In 2000, Cisco was selling into severe shortages. Customers who expected partial fulfillment ordered more equipment than they needed. As orders and backlog surged, Cisco and its suppliers expanded supply in response. By the time that supply arrived, the shortage had eased, protective orders disappeared, and apparent demand collapsed.
A Theory of Enterprise Coherence
Companies are dumber than the people who run them.
How intelligently a company behaves is a systems property, and systems can be improved.
Enterprises often behave stupidly even though smart people act rationally within capable systems.
Companies have their own cognition that is more than their people, more than their technology, and more than their AI.
Company cognition is really good at solving some types of problems and really bad at others.
Because companies are good at the easy stuff, the remaining value frontier tends to be the messy stuff.
Small increases in enterprise cognition are worth billions at enterprise scale, especially for the messy stuff.
AI accelerates some cognitive machinery. Whether the company gets smarter depends on how AI integrates. Misconfigured, it makes the company stupider faster and more convincingly.
Smart people, dumb companies
Given the informational scale, logistical burden, and dynamic complexity involved, it is remarkable that large organizations function as well as they do.
Many dumb behaviors result from smart people and competent systems doing the rational thing from each local perspective and producing something stupid at enterprise scale.
Sony entered the digital music era with the Walkman franchise, elite consumer electronics capability, and its own major music business. Sony’s hardware business pushed toward connected devices and hard-drive players. Sony Music wanted tighter control over copying. In trying to placate both, Sony built products around proprietary formats, restrictive rights management, and cumbersome software that made digital music harder to load, move, and use.
Both Cisco and Sony misread reality in consequential ways while intelligent people inside each company acted rationally on what they saw.
Definition
Rational incoherence describes behavior that makes sense locally even as it creates materially worse outcomes for the enterprise.
The stupidity does not necessarily belong to any person, function, system, or individual decision. It emerges from how locally rational actions, systems, and judgments combine, and from whether that combination fits the reality the enterprise is navigating.
Calling a company smart or dumb only makes sense relative to what it is trying to accomplish. Every enterprise is purposive, and purpose does not contain its own path to realization. Reality hands the organization a set of opportunities and constraints.
How well the organization navigates reality is the deciding factor for how well it realizes its goals.
Cohere to reality+
When an enterprise behaves in dumb ways, we call that incoherence. Coherence is the degree to which a company’s behavior fits the reality governing achievement of its purpose. Incoherence is misfit behavior.
Internal consistency is a legitimate but narrower definition of coherence. Internal consistency asks whether everything inside a system agrees. Coherence, as we define it in this work, asks whether the system fits reality to achieve its purpose. A company can be internally consistent and still incoherent with reality.
- If I need to get somewhere in a hurry and I slash my own tires, that’s incoherent.×
- If I take the wrong freeway, that’s incoherent, even if I believed it was the correct one.×
- If I take the fastest route and a drunk driver crashes into me, that’s still coherent, because randomness exists.✓
Coherent action gives us our best shot at the outcome we want. It never guarantees it.
Incoherence may be obvious, visible only in hindsight, or never visible at all. In many everyday human situations, incoherence tends to be blatant. In enterprise, it can be really difficult to spot, especially in real time.
Coherence is judged at the enterprise level. For example, if a department advanced its own agenda while materially harming the company, that would be incoherent.
Read the (reality) room+
Companies pursue growth, margin, resilience, service, safety, capability, and strategic freedom.
Relative to its aims, the enterprise must: read opportunities, obstacles, and options; discover or construct a viable path; commit; act; observe results; revise when new evidence warrants.
At every stage, the enterprise works from some treatment of the reality bearing on its purpose.
The different ways a want can meet reality
Possibility. How achievable something is. Reality may make a want easy to fulfill, difficult, conditionally possible, or impossible. It may afford one viable route, several, or none yet visible.
Legibility. How obvious the route is to a goal. A viable path may be obvious, partly visible, hidden, or discoverable only through inquiry, experimentation, or invention.
Probability. How likely something is to work. Even a known and viable path may be uncertain to deliver. Possibility does not imply certainty.
| Want | Possibility | Legibility | Probability |
|---|---|---|---|
| Scratch an itch on your belly | Easy | Obvious | Near-certain |
| Make a three-pointer | Achievable | Obvious | Somewhat |
| Improvise the CO₂ scrubber adapter during Apollo 13 | Achievable within severe material constraints | Non-obvious; a viable configuration had to be constructed from the materials on board | High once developed and tested |
| Get an object out of a tree branch slightly out of jumping reach | Achievable | Several obvious routes | High |
| Lift a 2,000 lb boulder with brute force, alone | Impossible | Obvious | N/A |
| Execute a movie-style heist | Conditionally possible | Illegible initially; clever criminals may devise a path | Low; many ways to go wrong |
| A fly buzzing against a windowpane with an open window nearby | Impossible via current route; easy via alternate route | Obvious to an observer, not to the fly | Current route: none; adjacent route: certain |
At enterprise scale, reality is distributed and largely illegible. More on this below.
