This is the main essay of a three-document set. Two companion applications publish under its umbrella, 2A on the post-2008 financial-sector instrument build and 2B on portfolio risk and optimization in consumer goods manufacturing. The public arc remains four essays. The companions deepen this one without renumbering what comes after it.
I. The architecture-governance gap
The autonomous systems arrived before the instruments meant to govern them. A firm of any size now runs optimization that searches a space of entangled decisions, returns a recommendation, and stands ready to act on it faster than any committee was built to match. Board oversight, the audit trail, the planning rhythm with major customers, the quarterly variance review, each was designed for a moment when the variables were few enough to enumerate, the interactions slow enough to argue one at a time, the question still long enough to be read before anything moved. The machines no longer hold still. They take in more of the firm at once than the governing bodies have language for, and someone who has spent years learning to read a firm’s decisions is asked to vouch for what those bodies cannot see.
The boards are diligent here, the audit functions careful, the senior people doing exactly what their roles describe. What they hold was built well for the world it was built to read. The world changed register underneath it, and a fine tool calibrated to the wrong scale reads the new world at the grain of the old, returning a clean answer to a question the firm has stopped asking. The gap is a failure of apparatus and a failure of vision at once, each feeding the other. Management cannot see what its tools do not show, and the tools are never asked to show what management does not yet know to look for. None of this indicts the people holding it. It is the shape the whole takes when an inherited way of seeing and an inherited way of measuring fall out of step, and the way out runs through both.
The thing that does the work here is dimensionality. A structure can be governed only at the resolution its tools can represent. A body that can hold four variables and their pairwise interactions can govern a decision living in four. Put that same body in front of a decision living in four hundred variables, entangled through the whole, where the interactions run combinatorially, and it does not govern the decision badly. It governs a reduction, a low-dimensional shadow, and mistakes that shadow for the whole structure because the shadow is the only part its tools can render. The recommendation was produced in the large space. The committee reads it in the small one. Between the two sits the firm’s exposure, chosen by no one, because no one can see the part of the decision that did not survive the trip across.
A hedge-fund desk in 2007 lived inside that trip without naming it. Its risk tool returned a clean number each morning, and the room read the number with confidence, because the number was clean. The number was clean. The book it summarized, rendered in pairwise form, was anything but. The desk governed the number it could see, and the book ran on below it, holding the whole of what no one could read.
Someone now stands where that recommendation returns. The boards, the audit functions, the optimization machinery are what she holds and finds wanting. She holds the integrated picture of a firm against bodies that read it one axis at a time, asked to certify a result the committee cannot reach in its own language. The gap reaches her first as a silence, the moment no one in the room can say what would have to be true for the recommendation to be wrong. The number is on the screen. The room is waiting. No question in the room’s vocabulary will open the part of the decision that matters. Something almost comic sits underneath the silence, a board convened precisely to govern the firm and unable to phrase the one decision in front of it. That silence is not ignorance. It is the absence of the tool that would let people put the questions their judgment is already straining toward. What did the recommendation assume about the parts of the world it could not measure. Which couplings did it treat as independent when they were not. Who answers if it moves the firm somewhere no one chose to go.
The reflex is to assume the gap closes on its own, that the apparatus will catch up because the machines are new. The history of the one sector that lived through this says otherwise. Tools catch up only when someone inside builds them, and only after the institution has the language for what they must read, which is exactly what the gap withholds. A board cannot ask for a tool it has no words to specify. An audit function cannot test against a standard no one has articulated. The machinery was built by people who understood the mathematics and not the governing. The governing bodies were staffed by people who understood the firm and not the mathematics. Whoever stands between them, reading both, must often say two things at once, that the recommendation is sound by every internal measure and that the firm cannot yet govern it.
One sector has already paid for this in the open. Before 2008 the largest financial firms governed an entangled book with a tool that could not read its structure. David X. Li published a Gaussian copula in 2000, and the desks applied it to collateralized debt obligations, pricing the correlation of defaults off a single number that did not move.1 A fixed correlation cannot represent the state a book enters when its parts stop moving independently and begin to move together. The tool read a world of loosely coupled holdings, and it read that world well until the variables tightened into one. When the housing names defaulted in concert it went blind precisely where the firm was most exposed, in the tail where everything moves at once. No one had stopped watching. The structure the tool could hold was simply thinner than the book had quietly accumulated. That same shape now waits on a screen in firms that are not banks, which got no interval to build ahead of the failure and may yet get one.
In a conference room early in 2026, two colleagues, one from a major management consulting firm and one from a major technology company, described the firm’s generative capabilities as “glass box,” fully explainable, and rehearsed how to walk the client through the explanation as a service the client was owed. The phrase sat on the slide between them. They were operating in good faith. What neither could see was that the thing they were preparing to explain sits beyond what the technology can do at the level of technical feasibility, that they were assembling a surface-level mirage of an answer in the language of fundamental understanding. The gap formed in the language of the meeting, and the one who had spent years reading firms named it, and described the tools that do read at the resolution the problem requires. They were open to it, eager to learn. This was the knowledge gap itself, the one that now sits between what artificial intelligence and advanced analytics are asked to govern and what the people deploying them understand those instruments can actually do. The high-dimensional, sometimes unexplainably complex world requires new tools and new techniques the room had not yet been shown.
II. The inheritance
Financial services is the only fully worked example of governing a high-dimensional complex system with a tool that could not read it. The inheritance the previous generation left us comes substantially from there, and its value is also its limit. The apparatus was calibrated for a problem of lower dimensionality and slower tempo than the one now in hand. The new world has rendered the calibration insufficient and the competence still sound.
A failure of representation runs underneath the whole pre-crisis record. The copula was one instance. The loss measure was another. Value-at-Risk named a loss threshold at a stated confidence and left the shape of the loss beyond it unrepresented, saying nothing about the size of the loss in the tail it had drawn a line in front of. Under ordinary conditions that silence did not matter much. Under correlated stress, when the entangled losses lived precisely past that line, the silence was the whole of the danger. The measure held the body of the distribution and left blank the part that ended firms.
The lesson is about calibration rather than negligence. The instruments failed because they were calibrated for one dimensionality and asked to govern a higher one. In both cases the gap ran between the resolution the tool could read and the structure the world actually had, and in both it was invisible from inside, because no instrument can show what it was not built to hold. Everything built afterward is response. The failure is the anchor.
The response, when it came, was institutional, and it can be read in what particular people argued. Daniel Tarullo joined the Federal Reserve Board in 2009 and built the stress-testing regime, having held before the crisis that the inherited framework treated the largest trading books as no special case, when the losses those books would take in a tail event were exactly the correlated losses the framework did not price. He set out to make the supervisor read the firm under a stress it had not yet suffered rather than against a ratio drawn from the calm. The macroprudential turn he pressed, toward the firms whose failure would move the whole rather than toward each read in isolation, was an attempt to raise the resolution the supervisor could read, to govern the entanglement rather than the parts.2 The same instinct ran through the diagnosis Andrew Haldane offered at Jackson Hole in 2012. He argued that complex environments are often governed better by simple rules than by models adding representation faster than reliable signal, and he showed it across domains, from the dog that catches the frisbee on a single heuristic to the regulator drowning in granular risk weights.3 A model adding detail faster than signal governs worse, not better. There is something bracing in a regulator standing at Jackson Hole to argue, against a room that had spent a decade adding instruments, that the dog catching the frisbee on a single rule had the better method. The lesson the two of them drew was that the instrument had grown ornate without growing wiser, and that the cure was to read the entanglement that mattered rather than to model every variable that could be named.
The institutional tools that came out of that decade are the inheritance now in hand. The Comprehensive Capital Analysis and Review made the supervisor read the firm’s capital against forward stress rather than a static ratio the crisis had discredited.4 In the narrow lane of market-risk capital, the trading-book rules finalized in 2019 moved the internal-models measure from Value-at-Risk to one that integrates the loss across the tail the earlier measure left blank.5 Each raised the grain at which the financial supervisor could read, built by people who had watched the old tools go blind and set out to build ones that would not. A kind of stubborn faith sits underneath that work, a generation that had watched its instruments fail and chose to spend the next decade building better ones rather than concluding the book could not be read. The work is honored and inherited. What is also plain, standing now where an autonomous system recommends a move across a structure that is not a loan book, is that the inheritance was built for one structure, governed by a specific institutional apparatus, on a tempo set by the supervisory calendar. The recommendation waiting on the screen reads more axes than risk, faster than any calendar, in a firm with no chief risk officer holding the whole of it at the board. The inheritance is load-bearing. It does not reach. The full account of how the lesson was paid for, with names and instruments across 2009 to 2019, is the work of 2A.
