By the time information has been made suitable for an executive committee, the part that would have changed the decision has usually been removed.
When a serious operational problem is investigated after the fact, the same finding appears with striking regularity: somebody knew. Not everybody, and rarely anybody senior — but somebody, usually close to the work, understood a good deal about what was going wrong and often said so.
The question that follows is almost always framed as why didn't they escalate? It is the wrong question, and it puts the burden in the wrong place. In most cases they did raise it, in some form, to someone. The information entered the organisation's reporting system and did not arrive at the top.
The more useful question is structural: what does our governance system do to information as it travels, and what does it systematically discard?
The Strategic Context
Every reporting chain is a compression system. At each level, a larger volume of detail is reduced to a smaller summary so that the next level can absorb it. This is not a defect. An executive committee cannot process the raw operational detail of an enterprise, and any organisation of scale must compress or drown.
But compression is never neutral about what it keeps. It is optimised — deliberately or by convention — for a particular kind of output, and everything that does not fit that output is lost. Enterprise reporting is optimised for the summarisable: things that recur, that aggregate, that can be counted and trended, and that fit an established category.
Which means the information most reliably discarded is the information that is singular, anomalous, hard to categorise, and not yet countable.
That is an exact description of early warning. A problem that is well-understood, recurring and quantified has already matured. A problem that is odd, isolated and difficult to articulate is the one where intervention is still cheap — and it is the one the compression system is least able to carry.
H. Kent Craig, in his 2004 column for mechanical contractors, made an observation that reads as a courtesy but functions as a systems argument: learn as much from the labourer digging your ditches as from the chief executive. He was writing about respect. What he was describing is where a particular class of knowledge exists.
What Leaders Commonly Misread
That the problem is the willingness to speak. Considerable effort goes into psychological safety, speak-up culture and escalation channels. This work is genuinely valuable and it addresses only one of two failure modes. The other is that a person speaks, is heard, is even thanked — and the information still does not travel, because the next layer has no format in which to carry it.
That an open door solves it. An open door serves the small number of people who will use it and the small number of matters that will drive someone to use it. It is a bypass, not a system, and its existence often reduces pressure to fix the system it bypasses.
That the summary preserves what matters. Each layer's summary is produced by someone judging what the next layer needs. Those judgements are individually reasonable. Applied four times in succession, they produce an output that reflects what the organisation expects to hear more than what is actually occurring — not through dishonesty, but through the accumulated effect of reasonable editing.
That richer data restores what was lost. Better instrumentation captures more of what is already measurable. Anomalous operational knowledge is largely not measurable at the point it matters; it exists as a practitioner's sense that something is not behaving as it should. There is no dashboard field for that.
Reframing the Issue
The reframe: information flow is an engineered system, and most enterprises have engineered exactly one channel — the vertical summary — while depending on it to carry two entirely different kinds of information.
The two kinds have opposite requirements. Aggregate performance information needs to be compressed, because its value is in the trend; individual data points are noise. Anomalous signal needs to be preserved intact, because its value is precisely in the detail that makes it anomalous; compress it and nothing is left.
Running both through one channel guarantees the second is destroyed. Not degraded — destroyed, because the compression step that makes aggregate information useful is the same step that removes everything distinguishing an anomaly from a rounding error.
Organisations that handle this well maintain a second, deliberately uncompressed channel. It is usually narrow, low-volume and direct. Its purpose is not to duplicate reporting but to carry the small number of things that would be eliminated by summarising.
Difference as a Design Property
Craig offered a metaphor alongside his advice about learning from everyone. The strength of steel, he suggested, comes not from purity but from the minute "impurities" bound within it, which hold the structure in a resilient matrix.
The metaphor is apt and the metallurgy is not. Steel is an iron–carbon alloy; carbon is a deliberate addition rather than an impurity, and ferrous oxide is not a constituent of it. Craig placed the word impurities in quotation marks, which suggests he knew he was reaching. [FACT CHECK REQUIRED: the metallurgical claim as stated in the source is inaccurate; the metaphor is used here with the correction made explicit rather than repeated as fact.]
The underlying point survives the correction, and is arguably strengthened by it: what gives an alloy properties that pure iron lacks is precisely the presence of elements that are not iron. Uniformity is not strength. It is brittleness with good documentation.