Which parts of reality matter to achieving a want?
Reality is vast. Almost all of it is irrelevant to any particular purpose.
The enterprise has to exclude nearly all of reality while retaining what actually bears on realization. That boundary is not static. Conditions that matter little at one stage may become decisive later, and action itself can change which conditions matter.
We don't get to decide what bears on realization. Reality does.
Three questions matter most:
- Which parts of reality actually bear on realization?
- Which of those parts can we identify and know, and which remain unavailable?
- Given those limits, how much confidence is justified?
If you want a glass of water in your kitchen, the relevant reality is narrow: the faucet, the glass, whether water is available, whether it is drinkable. The orbital velocity of Pluto and the mating patterns of the Tibetan yak can be ignored with confidence.
If you want a successful hiking vacation in the Dolomites, the boundary expands. Weather, airline logistics, trail conditions, lodging, health, transportation, and other conditions matter. Some are knowable in advance. Others remain uncertain. And some consequential conditions may never become visible until they affect the outcome.
The enterprise faces this filtering burden at far greater scale: retain the conditions that can materially affect realization, exclude the rest, distinguish evidence from uncertainty, and calibrate confidence accordingly.
At enterprise scale, reality arrives in fragments and those fragments are received by a distributed cognitive architecture.
The brainless company
The enterprise has no brain. There’s no main processor of company intelligence. No technology is the company’s cognition. And its humans aren’t it, either. Yet the enterprise clearly manages to make decisions, take action, and carry on.
So where does the thinking happen?
Nobody thinks for the company
The CEO doesn’t think for the company. The company thinks through the CEO.
The same is true for every role. Human intelligence does not sit above the company and think on its behalf. It is one part of what the enterprise thinks with.
By the time something reaches an executive, the enterprise has already thought it into shape. The executive does not receive reality raw, but as something the enterprise has already predigested. Then the enterprise uses the executive to keep on thinking.
Predigestion is necessary. If every consequential judgment arrived with the full underlying reality attached, the cognitive burden of a single decision would be incomprehensible.
But every act of predigestion is also an act of selection. A financial model preserves some features and discards others. So does an ERP schema, a KPI, a risk category, a product taxonomy, an organizational boundary, or an expert’s mental shorthand. These all have excluded massive amounts of reality.
The same machinery that makes the company capable also makes certain relationships, conditions, and possibilities invisible.
Additional synthesis does not restore information discarded earlier. An executive team can intelligently combine the representations that reach it while never seeing a consequential relationship, condition, or possibility excluded upstream.
The company uses you+
From the company’s perspective, people are part of its machinery. The enterprise uses human intelligence as one component of its massive thinking capability.
You’re being used by your company, even if you are the CEO. Your role and function are necessary ingredients to its cognition.
And while the company is using an executive to think, the executive can change how the company thinks. Executive roles occupy that privileged place in the machinery of cognition that can alter the machinery itself, and can even alter how the machinery thinks about the machinery.
In that case, you’re being used to improve enterprise thinking so that the company can more efficiently use you in pursuit of its goals.
Since your compensation and career improve from the company’s improvement, you should be happy to be used in this way.
Nobody knows what the company knows
No one could inventory everything an enterprise knows. Its knowledge is distributed across people, systems, operations, history, and relationships. No single view or catalogue contains it.
The company’s knowledge does not fit inside it.
The company thinks beyond its formal boundary via suppliers, partners, customers, and external systems.
Some of what the company knows isn’t known anywhere.
A company contains many facts that live in a specific place and can be looked up. A company knows other important things by inference and interpretation only. These things might not be captured and stored, even though many of the supporting facts are.
People inside the company know things the company doesn’t.
Private human thoughts are not enterprise cognition until they participate in the cognitive machinery of the company. The enterprise’s people know many things that the enterprise should know, but doesn’t.
The company has everything required to know something, and still doesn’t know it.
The company possesses every input for an important inference to occur, but the inference never occurs.
Knowing is only part of thinking
Enterprise cognition involves much more than “knowing” things. It’s an entire cycle by which the enterprise navigates reality to achieve outcomes that meet its purpose.
Encounter→Representation and interpretation→Form possibilities and option sets→Judgment→Commitment and configuration→Action→Observation→Learning and revision
This is not necessarily a linear sequence. The functions overlap, recur, and happen at different scales throughout the enterprise. A production change might simultaneously act on reality, reveal something the company did not know, test an assumption, and alter the next decision.
Nor does the enterprise perform each function in one place. Different people, systems, representations, technologies, routines, physical processes, and other machinery participate in different combinations at different times.
The enterprise cognitive apparatus is the machinery through which these functions happen.
Go deeper on enterprise cognition
The recurring functions, their machinery, and the relationships among them are developed further in The Enterprise Cognitive Apparatus.