The operator inside a consumer goods firm, an industrial manufacturer, an energy company, a defense supplier, holds the same kind of structure under the same condition. The categories are entangled, the channels too, the supply chain coupling decisions the teams treat as independent, and the tools built to govern that work, the annual plan, the category review, the portfolio committee, the variance review, were calibrated for a market a committee could read before it moved. They are low-dimensional instruments asked to govern a high-dimensional book at a tempo set by the analytics rather than the calendar. The rest of the corporate world holds positions its tools cannot read and has not yet paid. That same failure, now read on the ground in consumer goods, where one representation governs trade investment and category allocation under a binding finance constraint, is the work of 2B. The structure has outrun the instrument, and the difference between the financial sector then and the rest of the corporate world now is only that one has had its reckoning and the other has not.
The previous generation had a word for what the book held. It called it variety, in the sense Ross Ashby and Stafford Beer gave the term, the count of distinct states a system could enter. The word did real work in its time, and it named a real thing. It is not the word the present generation reads with. The instruments that now read a high-dimensional book do not measure variety. They read the structure of the book itself, at the depth the structure has. The two readings are not opposed. The newer one supersedes the older the way a finer instrument supersedes a coarser one, by reading more of what was always there.
III. The narrowed unification
Underneath all of this is a claim that comes slowly, through the work. It concerns two readings of a portfolio the firm has always treated as separate disciplines. One asks what the firm is exposed to. The other asks what the firm should become. The firm houses them in different functions, staffs them with different people, runs them on different clocks. Do both for long enough and the org chart’s separation comes to look like an accident of staffing, because the two stand on a single structural representation.
To read a book for risk, the operator must first hold it as one complex structure, positions and interactions together at enough resolution that the joint behavior can be examined, then interrogate that picture for exposure, asking how the structure behaves under stress, where its losses concentrate, what happens in the tail where the parts move together. To read the same book for its best configuration requires the identical first move, and the optimization read then searches the same picture for the best feasible configuration under the firm’s constraints. Neither can begin until the picture exists.
The picture is the expensive part. Build it, and either interrogation runs off it. That is the unification, and it is narrow on purpose. The two reads share one rendering. That is the whole of the shared ground, and it has to be guarded at once against what it is not, because the elision that follows is where firms lose money.
A shared representation is not a shared discipline. The distance between those two sentences is the distance between a firm that governs its autonomous systems and one that has merely connected them. The same tool reads a book for its risk and for its best configuration, and the institution does not let the two sit in one place. The risk read goes up to a chief risk officer with the board behind her. The optimization read goes across to the investment committee, where the portfolio manager owns the answer. One rendering, two readers, two clocks, two consequences when each is wrong. The apparatus never divided the discipline. The institution did, and the division has to be honored even on the days the tool makes it trivial to collapse.
Then the reads diverge, and the shared picture reconciles none of it. They diverge in what each is for. Risk minimizes a loss, asking how large the exposure could grow. Optimization maximizes a return-adjusted utility, asking what the firm should become. One protects what exists. The other changes it. They point in opposite directions, and a rendering that serves both does not tell you which way to face. They diverge in the clock each runs on. Risk runs through the cycle, asking how the structure survives stress that may arrive anywhere across a long uncertain period. Optimization runs toward a configuration the firm answers for over its planning horizon. A move optimized on the one clock can be a risk the firm cannot carry on the other, and the shared account reports both without warning that they are timed against each other.
They diverge in who owns the answer. Risk answers, in the sector that built the tools, to a chief risk officer holding the whole book against the board. Optimization answers to an investment committee, a category general manager, a brand president, an operating committee, whose mandate is to grow the thing rather than bound its loss. The picture does not assign the accountability. It only produces the reads. The assignment is a governance act, and it is exactly the act the autonomous systems are outrunning. And they diverge in what it costs to be wrong. In a leveraged book a single coupled exposure read as independent can end the firm, so the tolerance for that error is asymmetric, weighted hard against the loss that was missed. An optimization landing on the second-best configuration costs only the difference between best and second-best, and the tolerance there is closer to symmetric. A firm that governs both as though the penalties matched will under-protect against the error that ends it and over-invest against the error it could survive.
The cost of the elision keeps one shape. Two vehicles run on one road and still owe different duties, hold different speeds, answer to different authorities, carry different consequences when they crash. A firm that points at the quality of its road has said nothing about its traffic. So with the picture. Having paid for it, the firm assumes that because the reads share a rendering they share a governance. They do not, and the shared rendering is no more the governance than the road is the law that runs on it.
The order in which the divergences arrive matters to how the firm gets them wrong. The objective it sees first and underestimates. The horizon hides inside the objective, a move right on its own clock and wrong on the clock the firm actually lives by. Accountability it defers, because no one wants to assign who owns the autonomous answer before they have to. Tolerance stays invisible until one of its two errors arrives, sometimes only after a major problem. Each is a place where the shared picture says nothing, and that fourfold silence is what has to be filled before the recommendation is allowed to move the firm.
IV. What the governance instrument must satisfy
The rendering that holds a complex structure at the resolution it actually has does not come from nowhere, and whoever would govern an autonomous system is owed a way to specify what the tool that produces it must do, in terms a board can hold and an audit function can test. The methods that satisfy these criteria are several, and they change as the field changes. The criteria do not, because they are properties of governance rather than of any one technique.
The first is the structure read at the dimension it actually has. A board that can hold four variables cannot govern a recommendation living in four hundred entangled ones unless the apparatus first finds, inside those four hundred, what is load-bearing and what is noise. It must compress what cannot be reduced without losing what matters, because the easy compressions are exactly the ones that throw away the entanglement in the tail. An instrument that flattens the book to the handful of variables a committee can argue is reproducing the pre-crisis failure with newer tools, and a firm that accepts the flattening has learned nothing from the sector that already paid for it.
Then there is translation, the carrying of a board’s question down into the large representation and the answer back up. The operator has to be able to ask in the language she reasons in and receive the answer in the same language. A recommendation produced in a four-hundred-dimensional space and delivered as a single number has not been made governable. It has been made opaque with a confident face. The inherited tools never had to make that trip, because question and computation once lived in the same small space. An instrument that cannot make it has left the board governing the shadow again.
Explanation has to be native to the room that must accept it, a reading the room can reach while it decides rather than a rationalization assembled once the recommendation is already made. Cynthia Rudin argued, against the field’s drift toward explaining opaque models after the fact, that high-stakes decisions should be made by models interpretable by construction, and that the post-hoc explanation of a black box is not the same thing as a model a person can follow, because the explanation is itself a second model that can be wrong about the first.6 The distinction has teeth. A board that accepts an explanation of a decision it cannot itself reach has accepted a story about the decision rather than the decision. The public research program that set out to build systems whose reasoning a person could follow took the second path and learned how hard the first warning was.7 A number arrives in the room, a price or a hedge ratio or a category allocation, with no reasoning the room can follow, and the room accepts it on the confidence of the machine that produced it, which is a deferral and not a decision. Reachability is the precondition of accountability rather than a courtesy to the non-technical.
Every recommendation out of a high-dimensional optimization rests on assumptions about the parts of the world it could not measure, the couplings it estimated, the variables it treated as fixed, and most never reach the room, because the inherited tools have no slot for them. An apparatus that surfaces its own premises lets the board ask whether it believes them, the audit function check them against the record, the investor weigh them against the capital at stake. A recommendation whose assumptions cannot be surfaced is one the firm is accepting on faith, and faith is not governance.
Blindness has to be made visible too. The pre-crisis instruments went blind in the tail because they could not hold the entanglement that lived there, and no one could see the blindness from inside. A tool that declares the edge of its own representation, which couplings it held and which it treated as independent, which axes of the firm it took in and which it left out, lets the firm tell a decision it governed from a decision it reduced to a shadow. Without that declaration the firm learns the difference only when everything arrives at once, which is how the financial sector learned it.
Last, the record. A governance act that leaves no trace is a decision that happened to be supervised once. What the recommendation assumed, how much of the structure it read, what it could not see, and who was answerable for letting it move the firm is what the next operator inherits, the way this generation inherited the financial sector’s.
None of these stands alone. A board that cannot read the structure cannot ask the tool to translate it, and a recommendation no one in the room can reach is one whose premises cannot be held, whose blind edge cannot be seen, and whose record is a trace of a decision that was never governed. Most firms running autonomous systems today satisfy none of them in the form the structure requires, which is the plain measure of how far the governance has yet to come. To name the criteria is to begin building the tools.
V. What governance must be able to see
The criteria let the operator build the rendering. They say nothing about what the governing bodies owe once the recommendation arrives, and that is the harder half. The duties that follow had to be learned the way the financial sector learned its failure modes, by naming them before anything could be built against them.
The premises come first. An investor, a board, an audit function, and a senior operating team must be able to see the couplings the recommendation estimated, the variables it held fixed, the distributions it presumed. The criterion in the prior section asks the tool to surface these. This asks the room to do something with them once surfaced, to refuse a recommendation whose premises it cannot endorse rather than wave it through because the rendering was expensive. Surfacing without a standing duty to reject is documentation, not control.
