The captain's instinct
The storm is coming and everybody knows what to do: the captain takes the helm. Inside a company the scene repeats with every crisis —emergency committees, approvals that climb two levels, cash controlled from the top, "at a time like this we can't afford to have everyone doing their own thing"—. The instinct has an apparent logic: a crisis demands coordination, and coordination demands a single command.
A team of economists —Philippe Aghion, Nicholas Bloom, Brian Lucking, Raffaella Sadun and John Van Reenen; readers of these series already know three of them from the measurement of competent management— set out to test it against the most measurable crisis of the century: the Great Recession of 2008–2009. They had what they needed: data on ~1,300 manufacturing firms across ten countries (World Management Survey) with their degree of decentralization measured before the crisis —how much real authority the plant manager held over investment, hiring, product and pricing—, plus the chance to replicate it on ~8,800 US plants (the MOPS census). Then they measured who fared better when the shock landed.
The result runs head-on into the captain's instinct: in the hardest-hit industries, decentralized firms lost significantly less in sales than centralized ones —roughly 3.6 percentage points less of a decline— and the finding replicates almost identically (3.5 pp) across US plants. The advantage did not exist before the crisis: it appears with the shock, between 2008 and 2011. Decentralized firms also held their productivity better. And the detail that turns a correlation into a story: the advantage concentrates where there was more turbulence —measured as in Part 1: product turnover, volatility—; that turbulence explains a substantial share of the effect. The authors ruled out the obvious alternative explanations —it isn't that the decentralized firms were better managed overall (they control for management quality), nor that they carried less financial risk—.
In the aggregate, the effect moves countries: organizational structure accounts for something on the order of 15% of the post-crisis growth differential between economies. The obvious question is why. The answer is eighty years old.
Hayek's argument (the plant-manager edition)
In 1945, Friedrich Hayek wrote the essay that explains this result: the knowledge that matters for deciding well is dispersed —each actor knows "the particular circumstances of time and place" that no center can assemble in time—. The central planner does not fail because he is dim: he fails because local information arrives late, summarized and distorted.
Carried into the firm, the argument comes with a switch that the crisis flips on. In stable times, local information is worth little: what happens at the Rosario plant this month looks like last month, and head office decides well enough on reports and averages —the benefits of control (coordination, scale, capital discipline) outweigh its cost—. But turbulence inverts the equation: when demand shifts week to week, when the customer you always had disappears and an odd one shows up, when an input goes missing and the product has to be redesigned around whatever is available, fresh information from the ground becomes the decisive asset —and it is exactly the asset that does not survive the round trip up and down the hierarchy—. Every approval that climbs two levels is information that ages two weeks. The plant manager with real authority decides on what he sees today; head office decides on what the plant saw last quarter, condensed into three slides.
Seen through the lens of Part 2: decentralization is distributed optionality —every local manager with real authority is an option on a fast response that the company has already paid for—. And through the lens of the previous series: it is the structural answer to complex terrain, where the pattern emerges on the ground and central analysis arrives after the facts.
"This is a multinational problem" (it isn't). The evidence comes from large manufacturers, but there are good reasons to think the mechanism bites harder the smaller the company, because in a small or mid-sized business centralization has a name and a face: the owner. Every price waiting on his WhatsApp, every purchase waiting on his signature, is the same toll of aged information —with an aggravating factor: the central node is a single person, and in the storm that person is also busy renegotiating with the bank—. The SME version of this article is not an organizational redesign: it is three moves —small but real decision thresholds for two or three trusted people, with a dashboard in plain view; a handful of simple written rules (what can be promised to a customer, when a purchase goes ahead without asking, what triggers a call to the owner); and the discipline of not clawing back every delegation at the first mistake—. The test is question 7 at the end: if two weeks out of contact would sink the company, the org chart is decorative.
Honesty about the limits is in order, because the paper does not say what the tempting headline would say. It does not say "never centralize": it says that for a demand shock with high turbulence, local information beat central control. Other kinds of crisis call for something else: a liquidity crisis calls for centralized cash; a safety or compliance crisis, for direct command; the chaotic terrain of the previous series, for temporary command to stop the bleeding. The error in the captain's instinct is not centralizing something —it is centralizing everything, including the commercial and product decisions that live on local knowledge, at precisely the moment that knowledge is worth the most.
Delegating without letting go of the wheel
The attentive reader's objection arrives on its own: "if I hand every manager authority over investment and pricing, this thing falls apart on me". The objection is sound —and the answer to it is the bridge to everything these series have been building—. Decentralization that works is not autonomy plain and simple: it is autonomy inside an architecture, and the architecture has three pieces.
First piece: the management floor. Delegating without clear targets, without performance monitoring and without consequences isn't decentralizing: it is letting go of the wheel. The boring practices from the competent-management article —targets that bite, follow-up, correction of deviations— are the prerequisite that makes delegation safe: head office can release the decision precisely because it never released visibility. It is no accident that the same researchers were measuring both things: structured management and decentralization are complements, not substitutes.