Unintelligent BI+
Business intelligence is not enterprise cognition. A company’s immaculate dashboards might propagate stupidity and blindness if they’re not calibrated to the reality that matters. Dashboards sit downstream of assumptions, selections, representations, and other predigested cognition. They can make a good picture clearer, or a bad picture more convincing.
At its best, BI participates in enterprise cognition while remaining one component of it.
Artificial (non) intelligence+
AI isn’t Ozempic for enterprise cognition. It’s not the easy button that makes enterprises more intelligent overall. It improves the speed, fidelity, and economics of specific cognitive functions.
It can find information faster, interpret more of it, compose across larger bodies of evidence, generate possibilities, test alternatives, and carry forward more context than humans can manage unaided.
But enterprise intelligence does not live inside any one of those capabilities. AI can also make the company wrong faster, at greater scale, and with better arguments.
Better cognition inside the wrong frame is still incoherent.
I love AI and think it has an enormous role to play in enterprise cognition. The danger is mistaking that role for enterprise cognition itself.
Some thoughts weigh 20,000 pounds
Can a CNC machine think? No, but the company thinks through it. The machine doesn’t need a mind, because it’s already part of the company’s cognitive apparatus.
An enterprise has no separate layer where thinking happens. It thinks through the same distributed system through which it acts. Physical operations are part of that system.
Improve how the company thinks, and you improve how it acts.
Thinking is action, but not all action is thinking.
All enterprise cognition occurs through enterprise activity. Much activity happens that delivers no cognitive value.
For activity to participate in cognition, something about it has to become available to the company’s thinking. It needs to be measured, observed, captured, compared, interpreted, inferred from, or otherwise carried forward.
A person walking 50,000 steps a day across a factory floor generates activity. If those movements are recorded and used to expose wasted travel, the same activity also participates in cognition.
A plant reverses two production steps and throughput jumps. That can remain nothing more than an operational event. Or the result can be noticed, interpreted, carried forward, and used to change how the plant sequences work in the future. The same action has now contributed to what the enterprise knows and how it operates.
Paying for lessons you never learn
Companies contain enormous unharvested cognitive potential.
The enterprise is already paying for the activity in which potential cognition is embedded. Missed opportunity exists in information the company failed to collect and the activities that could have made knowable, inferable, or learnable.
When something useful dies where it occurs, the company may have to rediscover it later, rediscover it somewhere else, or never know it at all.
The gap between what enterprise activity could contribute to cognition and what actually does is the cognitive harvest left on the table.
Companies get stupid by becoming smart
Enterprises deal with a massively complex reality by making it manageable. They narrow focus, group things, compress detail, divide into separate domains, standardize recurring responses, specialize, and build simpler representations of reality. These reductions are good enough for the job at hand, until they aren’t.
A schematic is one kind of reduction. There is no schematic of a building that includes the whole building, because that schematic would be too noisy to be useful. An electrician needs one schematic, a plumber another, and a structural engineer another. Problems happen when schematics are inaccurate or when a trade has a different trade’s schematic.
Making enterprise reality manageable gets far more squirrely than representing a building’s electrical systems.
The machinery that makes a company capable also shapes the reality it becomes capable of seeing.
To see why, let’s take a little detour into WWII.
Did World War II actually happen?+
Obviously, WWII happened. Also, it sort of didn’t. A lot of events happened, and people put a label on a bundle of those events and called it WWII. But the underlying reality is a lot more unruly than the label suggests.
A general, a diplomat, and an epidemiologist walk into a bar.
Ask a general about WWII and you’ll get strategy, troop movements, engagements, terrain, casualties.
Ask a diplomat about WWII and you’ll get alliances, commitments, leverage, posturing, communication channels, spheres of influence, etc.
Ask an epidemiologist about the war and you’ll get sanitation, nutrition, migration, crowding, etc.
Ask a human rights advocate, intelligence officer, economist, nutritionist, or civic planner, and you’ll get very different accounts of the same bundle of events.
But it gets even weirder than “different perspectives.”
What’s part of the war and what isn’t?
Imagine a woman wrapping a Christmas present in 1941 rural Idaho. Is her activity part of the war? Most would say no. Now let’s play with some details to show how fuzzy the boundary of belonging is.
- 01
The Christmas present is a care package for her son fighting in the Philippines.
- 02
The Christmas present is a book for her brother, who happens to be a general. He reads it, gets an idea, and changes strategy in a way that alters the war.
- 03
Now place the woman in occupied Paris, and she’s wrapping a wooden puzzle for a neighbor’s child.
- 04
Now imagine that she is part of the resistance and a coded message about German troop movements is hidden inside the puzzle. Instead of giving it to the neighbor’s child, she’s delivering it to a local cafe, where the proprietor will pass it on to an intelligence officer in the resistance.