The boundary comes next. The room must see how much of the structure the recommendation read and what it could not, the same edge the financial precedent learned to declare. A firm blind to that boundary cannot tell a decision the tool governed from one it reduced to a shadow, and learns the difference only when the entanglement arrives at once. The machine that declares which couplings it held and which it treated as independent, which axes it took in and which it left out, hands the firm the one thing the pre-crisis instruments never gave their supervisors, a view of their own edge.
Accountability has to be named before the machine moves the firm, while the deciding is still ahead. An autonomous system acting on its own recommendation can make a move no governing body deliberated, because the body was reading the shadow while the machine acted in the full space. Assign the answer after, and the question of who owned it becomes the question of who is to blame. There is a particular vacancy this addresses, the moment a system has moved the firm and the room turns to find no one was designated to own what it did. Someone speaks first. A name attached to a consequence after the fact is not accountability. It is a search for someone to absorb it.
Refusal is the duty held most closely, because the capability conversation most wants to skip it. A firm that can read its book at full depth, recommend a configuration, and act on the recommendation has acquired a great deal of power and said nothing yet about its limits. The most consequential line written into an autonomous architecture is the one naming the configurations it will refuse even when they score well. That line is hardest to write because it costs something the moment it is honored, a return left on the table, a move the model favored and the firm forbade. Whoever builds the tool is positioned to write it in before the build, while the constraint is still cheap and abstract rather than expensive and specific. A firm that has fixed only what its machine can do has left that line unwritten.
Endorse the premises and you have read the structure, because the premises and the represented structure are the same thing seen from two sides. Once that boundary is visible, accountability can be named, because you can bound what the responsible party answered for. Name accountability and refusal becomes possible, because a limit means nothing without someone answerable for honoring it. A firm with only the first has begun. A firm with all four has governed. Almost no firm running autonomous systems today holds more than one, which is the distance still to travel.
VI. Closing
The governance that would close this gap has not yet arrived. Read this far and you know it for a description of where the work stands, neither a forecast nor a complaint. The autonomous systems are in the building. The instruments that would let a board, an audit function, an investor, and a senior operating team govern them are not, at least not in the form the structure now demands, and the gap between the two is held, day after day, by the people inside the institutions who can see both sides of it at once.
They are the ones holding it. Not the board, which cannot yet read the whole. Not the audit function, which cannot yet check what the rendering assumed. Not the optimization machine, which can search a space and return a configuration but cannot tell the firm what it should refuse. The person who holds the integrated picture of the firm against the tools that cannot read it is the one standing in the gap. No release engineers the holding away. It is the condition of the work in this decade, the standing in the gap while the instruments to close it are still being built, and built well, by the same people standing in it.
What the financial sector learned, it learned in the open and after the fact, when the entanglement it could not see arrived and the firms paid for the structure their tools had not held. The rest of the corporate world has not yet had its reckoning. The operator inside those firms is working in the interval the financial sector did not get, before the concert arrives, when the instruments can still be built ahead of the failure rather than after it. The tools will be built by practitioners or after a crisis that did not have to happen, and whoever can see the gap now is the one who can decide which.
The governance has not yet arrived. The work is already here. Holding the picture. Naming what the instruments must be able to see. Building the language before the procedure. Refusing to certify what cannot yet be read. That holding is the work, and the work is what the next instruments will be made of. There is no shortcut through it, and no one else positioned to do it. The financial sector’s reckoning is the one record we have of the gap paid for in full, and it is read in detail in 2A. The same failure forming now on the ground in consumer goods, before its reckoning, is read in 2B. The interval is still open in most of the corporate world, and the two companions are where the work of reading it begins.
Notes:
1. Felix Salmon, “Recipe for Disaster: The Formula That Killed Wall Street,” Wired, February 23, 2009 (retrospective on David X. Li’s 2000 Gaussian copula and its role in the crisis). https://www.wired.com/2009/02/wp-quant/
2. Daniel K. Tarullo, “Next Steps in the Evolution of Stress Testing,” Federal Reserve, September 26, 2016 (macroprudential supervision and the post-crisis turn toward systemically significant firms). https://www.federalreserve.gov/newsevents/speech/tarullo20160926a.htm
3. Andrew G. Haldane and Vasileios Madouros, “The Dog and the Frisbee,” Jackson Hole economic policy symposium, August 2012, Bank for International Settlements. https://www.bis.org/review/r120905a.pdf
4. Federal Reserve, “Comprehensive Capital Analysis and Review (CCAR),” supervision and regulation overview. https://www.federalreserve.gov/supervisionreg/ccar.htm
5. Basel Committee on Banking Supervision, “Minimum capital requirements for market risk” (Fundamental Review of the Trading Book), d457, finalized January 2019. https://www.bis.org/bcbs/publ/d457.htm
6. Cynthia Rudin, “Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead,” Nature Machine Intelligence, 2019. https://www.nature.com/articles/s42256-019-0048-x
7. DARPA, “Explainable Artificial Intelligence (XAI)” program. https://www.darpa.mil/program/explainable-artificial-intelligence
Companion applications
- 2A,The Post-2008 Instrument Build. Derives the claim in financial services, treating in full the technical failure of the pre-crisis risk models and the institutional instruments built in response across 2009 to 2019.
- 2B,Portfolio Risk and Optimization in Consumer Goods. Derives the claim in consumer goods manufacturing, where the same high-dimensional complex system governs trade investment and category allocation under a binding finance constraint.
Essay 2A — The Pattern Returning, Reprised
The Post-2008 Instrument Build
Financial portfolio risk and portfolio optimization.The technical failure of the pre-crisis instruments, the rebuild carried by the practitioners inside the institutions, the unification the practitioner reads first.
Companion application to Essay 2, The Governance ThatHas Not Yet Arrived. One of two derivations of the narrowed unification claim, this one in financial portfolio risk and optimization. The parallel consumer-goods companion is Essay 2B. The arc remains four essays publicly. 2A and 2B publish as companion applications under Essay 2’s umbrella.
I. The instrument that could not read the structure the system had.
A firm can be lost to an instrument the people on its floor distrusted the whole time. 2008 is the record of how that happens. The senior trader on a mortgage desk in 2006 already knew something his tools did not report. He knew it the way you know the weather is about to turn before the gauge on the wall confirms it. The names in the mortgage book were not independent. They were tied to one thing, the price of American housing, and if that moved the whole holding would move with it. What he could not do was make the apparatus say so. It read each position against a number fixed in calmer weather, and that number told him the portfolio was diversified when his own eye told him it was one bet wearing many faces.
The tool was honest about its limits before the desks stopped reading the fine print. David X. Li published the Gaussian copula function in 2000, treating the joint behavior of defaulting credits as something a single correlation parameter could capture.1 Pricing a collateralized debt obligation, the desk used it to ask how likely it was that many underlying credits would default together. It answered with one number for how tightly two were coupled, calibrated to a recent past in which they had not defaulted together. So it reported a future in which they were unlikely to. Whoever trusted that was trusting a function built for mild, stable correlation, asked to govern a world that had quietly become something else. The code was developed by people that neither understood the maths, the business, or the governance.
What the trader could not get the function to show was the state a book enters when its assets stop moving independently and begin to move as one. A fixed correlation claims the relationship between two credits holds in every weather. He had lived through more than one cycle. In a panic the relationships tighten until everything that can be sold is sold at once, and the diversification that looked real on the spreadsheet evaporates in an afternoon. The copula carried none of that tightening. When the housing names defaulted in concert he watched a number describe a world the firm was no longer living in. He had been handed something that saw less of the structure than the position he was being asked to hold.
The same blindness lived in what the risk report led with each morning. Value-at-Risk gave him one sentence for a meeting. It named a loss threshold at a confidence level and said that on all but the worst days the firm would not lose more than that. Read carefully, it said nothing about the shape of the loss past the threshold. It drew a line and left everything beyond it blank, as if how bad the worst day could be were not the question that ended firms. He was being asked to govern the tail with something built to leave it unwritten.
A firm can be governed only in the variety its instruments let it see, to borrow the term the previous generation reached for. This is older than the crisis that proved it. Anything that holds fewer distinct ways the book can move than it actually has will be confident exactly where it is blind. The pre-crisis tools held fewer. They saw a collection of positions with stable, separable relationships, and the firm actually held a high-dimensional complex system whose relationships changed under stress. That gap was the gap the crisis came through.
