could be reported, not the observations themselves.
From anomalous science, Westrum moved to aviation safety, then to healthcare, and eventually to a general theory of organisational culture that would, decades later, become one of the most empirically validated predictors of software delivery performance in the world. The thread connecting all of it is a single proposition: the way an organisation processes information, particularly information about problems, is the defining characteristic of its culture and the primary determinant of its performance.
1. Information Flow as Cultural Diagnostic: The Typology
Westrum’s central insight is deceptively simple: you can classify an organisation’s culture by observing how it handles information, particularly information about problems, anomalies, and failures. In a 2004 paper published in BMJ Quality & Safety, he formalised this into a three-part typology.
In pathological cultures, information is a weapon. The leader’s preoccupation is personal power and status; this preoccupation is absorbed by the workforce, who operate with it in mind. Messengers of bad news are punished or marginalised. Failure is concealed. Responsibilities are shirked. New ideas are crushed. Bridging between departments is discouraged because information is a positional good; sharing it makes you less powerful.
In bureaucratic cultures, information is channelled. The leader’s preoccupation is departmental turf and procedural compliance. Messengers are tolerated but frequently unheard, because they did not use the correct form or the correct channel. Failure leads to new process, not to inquiry. Responsibilities are compartmentalised. Bridging happens reluctantly, if at all; it looks like “going outside proper channels.”
In generative cultures, information flows to where it is needed, regardless of hierarchy. The leader’s preoccupation is the mission. Messengers are trained and rewarded. Failure leads to inquiry. Responsibilities are shared. Bridging is actively encouraged; when a crisis occurs, people reach across departmental lines without waiting for permission, because the mission takes precedence over the org chart.
The critical observation is that the culture type is determined by what the leader prioritises. Westrum is explicit: pathological environments are caused when the leader puts stress on their own advancement. Bureaucratic environments emerge when leaders put departmental goals ahead of organisational ones. Generative environments emerge when leaders subordinate everything to the mission. The culture is not a mysterious emergent property that nobody controls. It is a direct reflection of what leadership actually rewards, punishes, and ignores.
Anthony Giddens would recognise this as structuration in action. The leader’s preoccupation creates rules and resources (structures) that shape the behaviour of the workforce; the workforce’s behaviour reproduces those structures in daily interaction; and over time the culture becomes self-reinforcing, appearing “natural” rather than produced. A pathological culture does not feel pathological to those inside it. It feels like “the way things work here.” This is why culture change is so difficult: the structure that needs to change is the same structure that shapes the behaviour of the people who would need to change it.
2. Hidden Events and the Sociology of the Unsayable
Westrum’s typology gains its depth from his earlier work on hidden events. The concept is straightforward but its implications are profound: many phenomena exist in the world that are systematically denied, ignored, or misclassified because they conflict with prevailing beliefs about what is possible or acceptable. Child abuse was a hidden event for decades; physicians encountered the evidence repeatedly but classified it as “accident” because the alternative was socially unacceptable. Aviation near-misses were hidden events until reporting systems were designed to surface them without punishment.
The organisational application is direct. Every organisation has hidden events: problems, risks, failures, and innovations that exist in the experience of the workforce but do not appear in management’s picture of reality. The culture type determines whether these hidden events surface or remain buried.
This is where Westrum connects to Argyris with some precision. Argyris identified “undiscussables”; topics that everyone knows about but nobody raises because raising them would violate the governing norms of the organisation. Westrum provides the cultural mechanism that produces undiscussables. In a pathological culture, problems are undiscussable because raising them is dangerous. In a bureaucratic culture, problems are undiscussable because raising them is procedurally difficult. Only in a generative culture does the cost of raising a problem fall below the cost of concealing it.
Carol Dweck’s mindset research illuminates the individual psychology. In a fixed mindset culture, reporting a problem risks being seen as the person who caused the problem, or as someone who lacks the competence to solve it quietly. The messenger’s identity is threatened by the message. In a growth mindset culture, reporting a problem is evidence of attentiveness and professional commitment. The same act; radically different social meaning, depending on the culture.
For AI transformation, hidden events are everywhere. Teams that have discovered that AI-generated code requires extensive reworking do not report this if the organisational narrative is “AI makes us faster.” Engineers who have found that AI tools work well for some tasks and poorly for others stay silent if the governance framework treats AI as a uniform capability to be rolled out at scale. The gap between the official story of AI adoption and the lived experience of the teams doing the adopting is a hidden event, and its size is determined by the culture type.
3. Training the Messenger: The Operational Test
Westrum introduced the phrase “training the messenger” to describe what generative cultures do differently. In most organisations, “shooting the messenger” is recognised as bad practice, but the alternative is assumed to be merely tolerating them. Westrum argues that tolerance is insufficient; it is the response of a bureaucratic culture. Generative cultures actively train and reward the messenger, because they understand that early, accurate information about problems is the organisation’s most valuable resource.