Second piece: simple rules instead of a thick manual. How do you coordinate an organization in which every node decides? The classic answer —the two-hundred-page manual— dies in the first week of turbulence, because no manual anticipates tomorrow's case. Kathleen Eisenhardt and Donald Sull documented how fast companies coordinate in fast markets: with simple rules —a handful, typically between two and seven, learned from their own experience— that mark the edges of the field and leave the play free. Boundary rules (which opportunities don't even get a look: through its years of compulsive acquisitions, Cisco only evaluated companies of up to 75 employees, 75% of them engineers), priority rules (how the scarce resource gets allocated when requests compete), timing rules (Nortel: development cannot run past 18 months), exit rules (which signal closes a project, with no committee involved). The simple rule is decentralization's coordination technology: it turns senior management's judgment into something that travels all the way to the last node without needing its signature.
Priority rules also conceal the most spectacular case of well-designed decentralization on record. At Intel, production capacity was allocated across products by a simple rule: wafers went where the margin per wafer was highest. When memory prices collapsed in the 1980s, the rule —applied by middle managers, with no strategic decision from the top— quietly moved capacity toward microprocessors. By the time Grove and Moore asked their famous new-CEO question (previous series, Part 1) and "decided" to exit memories, the organization had already exited: memory production was down to a sliver. Robert Burgelman, who documented the case, summed it up in a lesson that is uncomfortable for the top: a company's real strategy is the one its resource-allocation rules execute, not the one its plan proclaims. A well-chosen rule can see the regime change before the board does.
Third piece: shared context. Decentralization fragments when the nodes decide from different maps. The countermeasure is not to recentralize the decision but to centralize the context: every node looking at the same dashboard —cash, demand, priorities—, working from the same strategic thesis and knowing what is sacred. It is the "shared mindset" that the teams series identified as the extra condition for dispersed teams, scaled up to the organization. The short formula: local decision, global information, common doctrine.
Exploring and exploiting under one roof
That leaves turbulence's hardest structural problem, and it predates any crisis: a company needs to exploit its current business (efficiency, continuous improvement, discipline) and to explore the one that will replace it (variation, trial, the tolerance for failure from Part 2) —and the two activities destroy each other when they live inside the same structure, because the big business always wins the budget fights against the embryo, and the metrics that make an operation efficient kill the experiment in its cradle.
Charles O'Reilly and Michael Tushman studied 35 breakthrough innovation attempts across 15 business units in 9 industries, comparing four ways of organizing them: inside the existing functional structure, as cross-functional teams, as unsupported standalone teams, or as ambidextrous organizations —exploration housed in a separate unit with its own culture, processes and metrics, but integrated at the top, in the executive team that protects it and gives it access to the house's assets (brand, customers, plant)—. The scoreboard was brutal: more than 90% of the ambidextrous designs hit their targets, against 25% of the functional ones and 0% —zero— of the cross-functional teams and the unsupported standalone ones. The pattern of the failures is instructive: inside the structure, exploration suffocates; outside it and without sponsors, it starves. It survives when it is separate below and integrated above. USA Today used the design to integrate newspaper, web and TV and make money straight through the dot-com collapse while its peers bled out; Ciba Vision launched from autonomous units the lenses that took it from US$300 million to more than US$1 billion in a decade.
Ambidexterity is the structural answer to the core rigidity of Part 2: it doesn't ask the efficient organization to "become innovative" —a request that always fails, because it asks one place to be two things at once— but to house the two logics in different places and integrate them under a single executive head. It is, once again, the portfolio from Part 2 turned into structure: exploitation is the commitment; exploration, the options; the executive team, the portfolio manager.
| Symptom today | What it will cost in turbulence | Countermeasure |
|---|---|---|
| Every investment, price or hire climbs two levels | Weeks of delay exactly when local information is worth the most | Real autonomy thresholds per node, with a management floor (targets + monitoring) |
| Thick manual, thin judgment | Paralysis in the face of the case the manual never foresaw | 2–7 simple rules: boundary, priority, timing, exit |
| Every area with its own map of the situation | Fast nodes running in opposite directions | A common dashboard, an explicit thesis, a doctrine of what is sacred |
| Innovation reports to the operation that will cannibalize it | The embryo loses every budget round to the big business | A unit separate below, integrated above (ambidexterity) |
| With every scare, approvals get recentralized "for now" | The "for now" turns permanent; the organization unlearns how to decide | Tell apart what the crisis should centralize (cash) from what it shouldn't (local commercial decisions) |
| The proclaimed strategy and the real resource allocation never speak to each other | The company executes a different strategy without knowing it | Audit the allocation rules: that is the real strategy (Intel) |
Table — Symptoms of a structure that will not hold up in the storm. 32sur, built on the evidence cited in this article.