It did not feel like ignorance at the time. It felt like a disagreement between two views, the trader’s and the machine’s, in which the institution was organized to believe the machine. The board reviewed its output. The capital was held against its measure. The committee deciding what the firm could carry deliberated over the numbers it produced. The risk officer who might have stopped the concentration, had he carried that view to the table, sat below the desk that wanted the trade and did not have the chair from which a trade can be refused. The man who tried to say those numbers described a world the firm was leaving had only a hunch and a memory of prior cycles, and the firm did not run on hunches. The screen was winning. That asymmetry is what the next decade had to fix, and fixing it was never a matter of a better number. It was a matter of building something that saw the structure he already knew was there.
What he carried out of 2008 was not that the mathematics had betrayed him. Li’s function did what it was specified to do, and the people who built Value-at-Risk had been candid that it described the body of the distribution and not the tail. The lesson was harder than betrayal. Anything honest about its limits becomes dangerous the moment an institution forgets them and governs as if it saw everything. The decade that followed was the institution’s slow, expensive admission that his private judgment had been the true one. The build was the attempt to give that judgment something it could finally use in the room.
II. The rebuild, in the practitioner’s hands.
The regulatory decade is usually told as rules arriving from above. Read it the other way and the people who carried it come into view. The capital rules, the stress tests, the new committees, the new measure of tail loss, none built themselves. Each was the institutional form of one person’s judgment, hardened into procedure so it would survive the next time the person who held it was overruled. The decade from 2009 to 2019 is better understood as operators finally given instruments that read at the grain they had always known the book had.
Daniel Tarullo held the clearest account of what had to change, and he held it before the crisis made it policy. In Banking on Basel, published in 2008, he argued that the inherited framework treated the largest trading books as no special case, when the losses those books would generate in a tail event were precisely the correlated losses it did not price.2 He joined the Federal Reserve Board in early 2009, into the supervisory portfolio, in the months when the firms that had carried the copula into the crisis were still being kept alive by public capital. From that chair he moved from diagnosis to build. He replaced a static view of capital with a forward-looking one, and he held the firms whose failure would move the whole to a standard the firms whose failure would move only themselves did not have to meet.3 It is the institutional version of what the floor trader could not get his tool to do. The man who had written that the largest books were the special case was now in the room where the special case could be written into law.
Stress testing is the most concrete version of the same work. The Comprehensive Capital Analysis and Review asked a firm to project its own book through a severe, correlated downturn and show the supervisor what its capital would be on the far side.4 The operator running it could no longer report a single threshold and leave the rest blank. He had to walk the book through the tail, name the scenario in which the correlations tightened and the losses arrived together, and hold capital against the walk. The state was now written into the test the firm had to pass, the very thing the previous generation had lacked.
To the firm’s highest table this work was carried by the chief risk officer, and the crisis is what priced that role into a seat with board access rather than a function buried below the desks. Before, the person who held the integrated picture often held it without the standing to make the institution act. The chief risk officer who now sits with the board, who can stop a trade or refuse a concentration, is the institutional answer to the asymmetry of 2006. The same work ran across the asset classes the risk report had separated. A treasury team that saw funding, trading exposures, and the lending book as one structure, rather than three reports from three functions, was reconstructing the integrated picture the old instruments had dismantled, building out of organization rather than mathematics.
Rebuilt most exposed of all was the measure that had left the worst case blank. The Fundamental Review of the Trading Book, finalized in 2019, moved the internal-models measure away from Value-at-Risk toward Expected Shortfall, which integrates the loss across the part the earlier one had left unwritten.5 The operator who had carried Value-at-Risk into meetings for years, knowing exactly what it did not say about the worst days, now had a measure that asked about the shape of the loss beyond the threshold. This was narrow, scoped to the market-risk capital measure and the firms using internal models. It was the closing of one specific blindness in the one place it had been most dangerous.
Andrew Haldane read the rebuild most skeptically, and the skepticism was part of the build rather than a dissent from it. At Jackson Hole in 2012, in “The Dog and the Frisbee,” he argued that an intricate environment is often governed better by a few simple rules than by a model that adds detail faster than it adds reliable signal.6 An instrument can hold more variety on paper, in the sense Haldane’s argument turned on, and see it worse in practice, drowning the signal in detail the operator cannot check. The cure for seeing too little was to read the structure that actually governed the book’s behavior, a harder target than mere elaboration and a different one from seeing more.
Where the risk had gone mattered as much as where it had been. Jeremy Stein was working that end of the problem. As a Federal Reserve governor in 2013 he argued that credit markets could overheat in ways the standard measures would miss, that risk reaches for yield and hides in places the instruments were not built to look.7 The copula had been blind to entanglement. Stein named a blindness to migration, to the way risk moves toward the yield the instrument has declared safe. He was extending the rebuild toward the next failure. The post-crisis instruments were better than before, and the people who built them knew they were not the last word.
Hold one thing above the rest. The institution did not lead and the practitioner did not follow. The practitioner led. The judgment came first, held by a person on a floor or in a supervisory chair who could see the whole structure before any rule required him to. The institution followed, slowly and at cost. Dignity belongs to that work, the supervisors who learned a firm’s book at the price of their weekends, the people who had been right once and overruled and came in the next morning to build the thing that would make the next refusal stick. The architecture of 2009 to 2019 is the sediment left by a generation that had been right and overruled and was determined to build instruments that could not be overruled the same way again.
Early in 2026, in the Gulf region of the Middle East, the question on the table ran the other way from the one 2008 left behind. How to build resiliency into the risk function itself rather than bolt it on after a loss. How to decode the data they already held into strategic risk read across the whole of their holdings, across the asset classes the old reports kept apart, with no idiosyncratic risk left undetected inside any one of them. The senior leadership of a very large asset management firm had arrived at the structural question rather than the surface one. Their holdings were written across a single sheet on the table. From there the discussion turned to a methodology of decision science that joins mathematical analysis to semantic, symbolic reasoning of the human-like kind. No one in the room was defending an inherited instrument. They were eager to learn how the high-resolution techniques applied to the book they held, asking at the start what the trader of 2006 could only have wished the room around him would ask. That is the path the financial sector now offers, the question reached for before a loss forces it.
III. The narrowed unification, derived from the financial side.
There is a discovery waiting for the practitioner who carries the post-crisis tools into the room where the firm decides what to become, and he is rarely told to expect it. The apparatus built to read the portfolio’s risk and the one that would read it for its best configuration are looking at the same object, and he already owns half of what the second needs. The risk rebuild forced him to hold the book as one high-dimensional complex system at the resolution it actually has. That picture, the system held whole with its variables entangled, is exactly what portfolio optimization stands on. He sees it before the textbooks tell him, because he is the one holding both on the same ground.
The shared ground is the picture itself. Harry Markowitz opened the tradition in 1952, asking how an investor should configure holdings given that the assets move together, that their covariances matter as much as their individual returns, and that the best configuration accounts for the coupling correctly.8 A capital-allocation model, a stress-testing apparatus, a risk committee weighing exposure across entangled positions, each requires the same single object whose parts move together. The stress test that reads that object under duress and the optimizer that reads it for configuration are built on one representation. This is the unification, and it is real. Two takes on one system held at the grain required to govern it.
What he also discovers, the moment he tries to run the risk tool as an optimizer, is that the shared picture does not carry a shared discipline. The two diverge at the points where he is accountable, which is why he feels the divergence as work rather than receives it as theory. Risk is built to minimize a loss, to ask what the firm is exposed to and hold capital against the answer. The optimizer is built to maximize a return-adjusted utility. Point the loss-minimizing tool at the maximizing question and the book comes back safe and going nowhere. Point the other at the loss question and the configuration comes back excellent on average and fatal in the tail.
He feels the next divergence in the clock each one keeps. Risk runs through the cycle, asking what the portfolio does in the severe downturn, the worst weather, because the worst weather is what ends firms. The work Tarullo built into the stress test is built to ask about the bad year that has not arrived. Optimization runs toward a steady state, asking what configuration serves the return objective under ordinary conditions sustained over time. The tradition Markowitz opened is built to ask about the good decade that usually does. Confuse the two and you have built a firm optimized for a weather it will not usually have, or stress-tested for a return it will never earn.
Then there is the question of who answers for each, the divergence the institution learned to seat. Risk answers to the chief risk officer and the risk committee, the seat the crisis built with board access precisely so the loss side could not be overruled by the desk that wanted the trade. Optimization answers to the investment committee and the portfolio manager, accountable for the return the configuration earns. The institution learned through the crisis to separate the seats so the loss side would have its own chair. That separation is governance the shared picture does not supply.
And there is how each tolerates being wrong, the divergence the crisis punished most directly. Risk tolerates error asymmetrically. A view that understates the tail can end the firm. One that overstates it merely costs some return. So it is built to fear the first far more than the second. This is the asymmetry Haldane was defending when he argued for fewer, sturdier rules over models that see too much and trust it too far. Optimization tolerates error closer to symmetrically, treating a configuration somewhat too aggressive and one somewhat too cautious as both simply suboptimal. Import that symmetric tolerance into the risk side and you have rebuilt the exact blindness the crisis punished.