- 05
Now imagine the message never gets delivered because the cafe owner is arrested.
- 06
Now imagine it does get delivered but changes nothing.
- 07
Now imagine it alters some important outcome.
Across all of these counterfactuals, when do the woman and her actions become “part of the war” vs “not part of the war”?
If she’s a fierce resistance fighter vs oblivious to her participation, does that change whether her actions count as part of the war?
If a group of historians agree that this woman’s efforts were decisive to the Allies winning the war, does that make her actions more “in” than if nobody remembers her existence?
There’s no “WWII” label attached to any of this, and there’s no cosmic authority that draws a boundary around the occurrences that belong in the category of WWII vs outside it.
Some of the most decisive facts to the war may not even be part of the war
Factors that influenced WWII may have originated far outside any conventional idea of the war’s boundary.
- ·
Steel production in Ohio and wheat production in Kansas supported the war
- ·
Ancient plankton formed petroleum that fueled the war
- ·
Tectonic activity made geologic features that later became decisive to key battles
- ·
A country’s literacy rate and education system determined the available talent pool for specialized roles in the war
Reality is so entangled that the details that determined the war’s outcome sprawl far beyond the thing we call WWII.
Nobody experienced WWII, only pieces of it
No contemporaneous person had a vantage point from which to observe WWII. Millions of them individually experienced a slice of related reality.
- ·
A soldier encountered a beach, a shell, an officer, a meal, a wound.
- ·
A diplomat encountered governments, cables, negotiations, threats, promises.
- ·
A factory manager encountered workers, materials, quotas, shortages.
- ·
Churchill, Roosevelt, Stalin, and Hitler each encountered partial slices of reality plus representations assembled from other partial slices.
WWII was a massive composite made of fragments spread across time and space. The war as a “thing” could not be located by anyone.
Scale changes relevance
A rock can be everything to one soldier in a firefight and of little importance to the war. A theater strategy can matter hugely to the war outcome and have little relevance to the soldier behind that rock. Relevance changes with the altitude of the thing you’re trying to understand.
Translation to enterprise reality
Just as “WWII” is an artificially tidy reduction for a bundle of events, enterprises must reduce complex reality into tractable objects and views.
A useful reduction tells us what to ignore. It does not make the ignored reality disappear.
It’s smart to dumb things down
We need to dumb reality down in order to talk about it, react to it, collaborate on it, and pursue goals.
“World War II” is a useful dumbing down that lets us reference a cluster of events as a single event in history.
Specialization is a really efficient tactic, most of the time.
Google Maps doesn’t narrate every fire hydrant, building, and blade of grass. It dumbs reality down to the turns I need to make, and I’m glad for that.
It’s really useful to dumb things down in this way. In fact, it’s the only way to operate.
Smart dumb vs dumb dumb
When you dumb things down and that dumbing helps achieve one’s purpose, that’s smart. Or as we say, coherent.
An electrical schematic is coherent because it helps electricians do their job. The inclusion of plumbing, structural, and a thousand other irrelevant details would be incoherent even if they’re accurate because they’d make the electricians’ job harder.
When you dumb things down in a way that thwarts a purpose, that’s dumb or incoherent. To understand why incoherence happens so often in enterprise, we need to explore systemicity and why it makes reality so difficult to reduce without losing something that matters.
When reality refuses to cooperate
Enterprises organize matter and intelligence at staggering scale because their reductions work so well, so much of the time.
As systemicity rises, those reductions are more likely to become incoherent. Think of systemicity as the inherent squirreliness of complex systems. It shows up as interdependence, configuration dependence, path dependence, feedback loops, emergence, and other forms of dynamism. Systemicity occurs in degrees. As it rises, simplification becomes more likely to discard relationships, states, sequences, or feedback that materially affect the outcome.
To see why systemicity isn’t just complexity or difficulty, consider a thought experiment involving the assembly of a 747.
747 assembly and systemicity vs complexity vs difficulty+
Imagine assembling a specified 747 with complete schematics, verified procedures, exact interface specifications, and an established sequence, but with every component scattered across a square-mile hangar and no indexing system. Finding a single bolt might take months. Complexity and difficulty are enormous. The systemic burden is comparatively low because the required configuration, dependencies, and sequence are already known.
Now make every component instantly retrievable but remove the schematics, procedures, and instructions. The search difficulty is gone. The assembler must now reconstruct component dependencies, assembly sequence, interface constraints, and permissible configurations. The systemic burden has increased dramatically even as one major source of complexity and difficulty has disappeared.
And even this more systemic version is still relatively tame compared to enterprise reality. The aircraft specs don’t change while you assemble it. The parts do not learn, adapt, retaliate, change their interfaces, or alter the goal. Effects do not arrive months after their causes. Small changes do not suddenly cross thresholds and behave differently. One subsystem does not quietly change the conditions facing another. The assembly process does not change the evidence you use to understand the aircraft. There are no customers changing behavior, competitors reacting, policies changing incentives, feedback loops changing later conditions, or yesterday’s decisions reshaping what is possible tomorrow. Much of the systemicity that makes enterprise reality so unruly still isn’t present.