Applied to information systems, this is not a sentiment about inclusion. It is a claim about reliability. A governance system in which every layer shares the same background, incentives and mental model will compress information the same way at every step, and will therefore be blind in a consistent direction — the most dangerous kind of blindness, because consistent bias is invisible from inside and produces no contradictory signal to investigate.
Difference in perspective functions the way redundancy functions in engineered systems: not as a courtesy, but as protection against correlated failure.
Where the Loss Is Most Expensive
Supply chain and logistics operations show the pattern with unusual clarity, because the distance between where knowledge exists and where decisions are made is both large and formalised. The illustration below is hypothetical.
The people who know first that a supplier is deteriorating are rarely in procurement. They are the people receiving the goods — noticing that packaging has changed, that documentation is arriving later, that the quality of communication has shifted, that a familiar contact has been replaced twice in a quarter. None of these are measurable. None fit a supplier scorecard. All of them are recognisable to someone with pattern experience of that supplier.
By the time the deterioration reaches a scorecard, it has become a delivery failure or a quality incident — which is to say, it has become countable, which is to say the window in which it was cheap to address has closed.
The organisation was not short of information. The information existed, in the right form, at the right time, in a place from which the reporting system could not carry it.
Decision Framework
Six tests of whether a governance system can hear from below.
| Test | Question | Failure signal |
|---|---|---|
| Second channel | Is there a route for uncompressed signal? | Everything travels through the summary |
| Distance | How many compression steps between the work and the decision? | More than three |
| Anomaly handling | What happens to something that fits no category? | Recorded as "other", or not recorded |
| Direct contact | Do decision-makers regularly hear unmediated from the front line? | Contact is arranged, prepared and represented |
| Model diversity | Do the layers share one mental model? | Same background, same incentives, same reading |
| Return path | Does the person who raised something learn what happened? | No feedback, so no reason to raise the next one |
The Return path test is the quietest and among the most consequential. Where raising something produces no visible result, people stop. The channel then appears healthy — low volume, no complaints — while having ceased to function.
From Strategy to Execution
Immediate. Take a recent operational surprise and trace it backwards. Establish who knew, when, what they did with it, and at which layer it stopped. That layer is the finding, and it is usually the same layer across multiple incidents.
Medium term. Establish a deliberately narrow uncompressed channel to the executive level — low-volume, high-signal, with a defined threshold for what belongs in it and a guaranteed acknowledgement. Its value depends entirely on staying narrow; the moment it becomes a general reporting route, it acquires a summary step and stops being a second channel. [Related article: Risk Is Not the Chance That Things Go Wrong]
Long term. Build genuine model diversity into the governance layers. This is slow work and it is the only structural protection against correlated blindness. It means deliberately including operational, technical and front-line perspectives in the forums where interpretation happens — not as consultation, but as participants whose reading of the evidence differs and is expected to.
Signals to Monitor
- Post-incident reviews finding that somebody knew. If this recurs, the finding is about the channel, not the incident.
- Escalation volume falling without operational improvement. Usually a dead return path rather than a healthier operation.
- Front-line contact that is arranged and prepared. A prepared visit conveys what the preparers decided to convey.
- Anomalies recorded as "other". The category itself is evidence that the system cannot carry the unfamiliar.
- Unanimous interpretation of ambiguous evidence. Where a governance layer consistently agrees on what mixed evidence means, it is likely sharing one model rather than testing several.
Questions for the Leadership Team
- In our last significant operational surprise, who knew first — and at which layer did it stop travelling?
- How many compression steps sit between the people doing the work and the people deciding?
- Does a person who raises an anomaly here find out what happened to it?
- When did any of us last hear something unmediated, unprepared and unsummarised from the front line?
- Do the layers of our governance system read ambiguous evidence differently — and if they never do, what does that tell us?
Closing Perspective
Executives are rarely uninformed. They are informed by a system that removed the unfamiliar in order to make the familiar legible, and that trade was made at every layer by people acting sensibly.
The organisations that catch problems early are not the ones with more reporting. They are the ones that recognised summarising as a design decision with a cost, and deliberately built a way for the small, strange, uncountable things to arrive intact.
Related article: Risk Is Not the Chance That Things Go Wrong
Related article: Why "The Principles Apply to Any Project" Is Only Half True
Related article: Accountability Without Authority: How Organisations Design Delivery Leadership to Fail
About the author
Kevin Jogin is Founder & Principal Advisor at EraNorth. Meet the Founder.