What he holds at the end, having felt the divergences in his own work rather than received them in a list, is the precise shape of the unification. The tools built after the crisis carry most of the structural picture optimization would need. They do not carry the governance it requires, because that governance runs the other way on every count, on what is asked, on the clock kept, on the seat that answers, and on how error is tolerated. The unification is structural and it is narrow. He sees it first because he is the one standing on the shared ground holding two tools that diverge in his hands.
IV. What the practitioner who carries the instrument now reads.
Hand a practitioner the tools the crisis built and he reads his own portfolio in a way the previous generation could not. That difference is the whole inheritance. Carrying them into the room in 2026, he holds a high-dimensional complex system whole, through the cycle, with the tail shown rather than left blank, and with a chair at the board table from which the firm can be stopped before it does something he can see and the desk cannot. The portfolio manager holds the risk view and the configuration view at once and feels, in the friction between them, the divergences the trader of 2006 could not have named because he did not yet have two to set against each other.
He sees the entanglement itself, the state in which the assets stop moving independently and begin to move together, the state the copula could not hold. He can see it before it arrives rather than discovering it in the afternoon it does. The stress test Tarullo built walks the book through the correlated downturn. The measure the Fundamental Review installed integrates the loss across the tail. Between them they let him do in the open what the trader of 2006 could only do as a private hunch he had no standing to act on. That is the practical meaning of the rebuild. It did not make him smarter. It gave his existing judgment a tool the room would accept, which is a smaller thing and the thing that mattered.
The divergence between his two takes now reads as a thing to govern rather than a confusion to resolve. The previous generation conflated risk and return because it had nothing that forced the distinction. He has both, knows they share a picture and diverge in governance, and reads the firm’s decisions with both clocks running and both accountabilities seated. A firm that conflates them courts the next crisis. He sees the elision before it costs anything, the one advantage the previous generation never had.
The limit of the inheritance is now visible. The post-crisis tools carry most of the structural picture the optimization side needs and none of its governance. The institution gave him the shared ground. It did not give him the second tool, the one that reads the same object for what the firm should become and answers for that to the seat that owns the return. He holds half of a thing, the risk half, built at great cost. The optimization half waits to be built on the same picture by people who understand that the picture transfers and the governance does not. He is further along than 2006, and he can see the unbuilt half from where he stands.
The last thing he reads is the one that travels furthest. The financial sector paid for this inheritance once, in the open, and what it bought is not the financial sector’s alone. He sees a complex system at the resolution it carries because a crisis forced his industry to build the tool. Any firm allocating constrained capital, shelf, channel, or attention across entangled claims under uncertainty is holding a portfolio in the same structural sense, whether or not it has been forced to build the thing that reads it. He has the view because his sector paid for it. The view itself was never sector-specific.
V. What the rebuild left behind.
A rebuild is remembered for what it closed, and this one closed the blindness that ended firms. The tail is shown now. The correlated downturn is written into the test a bank has to pass, the chief risk officer sits where the trade can be stopped, and the integrated picture the old instruments dismantled has been put back together in organization and in capital. None of that is small. The one who lived inside the failure does not undersell the decade that answered it. He also knows, better than the people who legislated it, exactly where the answer stopped.
It stopped at risk. The tools the institution paid for show what the firm is exposed to, they show it well, and they say almost nothing about what the firm should become. The optimization half of the unification, the one that takes the same high-dimensional complex system and asks for its best configuration under the firm’s own objective, was never built to the standard the risk half reached. There was no crisis to force it. A firm does not fail from a book merely configured short of its potential the way it fails from a tail it could not see. So the institution built what answered to fear and left what would have answered to ambition in the practitioner’s hands, unfinished, on the picture the risk rebuild had already paid for.
Two other things were left there with it. The translation across sectors, carrying the financial view into an industry that allocates shelf and trade and attention rather than capital, was never done, because each sector waits for its own expensive decade and the financial one does not transfer itself. And the decision discipline on the practitioner’s side, the habit of holding the two apart while running both, of refusing to let the shared picture collapse into a shared judgment, was never written down as anything an institution could teach. It lives where it has always lived, in the person who has felt them diverge in his own work.
So what the next sector inherits is not a finished tool. It is half of one, and the harder half is the one still missing. Whoever carries the financial view into consumer goods, into the operational sectors, into the governance of the systems now deciding faster than the institutions around them, carries also the knowledge of what was not done, and that knowledge is its own inheritance. He is handed a picture that transfers, a governance that does not, and a clear view of the unbuilt half from the ground the last decade cleared. The consumer-goods practitioner takes up that work in the next companion. Essay 3 carries it past any single sector. The financial essay closes here, on what it could not finish.
VI. Closing.
There is one thing the next generation should not have to relearn at the cost the last one paid, and the operator who lived through the failure and built the rebuild is the one who can hand it forward. A tool confident where it is blind is more dangerous than no tool at all, because the institution will believe the confident one and overrule the person who can see past it. He held the true view in 2006 and lost to the convenient one. He spent a decade building tools that read the structure his judgment had always known was there, so the judgment would no longer lose inside the room where the firm decides what it can carry.
The discipline he can state plainly. Build the tool that reads the portfolio at the depth the portfolio actually has. Govern the risk side and the optimization side as the separate disciplines they are, even though they stand on one picture. Never mistake the shared ground for a shared discipline, because the most expensive error in the record is the one that treated the fatal mistake as merely one among many. The person who holds that discipline holds the whole lesson of 2008 in a form he can carry into the next portfolio and the next room.
That same lesson is already being read on different ground. A practitioner in consumer goods holds a portfolio too, a high-dimensional complex system of brands and channels and trade investment and shelf, and reads it with tools that report it as a set of separable lines when it has long since become one bet wearing many faces. He has not had the crisis that forces the build. He has had its slower version, the impairment and the divestiture and the brand cull, the firm shrinking the book because it cannot govern the entanglement it cannot read. The view paid for in finance in a single expensive decade is the one he is being asked to find without the same forcing event, on the same structural ground, in the next companion. The system does not care which sector holds it. Only the practitioner does, and only he reads it first.
Notes:
1. Felix Salmon, “Recipe for Disaster: The Formula That Killed Wall Street,” Wired, February 2009. https://www.wired.com/2009/02/wp-quant/
2. Daniel K. Tarullo, Banking on Basel: The Future of International Financial Regulation. Peterson Institute for International Economics, 2008. https://www.piie.com/bookstore/banking-basel-future-international-financial-regulation
3. Daniel K. Tarullo, “Next Steps in the Evolution of Stress Testing,” remarks of September 26, 2016. Board of Governors of the Federal Reserve System. https://www.federalreserve.gov/newsevents/speech/tarullo20160926a.htm
4. “Comprehensive Capital Analysis and Review (CCAR).” Board of Governors of the Federal Reserve System. https://www.federalreserve.gov/supervisionreg/ccar.htm
5. “Minimum capital requirements for market risk” (d457). Basel Committee on Banking Supervision, January 2019. https://www.bis.org/bcbs/publ/d457.htm
6. Andrew G. Haldane and Vasileios Madouros, “The Dog and the Frisbee,” Jackson Hole economic policy symposium, August 2012. Bank for International Settlements. https://www.bis.org/review/r120905a.pdf
7. Jeremy C. Stein, “Overheating in Credit Markets: Origins, Measurement, and Policy Responses,” remarks of February 7, 2013. Board of Governors of the Federal Reserve System. https://www.federalreserve.gov/newsevents/speech/stein20130207a.htm
8. Harry Markowitz, “Portfolio Selection,” The Journal of Finance 7, no. 1 (March 1952), 77 to 91. https://onlinelibrary.wiley.com/doi/10.1111/j.1540-6261.1952.tb01525.x
Companion to
- 2.0, The Governance That Has Not Yet Arrived. The main essay of the triptych, where the architecture-governance gap, the practitioner subject, and the narrowed unification are established at structural register. 2A derives the claim in financial services; 2B derives it in consumer goods.
Essay 2B — The Pattern Returning, Completed
Portfolio Risk and Optimization in Consumer Goods
On the consumer goods firm as a high-dimensional complex system under binding finance and channel constraint, the structural representation it shares with a bank’s loan book, and the governance the financial-sector instruments cannot directly transfer because the objective, the horizon, the accountability, and the tolerance for error all run the other way.
I. The operating problem at breadth
A category general manager inside a large consumer goods firm sits down each quarter to allocate something that feels, on the surface, like a budget. A fixed sum of trade investment to place with retailers, a fixed quantity of shelf to argue for, a fixed amount of media and innovation funding, and a set of categories that all want more of each. The work looks like division. It is not. Every dollar placed against one category changes what the next dollar does somewhere else, and the manager who treats the categories as separate envelopes is making the one mistake the structure is built to punish.