A Swiss watch is complex.
An Ironman is difficult.
The weather is systemic.
The biggest opportunities are invisible and systemic
Companies are good at harvesting opportunities that fit their reductions of reality, which means that the biggest remaining opportunities for value capture tend to sprawl across the boxes the company has put reality into.
Not only are they messy, but these opportunities are invisible to the company’s reductive cognition. And they’re invisible in plain sight. To understand why, here’s a silly thought experiment about cars.
Counting Honda Civics+
Imagine you pay someone to sit on an overpass and count the number of Honda Civics that pass under in a day. They take a faithful tally and then you ask them how many Subaru Outbacks went under in the same interval. They’ll have no idea.
This illustrates selective blindness, but it gets much worse in enterprise selectivity. The car counting analogy starts to fail us, but imagine that solutions involve noticing subtle patterns of lane change, signal use, directionality, etc. Our Civic-counting hire will be even more oblivious.
When selective blindness gets expensive
Enterprises are missing out on much of what reality offers because it is invisible to their cognitive apparatus.
Here’s an example from manufacturing (we see this sort of thing a lot):
A machining cell has mediocre OEE, so managers work to improve the metric because OEE is the “Honda Civic” of this cell. Changeovers come down. Runs get longer. Utilization rises. Cost per unit improves. By the cell’s own measures, improving OEE is working exactly as intended.
Look at the larger system, and that same improvement (sometimes) becomes incoherent against enterprise purposes.
Longer runs send material downstream in larger waves. Inspection gets slammed, then goes quiet. Some jobs sit behind large batches that are efficient for the machine but badly timed for everything that follows. Work-in-process grows. Lead times stretch. Other operations alternate between waiting and overload.
OEE is a coherent simplification of machine performance. The bigger, messier thing that would actually improve enterprise outcomes is invisible and systemic.
Don’t trust reductions too much
A reduction might be coherent for a limited purpose and incoherent for another, larger purpose. Under high systemicity, excluded relationships usually decide the outcome.
A representation contains accurate facts and disciplined reasoning and still supports the wrong decision because the governing relationship sits outside the frame.
The failure may become obvious, or may remain invisible because the enterprise’s standing machinery never generates or perceives the better option.
A useful reduction becomes dangerous when the company keeps trusting it after omitted reality starts changing the outcome.
When should the company look again?
Be less wrong
In business, you wish you could have 100% coherence all the time. Impossible.
Coherence means doing what reality makes available to achieve a purpose. In simple situations, full coherence is often trivial. In highly systemic situations, we’re shooting for degrees of coherence.
At enterprise scale, even a marginal increase in coherence is worth a fortune.
The tent stake and the acquisition+
I’m camping and need to stake my tent down in a high wind. I’ve forgotten my hammer. I look down, and there’s a nice, fist-sized rock. I pick it up and pound in my stakes. I accomplish my purpose based on what was available. Full coherence. No systemicity.
Now let’s say I’m the CEO of a $10b company that’s acquiring a $2b company. To have perfect coherence I’d need a God’s-eye view of billions of factors, plus the ability to forecast how they all interact dynamically and the capacity to enforce a detailed regimen across the entire configuration. Even though perfect coherence is impossible, incremental coherence is worth a fortune.
Practical coherence means finding consequential misfit the enterprise can reasonably discover and correct.
Two incoherent responses to systemicity
Business as usual.
Trust the dumbed-down reductions.
Paralysis by analysis.
Avoid making the choice (which is also a choice).
Know when to look again
At enterprise scale, you need to run on trusted reductions without always questioning them. Reopening should happen when dealing with an obviously systemic question or decision, because that’s when you have a high likelihood that the reductions are missing something important.
Repeated exceptions are another signal to delve deeper. When forecasts miss in the same direction, workarounds multiply, an intervention improves its target while the enterprise outcome does not, or every contradiction gets explained away as unusual, the frame itself deserves scrutiny.
Cross-boundary economics gives another reason to reopen. If the value of a move changes materially when an adjacent workflow, system, authority rule, customer behavior, or constraint changes, the administrative boundary may be too small for the judgment. The same warning applies when an option looks unattractive only because the evaluation assumes the current configuration stays fixed.
Hardening raises the cost of being wrong. An omitted relationship matters more when a judgment is about to become an acquisition, architecture, contract, facility, organizational structure, or standard that later decisions will inherit.
None of these proves the current frame is wrong. They are reasons to stop granting it automatic authority.
Practical coherence
Once a frame deserves another look, reopen only what is likely to govern the outcome. You may need to redraw a boundary, revise a representation, recover a relationship lost through decomposition, construct an option the standing machinery never generated, or let new evidence overturn a judgment already being realized.