Questions for Monday
For your organization, exactly as it decided the last time there was a jolt:
Seven questions for Monday
- In your last crisis —a devaluation, a large customer lost, a critical shortage—, which decisions were recentralized? Which of them depended on fresh local information —and what did the delay cost?
- What can one of your branch or plant managers decide today without asking permission: hire, adjust a price, switch a supplier, invest how much? Was that threshold designed, or is it the sediment of old scares?
- Does your delegation have a management floor —targets, dashboard, consequences— or would delegating today amount to letting go of the wheel? (If it's the latter, the problem isn't decentralization: it's the floor.)
- Could you write, on an index card, the 2–7 rules your middle managers use to decide without you? Do they know them —or is each one improvising a set of their own?
- Which resource-allocation rule really governs your capacity, your cash and your people —and what strategy is that rule executing, with or without your signature?
- Your most important exploratory bet: does it report to the operation it is going to cannibalize, does it live isolated and without sponsors, or is it separate below and integrated above? What metrics is it judged on —the mature business's?
- If you and your inner circle were out of contact for two weeks starting tomorrow, would the company keep deciding sensibly —or would you discover that the "organization" was you plus assistants?
If several of the answers are uncomfortable, there is good news: unlike culture, structure changes by decision —thresholds, rules, dashboards and reporting lines are ink, not genetics—. The hard part isn't the redesign: it is getting the top to accept that its value added in the storm is not to take every decision, but to build the system that decides well without it.
The series closes here. Part 1 showed that turbulence is not the universal weather the conference circuit sells but a local regime that can be measured —and that what really has shortened is the life of each individual advantage—. Part 2 turned that measurement into a portfolio: commit all the way where irreversibility protects you, take options on what uncertainty won't let you decide yet, and fund by milestones that buy information. This part supplied the structural condition: none of it runs on an organization where every decision climbs two floors —a company's speed is the speed of its slowest node with authority—. The three parts share a single conviction, which is also the conviction of the earlier series: in turbulence the winner is not the one who forecasts the storm best —it is the one who built, beforehand, the ship that sails it: a thermometer of its own, a portfolio with options, and a crew that knows how to decide when the bridge doesn't answer.
Does your company depend too much on its bridge?
32sur works on decision architecture as part of your management processes: autonomy thresholds with a floor of targets and monitoring, simple rules for prioritizing and exiting, dashboards that share context without recentralizing the decision, and the design of exploratory units kept apart from the operation that would suffocate them. We don't sell org charts: we install the structure that decides well when you are not in the room. If your company depends too much on its bridge, let's talk.
References
- Aghion, P., Bloom, N., Lucking, B., Sadun, R. and Van Reenen, J., "Turbulence, Firm Decentralization, and Growth in Bad Times", American Economic Journal: Applied Economics, 13(1), 2021 — WMS (~1,300 firms, 10 countries) and MOPS (~8,800 US plants): −3.6/−3.5 pp in the sales decline of decentralized firms over 2008–2011; effect concentrated where there was turbulence; ~15% of the growth differential between countries.
- Hayek, F. A., "The Use of Knowledge in Society", American Economic Review, 35(4), 1945 — the knowledge of the particular circumstances of time and place; why the center decides late.
- Sadun, R., Bloom, N. and Van Reenen, J., "Why Do We Undervalue Competent Management?", Harvard Business Review, September–October 2017 — the management floor (targets, monitoring, consequences) that makes delegation safe ("Deciding under uncertainty" series, Part 2).
- Eisenhardt, K. M. and Sull, D. N., "Strategy as Simple Rules", Harvard Business Review, January 2001 — 2 to 7 rules learned from experience: boundary (Cisco: ≤75 employees, 75% engineers), priority, timing (Nortel: <18 months), exit (Oticon).
- Burgelman, R. A., "Fading Memories: A Process Theory of Strategic Business Exit in Dynamic Environments", Administrative Science Quarterly, 39(1), 1994 — Intel: the margin-based allocation rule moved capacity to microprocessors ahead of the formal decision at the top.
- Grove, A., Only the Paranoid Survive, Currency, 1996 — the new-CEO question ("Deciding under uncertainty" series, Part 1) and the exit from memories.
- O'Reilly, C. A. and Tushman, M. L., "The Ambidextrous Organization", Harvard Business Review, April 2004 — 35 breakthrough innovation attempts, 15 units, 9 industries: >90% success in ambidextrous structures vs. 25% functional and 0% in cross-functional or unsupported standalone teams; USA Today and Ciba Vision.
- March, J. G., "Exploration and Exploitation in Organizational Learning", Organization Science, 2(1), 1991 — the founding tension between exploring and exploiting.
- Leonard-Barton, D. (1992) and Ghemawat, P. (1991) — core rigidities and commitment (Part 2 of this series).
- Haas, M. and Mortensen, M., "The Secrets of Great Teamwork", Harvard Business Review, June 2016 — shared mindset as a condition for dispersed teams ("High-performance teams" series), scaled here into organizational doctrine.