The categories’ variables are entangled, and the coupling is not a metaphor. A deep promotion on one brand pulls volume forward and steals it from the brand beside it on the same shelf, so the lift booked in one category is partly borrowed from a category another manager answers for. A price cut on a flagship trains the shopper to wait for the next one, eroding the equity the brand was built to carry. A retailer who funds a feature on one line expects a feature funded on another. The plant runs under shared lines, shared logistics, shared procurement, so pushing one category through in a quarter quietly removes the capacity another was counting on. The teams treat these decisions as independent because their accountability is drawn that way. The decisions are not. They are one high-dimensional complex system the organization has been cut into pieces to run.
What defeats the inherited tools is the scale of that interweaving. A major manufacturer runs dozens of categories with coupled variables, tens of thousands of stock-keeping units, hundreds of trade events in a quarter, and dozens of retail relationships each with its own bargaining power. McKinsey reports that between twenty-eight and fifty percent of retail sales volume across consumer goods categories in Europe is sold on promotion, that some manufacturers invest up to twenty percent of gross revenue in promotions, that visibility into the net effect of that spending is poor relative to its size, and that nearly half of large European manufacturers report no trade-promotion optimization tool at all.1 The largest discretionary line on the profit and loss moves the most money across the most decisions in the shortest time, and what governs it reads almost none of the structure it moves through.
Dimensionality is what is missing. The previous generation would have counted the distinct states a thing can be in, the range a picture has to tell apart if it is going to govern the thing at all. A brand portfolio at this scale holds more distinct joint configurations than any committee can enumerate, because the categories interact, the interactions interact, and the count of interaction terms grows combinatorially rather than line by line. The manager works hard enough. The manager is handed a problem whose breadth exceeds what the tools can represent, and a state they have no symbol for is a state the firm cannot price or govern.
The one who lives inside this is a small set of people, their accountabilities drawn across the coupling rather than along it. The category general manager owns the growth of a set of categories and allocates the trade, shelf, and innovation that competes for the firm’s finance. The brand president owns the equity of a brand that lifts or drags the categories around it, held against the quarterly volume the same brand is asked to deliver. Constraints belong to the supply chain VP, the plants and lines whose limits bind decisions the category teams make as if they were free. The binding finance constraint belongs to the CFO, since trade investment, capital, and working capital are drawn from one pool that clears across the whole firm. Each reads a true picture of part of the lineup. None reads the whole, and the gap between the one structure and the partitioned accountability is where the firm is actually run.
What makes the gap structural rather than personal is that each is doing the job correctly by the measures the organization built to evaluate it. The category manager rewarded for assigned growth pursues it even when it is borrowed from a neighbor. The brand president is right to protect equity even when it costs the quarter. The supply chain VP is right to run the plants efficiently even when efficiency removes the flexibility another category needed. The CFO is right to hold the finance constraint even when holding it forces a choice no single owner has the standing to make. The firm has optimized each part against its own objective and has nothing that reads the parts as one. The result is a lineup that performs adequately in ordinary conditions and reveals, when conditions change, that no one was holding the whole.
A consumer goods firm, then, is a high-dimensional complex system under a binding finance constraint and a binding channel constraint, run at a breadth its inherited tools cannot represent, by people whose accountabilities are drawn across the coupling rather than along it. The manager allocating trade each quarter is doing portfolio work in the strict sense, even though the word is rarely the one used in the room.
II. The inherited governance instruments, calibrated for a smaller problem
The instruments the operator inherits to govern this book were built with care, for a market that moved at a pace a committee could read. An operator who sits through their cycles for a few years learns the calibration in the cadence. Annual joint business planning sets the terms with each major retailer once a year and presumes those terms will hold for the year. The category review brings a category’s leadership together quarterly and presumes a category can be read on its own, without holding the cannibalization those lines are working on their neighbors and the supply they are consuming. The brand portfolio committee deliberates which brands to keep on a rhythm measured in quarters, when the equity it decides about erodes or compounds continuously. The finance variance review reads the gap between plan and outcome after the quarter has closed, when the decisions that produced the variance are already months behind the people explaining them. Each does real work. Each was built to read a market that holds still.
The market does not hold still.
That market has a name the readers of this work already carry. It is a BANI environment, brittle, anxious, non-linear, and incomprehensible.2 The book is brittle in that an arrangement that worked last year fails this year without the warning the annual cycle was built to provide. The anxious part is the loss of signal, the firm’s inability to tell which change deserves action and which is only noise. Non-linearity shows up when a moderate move in one category produces a disproportionate effect three steps away, through cannibalization, through retailer reaction, through a supply constraint no one was watching. The whole becomes incomprehensible when the committee cannot hold the joint behavior of the lineup in the language it meets in. The operator is asked to vouch for a plan the committee can no longer read at the rate the market changes it.
The strain is written into the public record of the largest firms in the sector, read here as public record rather than as verdicts on the firms. On February 21, 2019, when Kraft Heinz reported its results for 2018, it recorded a non-cash impairment of more than fifteen billion dollars against the carrying value of its Kraft and Oscar Mayer brands.3 The equity in those brands had deteriorated against private-label pressure and channel shift for the better part of two years, while the category-level reports the governance bodies read carried that equity as sound. The number held that value right up until the morning it reported, all at once, that the value had not been there.
Six years later the same firm yields the structural version of the lesson. On September 2, 2025, Kraft Heinz announced a plan to separate into two independent public companies, one built around its global sauces and shelf-stable meals and one around its North American grocery staples.4 The complexity of the current structure, the firm said, made it challenging to allocate capital effectively, to prioritize initiatives, and to drive scale in its most promising areas.4 That is a company’s own account of a system whose breadth had grown past what its allocation could hold, and of a remedy that made the portfolio smaller instead of building a richer apparatus to govern it whole.
The sector’s other large manufacturers supply one pattern at several stages. They are stages of the same story. On March 19, 2024, Unilever announced that it would separate its ice cream business into a standalone company, since ice cream ran on a distinct operating model and supply chain, and the unit that resulted is The Magnum Ice Cream Company.5 Nestlé moved to narrow toward four core areas, coffee, pet care, nutrition, and food and snacks, while divesting its ice cream and water operations and running a program to bring its brand count down from more than four hundred toward roughly one hundred fifty.6 Procter and Gamble had taken an early version of the same move a decade before, announcing in 2014 a plan to shed up to a hundred brands and concentrate on the seventy to eighty that carried the overwhelming share of its sales and profit.7[ These are not firms that failed. They are the largest manufacturers in the world arriving, by different routes and in different years, at the same recognition, that the lineup had accumulated more variety than the inherited tools could hold, and that the move those tools allowed was to reduce it.
The channel makes the constraint tighter. Private-label share, the retailer’s own brand competing directly against the manufacturer’s, has reached a level that changes the firm’s whole position in the negotiation. Circana puts private-label value share in the United States at roughly twenty-two percent and unit share higher still, and in the European Union at thirty-nine percent of value and close to half of all units sold.8 The retailer is now also a competitor on the same shelf, with the data, the placement, and increasingly the consumer trust to take volume directly. Retail has consolidated alongside this, pooling the bargaining power into fewer and larger hands. On December 10, 2024, a federal court granted the Federal Trade Commission’s request to block the proposed merger of Kroger and Albertsons, the largest grocery merger the country had attempted, on the finding that the combination would have crossed monopoly thresholds in more than two thousand local markets.9 The counterparty grows more powerful and more vertically capable on a clock faster than the annual cycle was built to track, while the firm’s negotiating tool still resets once a year.
Read all the way through, the public record gives a portrait of a high-dimensional complex system governed with inherited instruments. It is not a catalog of failures. The instruments produce correct numbers from their inputs. They are calibrated for a lower-dimensional world, and they go quiet exactly where the coupled variables live, in the interaction between categories, in the erosion of equity, in the binding of supply to decisions made as if it were free, in the channel that has become a competitor. The firms that have read this most clearly reach for the remedy their tools allow, which is to make the stable smaller rather than build a picture that could carry it whole. What the record does not yet show is a firm that built the apparatus that could have held the whole.
III. The narrowed unification, derived from the consumer goods side
Stand inside the firm and read the book twice, once for risk and once for its best configuration, and what becomes visible first is how much the two share. The risk reading asks what the firm is exposed to, taking in the category-level loss tolerance, the erosion of brand equity, and the uneven and poorly understood return the largest discretionary line on the profit and loss produces across the events it funds. The optimization read asks what the firm should become, taking in the allocation of trade across the coupled categories, the shaping of the lineup toward the configuration that compounds instead of cannibalizing, and the allocation of capital under the finance constraint. These look like two different jobs in two different parts of the organization. Underneath, they require the same prior achievement.