Practical coherence is not:
- ×
Question everything (who has time for that?)
- ×
Break down the silos (sometimes the silo is useful)
- ×
More data (not helpful when the failure lies in inference, interpretation, or a missing relationship)
- ×
More analysis (useless if you’re analyzing the wrong thing)
- ×
Map everything (impossible, and mostly unnecessary)
The enterprise does not need permanent holism. It needs to recognize when an ordinary reduction has stopped preserving something significant to a specific decision or outcome.
Find the economic object
Projects need scopes. Budgets need owners. Systems need names. Accountability requires bounded objects.
Reality doesn’t respect those administrative boundaries. The relationships that matter probably spill over.
Calling an initiative an AI implementation, ERP project, acquisition integration, capacity expansion, or workforce initiative tells us how the company has chosen to manage it. The label does not establish the full set of relationships that determine whether the initiative creates value.
The right boundary is the smallest one that preserves the relationships capable of materially changing the economic judgment.
If the value of an AI investment depends on workflow, authority, data, another system, and retirement of an existing process, those conditions belong inside the economic object even when they are outside the project plan.
If an adjacent condition cannot materially change the economics, leave it outside.
Reduce aggressively, but justify every boundary.
Enterprise-scale consequence does not require enterprise-scale intervention.
Widening the frame may reveal that the highest-leverage change is one interface, one authority rule, one dependency, one sequence, one representation. A broad transformation might miss the binding limitation just as easily as a narrow project can.
Understand the system broadly enough to find the intervention that actually changes the economics.
Go deeper on systemicity
The relational structure that makes boundaries economically consequential is developed further in Systemicity.
Sometimes you should change the system instead
Some situations become easier to navigate by changing the system rather than understanding it more completely.
That may mean removing a dependency, changing an interface, improving observability, delaying a decision until relevant evidence exists, or making a commitment easier to reverse. The enterprise may even be able to eliminate a constraint it was preparing to forecast more accurately.
The gain comes from reducing how much unresolved reality must be successfully navigated, not from understanding the old configuration more perfectly.
AI and M&A expose two different limits of the ordinary frame. AI is often evaluated inside a managed object that is smaller than the economic object. An acquisition is systemic in different ways, with part of the reality governing the economic object deferred until the companies are combined.
AI is not a thing
Companies sometimes evaluate AI capabilities as though they’re buying an object, for example, a laser printer with a defined function and set of attributes. AI is instead an open-ended, configurable technology, the potential value of which depends heavily on the system around it.
What does it do?
How much does it cost?
How secure is it?
How hard is it to integrate?
Who will use it?
How difficult is it to use?
What do we expect it to accomplish?
What would have to be true for that value to appear?
Are those conditions in place?
What else would have to change?
What additional value is available that we may not have considered?
A company can perform excellent technical analysis of an AI capability and still destroy value because it evaluates the technology as a bounded purchase rather than as part of an operating configuration.
A common failure mode is applying AI to an inherited process without reconsidering why that process exists.
A workflow may contain steps created by an old human limitation. A report may exist because information once had to be gathered and summarized manually. A sequence of approvals may compensate for weak visibility. A role may spend much of its time translating between systems that no longer need a human translator.
Automating those steps can make the existing configuration cheaper or faster. It can also preserve a configuration whose original reason has disappeared.
The larger value may lie in redesigning the workflow, changing the decision architecture, collapsing steps, moving judgment, altering the product, or eliminating an activity altogether.
The enterprise can therefore become much better at doing something that no longer deserves to exist in its present form.
This is not peculiar to AI. A sufficiently important new capability can change more than the performance of one component. It can change which arrangement of components is economically superior.
How much better can this component become?
Given the new capability, should the surrounding system still be arranged this way?
Enterprises usually have stronger machinery for the first judgment. The second becomes decisive when technology, markets, capabilities, constraints, or knowledge move far enough that an inherited configuration loses its economic advantage.
Nothing has to break. The old arrangement can become wrong before it becomes broken.
Go deeper on enterprise possibility
How changing capabilities alter the economically superior enterprise configuration is developed further in Enterprise Possibility.
This is why AI is not a thing.
Its value does not sit inside the AI technology itself. It emerges from the configuration created around the capability and from the changes that capability makes possible elsewhere.
The enterprise has to cognize both.
When you can’t know
Even with the right economic object, some governing reality does not exist yet.
When one company acquires another, it is pursuing a post-acquisition state that satisfies its aims. The combined enterprise whose behavior will determine much of that value has never existed.
Diligence makes that uncertainty tractable by reducing the acquisition into analyzable forms. Revenue gets modeled. Contracts get reviewed. Customers get segmented. Systems get inventoried. Talent gets assessed. Facilities, liabilities, suppliers, intellectual property, operating processes, forecasts, and integration costs receive their own treatments.