Consider the practical test. Someone stands up in a town hall and says, “This AI transformation strategy is not working for my team.” What happens next?
In a pathological culture: “You are not a team player.” The messenger is marginalised. Others learn the lesson: do not speak.
In a bureaucratic culture: “Please submit that through (some) feedback process.” The messenger is redirected to a channel designed to slow information flow to a manageable pace. By the time the feedback is processed, the moment for action has passed.
In a generative culture: “Thank you for telling us. What are we missing?” The messenger is rewarded with attention and the information is acted upon. Others learn a different lesson: speaking up produces response.
This is what Heifetz calls “protecting voices from below.” Adaptive leadership requires that the people experiencing the adaptive challenge, the people for whom the old way of working no longer fits, can surface their experience without being punished for disrupting the official narrative. Heifetz provides the leadership practice; Westrum provides the cultural precondition. You cannot protect voices from below in a pathological culture, because the power dynamics make protection impossible. You cannot protect them reliably in a bureaucratic culture, because the channels designed to process feedback are also channels designed to filter it.
Tom Peters would put it more bluntly: stop reading the dashboard and go sit with the team. Management by Walking Around, for all its simplicity, is an attempt to bypass the information-filtering mechanisms that bureaucratic and pathological cultures install between the leader and the reality of work. Peters’s instinct was right; Westrum provides the theoretical explanation for why it matters.
4. Bridging: Why Transformation Happens Between Silos
Westrum observed that in crises, standard communication channels consistently fail. The information needed to resolve the problem exists in the organisation, but it is distributed across departments that do not normally talk to each other. In bureaucratic cultures, this is fatal; the information cannot cross departmental boundaries quickly enough.
In generative cultures, people reach across departmental lines to solve problems without waiting for permission. Westrum calls this bridging.
Bridging is not networking. It is not the informal coffee-machine conversations that management consultants celebrate. It is the deliberate, mission-driven act of connecting information across boundaries that the organisation’s formal structure has created. It requires two conditions: first, the person bridging must believe that the mission is more important than the org chart; second, the organisation must not punish them for “going outside proper channels.”
This is the cultural foundation for what Matthew Skelton and Manuel Pais describe as collaboration mode in Team Topologies. Their framework assumes that teams will sometimes need to work closely together across boundaries, particularly when building new capabilities that do not fit neatly within existing team structures. But collaboration mode is expensive and cognitively demanding; it only works when the culture supports it. You cannot have fluid team interaction if the culture treats crossing a boundary as an act of insubordination.
Richard Normann saw the same dynamic at the strategic level. His shift from value chain to value constellation assumes that value is co-produced across a network of actors; but a network only functions when information flows freely between its nodes. A value chain can survive bureaucratic culture because it is sequential; each link only needs to communicate with the link before and after it. A value constellation requires generative culture because every node needs to communicate with every other node as conditions change.
For AI transformation, bridging is not optional. AI adoption inherently crosses the boundaries between business domains and technology teams, between the people who understand the problem and the people who understand the tooling. The specification-driven approach that this series describes, where domain experts specify intent and AI generates implementation, is a bridging activity. It requires business analysts, engineers, and domain experts to work together in ways that most organisational structures actively discourage. If the culture punishes bridging, the approach fails regardless of the quality of the tooling.
5. From “Soft” to Statistical: The DORA Evidence
For years, Westrum’s typology was respected in safety circles but dismissed in mainstream management as “soft.” Culture was something you talked about at offsites, not something you measured or managed. The State of DevOps Reports, produced by the DORA (DevOps Research and Assessment) team and synthesised in Forsgren, Humble, and Kim’s Accelerate (2018), changed this decisively.
Using Westrum’s model as a measurement instrument, the DORA research demonstrated statistical relationships between generative culture and organisational outcomes. The survey items are deceptively simple:
“On my team, information is actively sought.”
“On my team, failures are learning opportunities, and messengers of them are not punished.”
“On my team, responsibilities are shared.”
“On my team, cross-functional collaboration is encouraged and rewarded.”
“On my team, failure causes enquiry.”
“On my team, new ideas are welcomed.”
Respondents rate agreement on a Likert scale; the aggregate score predicts performance.
The findings were striking. Generative culture, as measured by these items, is predictive of higher software delivery performance across all four DORA metrics: lead time for changes, deployment frequency, mean time to restore service, and change failure rate. It is also predictive of higher organisational performance measured by profitability, market share, and productivity. And it is predictive of lower burnout and higher job satisfaction among the people doing the work.
The DORA research also confirmed a finding that Deming had insisted on for decades and that most organisations still refuse to believe: speed and stability are not trade-offs. High performers achieve both shorter lead times and higher deployment frequencies while simultaneously maintaining higher stability and lower failure rates. The “move fast and break things” framing and the “move slowly and maintain control” framing are both wrong. The organisations that move fastest are also the most stable, and the cultural precondition for both is generative information flow.