That achievement is a representation of the high-dimensional complex system at the resolution it actually has. You cannot read the risk of a book you cannot picture at the breadth it has, because the loss arrives from the coupling the picture left out, and you cannot optimize one either, for the best configuration depends on the same coupling. A loan book under a capital constraint and a category set under a trade, channel, and finance constraint are the same kind of object, an interwoven whole whose joint behavior has to be held before any governing can touch it. Harry Markowitz gave the form in 1952, showing that a portfolio’s risk depends on the covariance between its parts rather than on the holdings read one at a time, so the right object of attention is the portfolio whose variables are bound up together rather than the individual position.10 Applied to product lineups rather than securities, and extended through half a century of work on constrained allocation in the tradition running from Dantzig’s linear programming forward, the same mathematics describes a product lineup whose members covary through cannibalization, equity spillover, and shared supply.11 A promotion that moves volume on one line and steals it from the next is a covariance term in exactly the sense Markowitz meant. The shared structure is real and old. It is the ground both readings take their stand on.
Hold this where a category manager actually meets it. There is a slide. It presents four categories that grew on their own lines last quarter, and the room reads four businesses, each with its own owner and its own target. What the slide cannot show is that the four are one structure under a single trade-investment pool, a single retailer who allocates the shelf, and a finance constraint that clears across all four at once. They are knit together through the whole, so a win booked on the first line was paid for, in part, on the third, and the manager who lived the quarter knows which line paid. The operator reads the one system the slide was never able to draw.
That is the whole of what is shared, and anyone who has held both books knows the sharing stops at the picture. The consumer goods take diverges from the financial on four counts, and on each it runs opposite.
Start with what each is for. The financial risk read minimizes a loss tail, succeeding by making the worst plausible outcome survivable. The consumer goods one is built around the return on category growth and brand equity compounded over years, succeeding when the categories grow without borrowing from each other and the equity strengthens rather than erodes under the volume it carries. The objective is a maximizing aim pointed at what the firm should become, not a minimizing aim pointed at what could end it, and a tool tuned to hold down a loss tail is tuned away from that. The clock diverges the same way. The financial read is through-cycle, built to hold across the regime change in which correlations rise toward one. The consumer goods one runs quarterly at one end and multi-year at the other, the manager allocating trade for the coming months while the brand president holds an equity question that compounds across years. It is the horizon of an ordinary working year, not a stress event.
The sharpest break is in who answers for the whole. After 2008 the financial sector built a chief risk officer with direct access to the board, a single office holding the integrated picture and answerable for it at the highest level. Consumer goods has no equivalent. Accountability is distributed across the category managers who own growth, the brand presidents who own equity, the supply chain VPs who own the constraints, and the CFO who owns the finance, and none of them holds the entire book at the board the way a chief risk officer holds the bank’s. The integrated picture falls between the offices that exist. The tolerance for error inverts the same way. A leveraged financial book can be destroyed by a single coupled exposure read as independent, so its discipline must be intolerant of being wrong once in the place that counts. A consumer goods firm absorbs category-level missteps that would be catastrophic in leveraged finance and survives them. A promotion that loses money, a brand that underperforms for a year, a category that gives up share, these are recoverable costs, and a firm that is right on balance across many such decisions does the job. A tool calibrated to the financial tolerance would govern this book with a caution it does not need and at a cost it should not pay.
So the inheritance offers the shared structure, the discipline of holding the system at the resolution it carries before any governing touches it, which is the hard half the inherited consumer goods tools never built. What it does not offer is the governance. An operator who imported the financial instruments whole would have mistaken the shared floor for the building the sector still has to construct on it. The representation transfers. The governance does not.
IV. What the practitioner now reads
Read now what each of the four operators faces when the whole is put in front of them at the breadth it has.
The category general manager faces the allocation of trade across categories whose variables are entangled. What looked like dividing a budget is the placing of the largest discretionary line on the profit and loss across categories that answer each move in ways no plan can specify. Read them as independent envelopes and each is optimized against its own target while the whole degrades. Read them as one structure and the right allocation comes into view, the one that compounds across the categories instead of maximizing any single quarter, and the manager who can hold it is holding the book the governance was cut into pieces to avoid holding. It is worth pausing on how much skill that takes, and how little of it the org chart was built to see.
Harder still is the version the brand president carries, holding brand health and category growth in a single picture at once. Equity is what lets a brand command price, hold shelf, and lift the categories around it, and it erodes under exactly the promotional pressure that delivers the near-term volume the same brand is asked to produce. A president who protects equity at the cost of the quarter and one who delivers the quarter at the cost of equity are both doing the job. The conflict is structural, the fact that equity and volume are interwoven while the governance reads them apart. The Kraft Heinz write-down is what the record shows when an equity reported as sound had been eroding until the audited figure caught up all at once.
The supply chain VP faces the constraints that bind the decisions the category teams treat as independent. The plants, lines, logistics, and procurement are shared, so a decision to run one category through the network in a quarter removes the capacity another was counting on, and the constraint is the physical form of the coupling the teams keep trying to manage away. The VP reads the book from the one vantage that cannot pretend the categories are independent, because the steel and the trucks make the coupling concrete. In most firms that vantage sits downstream, consulted after the category plans are set rather than shaping them.
Above all of them sits the binding finance constraint, and the CFO holds it. Trade investment, capital, and working capital are drawn from one pool that clears across every category, so the finance constraint is where the coupling the others feel locally becomes a single firm-wide problem. The CFO is the closest thing the firm has to a person reading the whole book, and the limit of that seat is the limit this companion has been tracing. Holding the finance constraint is not the same as holding the integrated picture of growth, equity, and supply, and the CFO’s tools read the book after the quarter rather than at the rate the coupling moves. The CFO owns the constraint that makes the book one structure and inherits nothing built to govern it as one.
It is worth being plain about the standing from which this is written, and equally plain about its limits. Three decades inside this kind of work is the source of the confidence that the entanglement is real, that the operators are doing their jobs correctly against measures that cannot see the whole, and that the gap between the one structure and the partitioned accountability is where the firm is actually run rather than a defect to be cleaned up on a Tuesday. That experience is the witness register of this companion, and it is not the source of any named texture. There are no scenes from inside a client here, no specific allocation that could only have come from a particular firm, no executive playing a role, because the moment the prose reaches for that specificity it stops being an honest account of a structural condition and becomes a disclosure the work does not permit. The reading is meant to be recognized from the inside by anyone who has done it. The recognition is the demonstration. The naming would be the credential, and the credential is the thing this register refuses.
What the practitioner reads, then, is one structure held at a breadth the inherited governance cannot represent, by four offices each accountable for a true part and none for the whole, under a finance constraint that makes the parts one and a channel that grows more adversarial faster than the planning cycle tracks. The financial sector built the picture that lets such a book be read at breadth and paid for it in the open. Consumer goods has the same problem in front of it and has not yet built the governance its own objective, horizon, accountability, and tolerance for error require. The operator inside the firm in 2026 is the one who can already see this, and that gap is the precise place the firm will be governed or lost.
V. What consumer goods knows that finance does not
The inheritance has run one way so far, the financial sector having built the picture first and paid for it in the open. Read it the other way. The consumer goods operator carries operational knowledge the financial-sector one never had to acquire, and it is not a lesser version of risk discipline. It is a different discipline, learned on a different machinery.
Begin with time. Brand equity compounds or erodes across years, and stewarding it is a muscle the quarterly desk was never built to develop. The operator who has held a brand through a decade has watched a price decision made for one quarter cost three years of pricing power, and has learned to read the slow variable underneath the fast one. The equity does not announce its erosion. It declines quietly, below what any quarterly report can show, until the decline is large enough to surface as a number, and by then it has been true for years. The leveraged book is disciplined toward the tail and the through-cycle stress, not toward the patient accrual of an intangible no quarter can show. That patience is knowledge the consumer goods operator carries as a matter of course.
Then there is the matter of who holds the whole, which finance solved by building one office and consumer goods never did. The easy reading calls that a deficit. Watch how the operator lives it and something else appears. Accountability distributed across the category manager, the brand president, the supply chain VP, and the CFO is itself a discipline, because no single office can collapse the whole into one number and govern the number instead of the thing. The single risk office is a strength and also a temptation, a seat from which the whole can be flattened on a bad day. The consumer goods operator governs a structure no one office can flatten.
Finance built an apparatus to fuse the two readings. Consumer goods kept the person who holds the seam. No one designed it that way. It simply happened. The work simply demanded that a human hold the integrated picture of trade, equity, and constraint together before any tool existed to do the fusing, and so a human did, and still does. Something rare lives in that, a person carrying in his head a structure the firm has no chart for, holding it steady year after year while the institution around him still lacks the words to name what he is doing. That holding is not written down anywhere. It lives in the judgment of whoever is doing it, accrued over the years it took to learn the book. The person is portable where the apparatus is not.