That decomposition is indispensable. Nobody can apprehend a $2 billion acquisition whole.
But the acquisition is not the sum of its diligence files.
Its economics depend on the combined enterprise that emerges when customers, people, systems, products, incentives, contracts, operating processes, capital, authority, and competitive response begin interacting in a new configuration.
A company can therefore conduct excellent diligence and still destroy value.
This is rational incoherence returning at a much larger scale.
Diligence can reduce uncertainty about the two companies that exist. It cannot inspect the combined enterprise because the combined enterprise does not exist yet.
Customer response, employee adaptation, system interaction, political dynamics, new bottlenecks, and new possibilities only become observable after the configuration begins to exist.
More analysis cannot retrieve evidence from a future that has not occurred.
So practical coherence has two different jobs.
Reduce the uncertainty that is actually reducible. Test important assumptions. Search for relationships capable of changing the economics. Spend cognitive effort where being wrong will become expensive to unwind.
Treat realization as a source of new cognition rather than mere execution of old cognition.
Closing the deal changes the epistemic situation. The combined enterprise begins producing evidence.
Customers react. Systems interact. Employees adapt. Synergies appear, disappear, or change form. Constraints become visible. Assumptions collide with operating reality. Opportunities emerge that were invisible from outside the combined system.
The enterprise now has access to evidence no diligence process could have produced.
The value of that evidence depends on the company’s ability to interpret it without stripping away the relationships that produced its economic significance.
- →
If a failed assumption becomes merely an integration issue, the company may keep executing the thesis that generated the failure.
- →
If new evidence remains trapped inside the activity that produced it, the enterprise possesses the evidence without using it to revise the relevant cognition.
- →
If approval converts a hypothesis into doctrine, implementation becomes machinery for defending yesterday’s model rather than learning from today’s reality.
A company can know enough to change course and still fail to change course.
Practical coherence does not end at the acquisition. The enterprise has to preserve enough authority, flexibility, and continuity of reasoning for new evidence to alter the configuration being realized.
High systemicity forces action before the picture is complete. Coherence therefore depends on recognizing when the current frame deserves another look, widening it only where the economics require, and letting action reveal what analysis could not.
If the enterprise learns something worth carrying forward, the next problem begins.
How should that learning change the machinery that encounters the next situation?
A smarter company changes how it will think next time
Correcting one decision is useful. Having to rediscover the same correction every time is expensive.
Suppose a manufacturer spends weeks tracing an intermittent defect. No individual variable is out of spec. The failure appears only when a particular material condition, humidity range, and line speed occur together. Eventually the engineers find the relationship.
The enterprise has learned something. Will that learning becomes a one-time victory or standing capability?
The finding could remain in the people who discovered it, in a report, in meeting notes, or in an email. A future team may find it if they know where to look and recognize the situation well enough to go looking.
The company could also change company cognition and the operational state from which future decisions are made. The next shift inherits cognition from the previous one without having to reconstruct it.
Past cognition hopefully becomes future machinery.
Companies remember in their structure
People remember by carrying something forward in themselves. Companies also carry cognition forward by changing the environment in which future people and systems operate.
A policy carries an old judgment. A workflow carries a decision. A database field preserves a distinction. A software rule repeats an inference. A contract preserves a negotiated understanding. A budget carries expectations. A machine setting embodies a discovered operating condition. An organizational boundary preserves a judgment about which things belong together.
A document can remind someone what the company once learned. Structure makes that learning act again.
A software rule may apply an old judgment thousands of times after the people who made it have left. A default can shape behavior without anyone consciously recalling why it exists. A workflow can coordinate expertise that once had to be assembled manually.
Earlier cognition has changed the conditions under which later cognition and action occur.
A company might learn the wrong lesson from a right decision
Every time cognition becomes structure, some context gets compressed.
The original judgment arose under particular conditions and for particular reasons. A reusable rule cannot carry every detail of the situation that produced it.
A buyer reroutes shipments when a port closes. “Use the other port” preserves the decision but not necessarily the operative logic of the decision.
A company adds three approvals because weak visibility makes unilateral decisions dangerous. The approvals work. Years later, better systems remove the visibility problem while the approval chain survives.
A sales organization creates a customer category that captures an economically important distinction. The market moves. The category remains.
A recurring, stable relationship may deserve to become policy, software, routine, training, or physical design. A volatile conclusion may be better regenerated from current evidence. A one-time judgment may deserve a record of its reasoning without becoming a standing rule.
Good embodiment preserves enough of the relationship that made a judgment useful to keep the structure answerable to the conditions that justified it.
The company forgets why before it forgets what
Once a judgment becomes ordinary machinery, it stops arriving as an argument.
This is the approval threshold. This is the customer segment. This is how the system classifies the transaction. This is the field everyone fills in. This is the sequence the workflow follows.
The action persists while the reasoning that produced it fades.