This is the empirical validation that connects Westrum to every thinker in this series. Argyris’s double-loop learning requires that governing assumptions can be surfaced and questioned; generative culture is the environment where this becomes possible. Weick’s High Reliability Organisations exhibit generative characteristics by definition: deference to expertise, preoccupation with failure, and reluctance to simplify are all features of generative information flow. Stacey’s emphasis on the quality of conversational interaction as the source of organisational change presupposes an environment where honest conversation is possible. Westrum provides the measurable, predictive variable that underlies all of their frameworks.
6. Westrum and His Limits: Culture Without Structure
Westrum must be read with his limitations visible. His typology is a classification, not a causal model; it tells you what generative culture looks like but is less precise about the mechanisms that produce it or the interventions that shift an organisation from one type to another. The three types are ideal categories; most real organisations contain elements of all three, often varying by department or level.
More significantly, Westrum’s framework operates primarily at the level of information flow and leader preoccupation, and is less attentive to the structural conditions that constrain both. Giddens would note that culture is not independent of structure; it is produced and reproduced through structural conditions that include incentive systems, reporting lines, budget allocation, and the material organisation of work. A leader who wants to create a generative culture but operates within an incentive structure that rewards departmental optimisation will find that the structure overwhelms the intention. Westrum tells you what kind of culture you need. He is less helpful in telling you how to change the structures that are producing the culture you have.
Heifetz would add that shifting from bureaucratic to generative culture is itself an adaptive challenge, not a technical one. There is no procedure for “becoming generative.” The change requires people, including leaders, to give up familiar sources of authority and control. The bureaucratic leader’s power depends on being the gatekeeper of information; generative culture dissolves that gatekeeping function. The adaptive loss is real, and Westrum’s framework does not adequately address it.
And Stacey would observe that “creating a generative culture” is itself a paradoxical instruction. Culture is not something leaders create; it is something that emerges from the ongoing patterns of interaction among everyone in the organisation. A leader who announces “we will now have a generative culture” has issued a bureaucratic instruction to become non-bureaucratic. The paradox does not make the aspiration worthless; it makes the path more complex than the typology suggests.
Despite these limits, Westrum provides something that few organisational theorists offer: a framework that is simultaneously grounded in decades of research, simple enough to be immediately useful as a diagnostic, and empirically validated as a predictor of outcomes that leaders actually care about. That combination is rare, and it is why Westrum’s typology has traveled further than most academic frameworks; from sociology to aviation to healthcare to software engineering to the broader question of how organisations learn.
(An Organisational Prompt is something you can do now…)
Organisational Prompt
Westrum’s insight is that information flow is not merely a feature of culture; it is the culture. And the single most revealing test of information flow is what happens when someone delivers bad news.
In your next team meeting or retrospective, ask this question: “What is one thing about our AI adoption that we all know is not working but nobody has said out loud?” Then be silent. Wait. The silence will be uncomfortable. That discomfort is the sound of the culture being tested.
If someone speaks, respond with curiosity, not solution. Ask: “What do you see that we are missing?” Then ask the rest of the team: “Does this match your experience?” Do not defend. Do not explain. Do not redirect to a feedback portal. The act of asking, listening, and visibly valuing the response is the smallest unit of generative culture. It is also Argyris’s double-loop learning in practice: you are not fixing the problem. You are changing the conditions under which problems can be discussed.
If nobody speaks, that silence is also data. It tells you where you are on Westrum’s typology more accurately than any survey.
Further Reading
Ron Westrum: A Typology of Organisational Cultures (BMJ Quality & Safety, 2004). The short paper that launched a thousand DevOps presentations. Read it for the typology itself and for the foundational insight that information flow is both influential and indicative of other aspects of culture; it predicts how organisations will behave when trouble arises.
Ron Westrum: The Study of Information Flow: A Personal Journey (Safety Science, 2014). Westrum’s own account of how his intellectual journey from the sociology of anomalous science to aviation safety to organisational culture came together. Read it for the deeper context behind the typology and for the connection between “hidden events” in science and hidden problems in organisations.
Nicole Forsgren, Jez Humble, and Gene Kim: Accelerate: The Science of Lean Software and DevOps (2018). The empirical validation of Westrum’s model in software delivery. Read it for the statistical evidence that generative culture predicts both speed and stability, and for the DORA metrics framework that makes organisational performance measurable.
Gene Kim: The Unicorn Project (2019). Kim’s novel dramatises Westrum’s typology through the “Five Ideals,” with the Third Ideal (Improvement of Daily Work) and the Fourth Ideal (Psychological Safety) drawn directly from Westrum’s generative culture. Read it as the narrative companion to Accelerate‘s data.
Disclaimer
I write about the industry and its approach in general. None of the opinions or examples in my articles necessarily relate to present or past employers. I draw on conversations with many practitioners and all views are my own.
"when a person perceives a phenomenon that their community deems implausible, the social context will cause the observer to doubt their own perceptions, leading to denial or misidentification of what they have seen." - interesting.