What sat on the table early in 2026 was a portfolio run at a breadth no vendor’s account of it could hold, and the working problem the people responsible for it carried. How to read one structure at the resolution it carries, then optimize and decide across it on a global scale, and build a platform that activates locally at every level of the enterprise, from the mature markets rich in data down to the smaller ones that have little. The setting was a bright morning at Palmer Square, senior people from a large multinational consumer goods firm at the white marble of the table with mathematicians and AI specialists. They had gone direct to the source of the science, to the people who do the mathematics rather than to a vendor selling its result. They had read enough of the technology to ask the question at the texture the problem requires, which is the question the impedance layer between the operator and the mathematics is built to keep from being asked. That morning the layer was not there. The work was the two sides reading the book together.
Tolerance for error completes the inversion. A consumer goods firm is built to be wrong sometimes and right on balance, and the operator who has run that way for years has a calibrated relationship to recoverable failure the leveraged desk, intolerant of being wrong once at the tail, never had to develop. A category that gives up share for a year is a cost to be read and recovered. It is not a detonation. That tolerance is not laxity. It is the knowledge a governing tool needs when the thing it governs is allowed to be wrong and asked only to be right on balance.
None of this is a deficit measured against finance. The consumer goods machinery is governed by different muscles, of multi-year stewardship, of accountability held without a single office to collapse it, of living calibrated to recoverable error. The one who carries this forward, into the governance of structures that will run faster and reach further than any loan book, brings something the financial inheritance did not have to learn and does not contain.
VI. Closing
The firms that have read the problem most clearly have so far reached for the one remedy their inherited tools allow, which is to make the lineup smaller. Kraft Heinz moved to split in two. Unilever set its ice cream business loose to run on its own model. Nestlé narrowed toward four categories and began cutting its brand count by hundreds. Procter and Gamble shed the long tail a decade ago and has kept taking complexity out since. Each is a defensible answer and in several cases the right one. Each is also the same answer, the answer of an organization that cannot govern the variety it accumulated and reduces it instead of building the apparatus that could hold it.
Reducing the portfolio is not the only answer available, and it is not the answer the next generation of work is about. The shared picture the financial sector built, the discipline of holding a high-dimensional complex system at the depth it carries, is the half of the inheritance the consumer goods sector can take. What it cannot take is the governance. The objective compounds growth and equity over years instead of controlling a loss tail. The horizon is the operating one, where the financial book is read through the cycle. The accountability is distributed across offices that do not concentrate it in a single seat with board access. The tolerance for error runs toward right-on-balance, never toward never-wrong-at-the-tail. The representation transfers and the governance has to be built for the consumer goods problem.
The practitioner inside the firm is the one who has to build it, and is the only one positioned to, because he is already holding the integrated picture of growth, equity, supply, and finance against tools that read only its surface and cannot yet govern it. What he has to construct is the tool that lets the category general manager, the brand president, the supply chain VP, and the CFO read one system at the resolution it has, on the clocks they run, answerable for an objective the financial tools were never asked to serve. No office above the operator can specify it, because no office above the operator is standing where the entanglement and the accountability come apart. That is the seat the consumer goods firm has not yet filled, and the practitioner who can already see the whole is the one the next instrument will be built by, or it will not be built at all.
Stephen DeAngelis
Princeton, NJ
June 2026
About the Author
Stephen F. DeAngelis is the founder, president, and CEO of Enterra Solutions and Massive Dynamics, two companies that apply artificial intelligence and advanced mathematics to complex enterprise challenges. His work spans international relations, national security, and commercial technology, with visiting research affiliations at Princeton University, Department of Chemistry, the Computing and Computational Sciences and National Security Directorates of the Oak Ridge National Laboratory, the Software Engineering Institute at Carnegie Mellon University, and the MIT Computer Science and Artificial Intelligence Laboratory. He holds patents in autonomous decision science.
Notes:
1. McKinsey & Company, “How precision revenue growth management transforms CPG promotions,” 2021. McKinsey reports that between twenty-eight and fifty percent of retail sales volume across CPG categories in Europe is sold on promotion, that some CPG companies invest up to twenty percent of gross revenue in promotions, that visibility into the net short- and long-term impact of promotions is poor relative to the size of the investment, and that nearly half of large European CPG manufacturers report having no trade-promotion optimization tool. https://www.mckinsey.com/capabilities/growth-marketing-and-sales/our-insights/how-precision-revenue-growth-management-transforms-cpg-promotions
2. The BANI framework, an acronym for Brittle, Anxious, Non-linear, and Incomprehensible, was introduced by the futurist Jamais Cascio in 2020 as a successor frame to VUCA. Institute for the Future, “Navigating the Age of Chaos.” https://www.iftf.org/insights/navigating-the-age-of-chaos-a-sense-making-guide-to-a-bani-world-that-doesnt-make-sense/
3. Kraft Heinz Company, “Kraft Heinz Reports Fourth Quarter and Full Year 2018 Results,” February 21, 2019. The company recorded a non-cash impairment of more than fifteen billion dollars against the carrying value of its Kraft and Oscar Mayer brands. https://www.sec.gov/Archives/edgar/data/1637459/000163745919000010/ex991-erq42018.htm
4. Kraft Heinz Company, “The Kraft Heinz Company Announces Plan to Separate into Two Scaled, Focused Companies to Accelerate Profitable Growth and Unlock Shareholder Value,” September 2, 2025. The company stated that the complexity of its current structure made it challenging to allocate capital effectively, prioritize initiatives, and drive scale in its most promising areas. https://news.kraftheinzcompany.com/press-releases-details/2025/The-Kraft-Heinz-Company-Announces-Plan-to-Separate-into-Two-Scaled-Focused-Companies-to-Accelerate-Profitable-Growth-and-Unlock-Shareholder-Value/default.aspx
5. Unilever, “Unilever to accelerate Growth Action Plan through separation of Ice Cream and launch of productivity programme,” March 19, 2024, and “The Magnum Ice Cream Company demerger.” Unilever separated its ice cream business, which operated on a distinct supply chain and operating model, into a standalone listed company, The Magnum Ice Cream Company, reporting ice cream thereafter as a discontinued operation. https://www.unilever.com/news/press-and-media/press-releases/2024/unilever-to-accelerate-growth-action-plan-through-separation-of-ice-cream-and-launch-of-productivity-programme/
6.Nestlé public reporting, 2025 and 2026, on the narrowing of its focus toward four core areas (coffee, petcare, nutrition, and food and snacks), the divestiture of its ice cream and water operations, and the “Fuel for Growth” program intended to reduce its brand count from more than four hundred toward roughly one hundred fifty. Reuters and trade-press reporting, 2026. https://finance.yahoo.com/news/nestl-refocuses-portfolio-governance-earnings-111208387.html
7. Procter & Gamble announced in 2014 a plan to divest up to one hundred brands and concentrate on the seventy to eighty core brands that accounted for the large majority of its sales and profit, and has since run successive productivity programs. Reuters, “P&G to sell up to 100 brands to revive sales, cut costs,” August 4, 2014; P&G 2024 Annual Report on its productivity program. https://www.reuters.com/article/business/pg-to-sell-up-to-100-brands-to-revive-sales-cut-costs-idUSKBN0G13WW/
8. Circana, “Circana Global Research Illuminates New Era of Private Label Transformation,” 2025. Circana reports U.S. private-label value share at twenty-two percent and unit share at twenty-four percent, and European Union private-label value share at thirty-nine percent and unit share at forty-seven percent. https://www.circana.com/post/circana-global-research-illuminates-new-era-of-private-label-transformation
9. U.S. Federal Trade Commission, “Statement on FTC Victory Securing Halt to Kroger-Albertsons Grocery Merger,” December 10, 2024. The U.S. District Court for the District of Oregon granted the FTC’s request for a preliminary injunction blocking the proposed merger, the largest grocery merger in U.S. history. https://www.ftc.gov/news-events/news/press-releases/2024/12/statement-ftc-victory-securing-halt-kroger-albertsons-grocery-merger
10. Markowitz, Harry. “Portfolio Selection.” The Journal of Finance 7, no. 1 (1952), 77-91. Markowitz showed that the risk of a portfolio depends on the covariance among its holdings rather than on the holdings read independently, establishing the portfolio whose variables are entangled as the proper object of allocation. https://www.jstor.org/stable/2975974
11. The operations research literature extending constrained portfolio allocation to product and category portfolios, in which categories covary through cannibalization, brand-equity spillover, and shared supply constraints, and trade-promotion optimization research in the marketing-science literature. See, in particular, the multi-product allocation and resource-constrained optimization work in the tradition of George B. Dantzig, Linear Programming and Extensions (Princeton, Princeton University Press, 1963), and the trade-promotion-optimization findings summarized in the McKinsey work at endnote 1 and the constrained-allocation tradition originating with Markowitz at endnote 10
Share this post