That is part of how scale works. Every employee cannot reconstruct the history and logic behind every standard before using it. Much of competence comes from inheriting prior cognition without reopening it.
The enterprise can therefore preserve what to do while losing access to why it was once the right thing to do.
Past cognition acquires authority by disappearing into normal operation.
A control could outlive the constraint it was designed around. A threshold can keep firing after the underlying distribution changes. An organizational boundary might preserve a division that no longer matches the economics. A process can keep solving a problem the enterprise has become capable of eliminating.
The structure still looks like competence because it was competence.
Yesterday’s intelligence becomes tomorrow’s stupidity
The more deeply a judgment becomes embodied, the less often the company has to think about it again. More of the enterprise also begins depending on it.
Systems integrate around it. Roles adapt to it. Metrics assume it. Training teaches it. Contracts encode it. Other decisions take it as a fixed condition. Changing it becomes progressively more expensive.
Then reality moves.
The machinery keeps doing exactly what it was designed to do.
Successful cognition does more than guide current behavior. It becomes part of the cognitive apparatus itself. Today’s answer becomes one of tomorrow’s categories, rules, workflows, assumptions, interfaces, and defaults.
Embodied cognition also changes the evidence available to challenge it.
Old intelligence becomes self-reinforcing without anyone consciously defending it. The machinery has changed what reality is able to show the company.
Every improvement therefore changes the substrate on which later enterprise cognition occurs. That is how intelligence compounds. It is also how old intelligence gains the power to create new stupidity.
Good structure stays vulnerable to reality
A company that constantly reconsiders every rule loses much of the value of having rules.
You want durable structure that reality can still challenge.
A violated assumption, accumulating exception, changed condition, or surprising outcome needs some way to register as evidence against the machinery itself. If embodiment changes what the company observes, that evidence channel becomes part of the design.
Evidence then needs a path to effective authority. A company may contain abundant proof that a standing rule has become wrong while no person, process, or system has the power to reopen it.
The enterprise also needs enough ancestry to understand what it is reopening. A policy whose original purpose has vanished from institutional memory is hard to revise intelligently. So is a software rule whose assumptions are invisible, a metric whose causal logic has been forgotten, or an organizational boundary everyone experiences as natural rather than designed.
And revision has to remain economically possible. The more commitments accumulate around an embodied judgment, the more expensive correction becomes. Architecture, contracts, physical infrastructure, organizational structure, and other hard-to-reverse machinery deserve greater care because many later choices will inherit them as facts.
Make intelligence accumulate without making it permanent
The point of embodiment is to stop paying repeatedly for cognition the enterprise has already earned. Also, we should stop treating old cognition as permanently entitled to govern new reality.
Enterprise intelligence is recursive. The machinery through which the company thinks changes the machinery through which it will think later. Every improvement creates new machinery. Every new machinery carries assumptions, boundaries, and exclusions of its own.
A smarter company lets what it learns change what happens next, then lets reality change the machinery when the fit breaks.
A smarter company changes how it will think next time.
Purpose, representation, and distributed cognition
- George E. P. Box, “Science and Statistics” (1976)
- Felipe A. Csaszar and Daniel A. Levinthal, “Mental Representation and the Discovery of New Strategies” (2016)
- Herbert A. Simon, The Sciences of the Artificial (1969; 3rd ed., 1996)
- F. A. Hayek, “The Use of Knowledge in Society” (1945)
- Edwin Hutchins, Cognition in the Wild (1995)
- Richard L. Daft and Karl E. Weick, “Toward a Model of Organizations as Interpretation Systems” (1984)
- William Ocasio, “Towards an Attention-Based View of the Firm” (1997)
Systemicity, configuration, and dynamics
- Herbert A. Simon, “The Architecture of Complexity” (1962)
- Paul Milgrom and John Roberts, “Complementarities and Fit: Strategy, Structure, and Organizational Change in Manufacturing” (1995)
- Rebecca M. Henderson and Kim B. Clark, “Architectural Innovation” (1990)
- John D. Sterman, “Learning in and about Complex Systems” (1994)
- Jens Rasmussen, “Risk Management in a Dynamic Society: A Modelling Problem” (1997)
Learning, adaptation, and path dependence
- Peter F. Drucker, “The Theory of the Business” (1994)
- Chris Argyris, “Double Loop Learning in Organizations” (1977)
- James P. Walsh and Gerardo Rivera Ungson, “Organizational Memory” (1991)
- Donald T. Campbell, Assessing the Impact of Planned Social Change (1976)
Case sources
- Cisco: Scott Berinato, “What Went Wrong at Cisco in 2001”, CIO (2001)
- Cisco Systems, 2002 Annual Report, including the fiscal 2001 excess-inventory charge (2002)
- Sony: Henrich Greve, “Strategy and Organization in Sony”, INSEAD Knowledge (2012)
- Sony: Frank Rose, “The Civil War Inside Sony”, Wired (2003)