Free
Illusions in the Boardroom
The board-level diagnosis
Illusions of Work
The operator manual
What the Competition Does Instead
The strongest objections to the argument: genuinely irreducible complexity, scale and legacy, why large firms centralised in the first place, and how political immune systems kill reform before evidence can appear.
The argument is deliberately forceful. Before anyone acts on it, it deserves a stress test.
Not every shared service is a power grab. Not every coordination role is busy work. Some organisations are managing genuinely irreducible complexity. The question is how to tell the difference.
Anti-money-laundering in a bank touches every product, every customer relationship, and every transaction. Assigning AML to a single autonomous unit would either duplicate the capability across every unit or create a central function with cross-cutting authority, which is what the bank already has. AML is a cross-cutting concern that requires central coordination. The structural correction is to make that coordination machine-readable and contract-mediated rather than pretending it can be eliminated.
The test: remove the coordination role and ask what breaks. If the answer is “the two teams would make contradictory decisions about the same customer,” the coordination is real. If the answer is “they would deploy without asking permission,” it is governance theatre.
The autonomous unit model works cleanly for a mid-sized insurer with five core processes. A global bank with forty thousand employees and operations across thirty countries needs a longer timeline, a more distributed authority structure, and a willingness to accept that some parts will operate in the old model while others transition. The structural argument holds at scale; the implementation becomes harder and more politically expensive.
Some systems cannot be restructured at all because the cost exceeds the remaining lifetime of the system or the institutional knowledge needed to restructure it has already left. The honest response is to draw the boundary around what can be corrected and start there. The competitor’s advantage does not require perfection. It requires starting.
The strongest objection to this argument is not “this is hard.” It is: “you are underestimating why large firms centralised these controls in the first place.”
Large organisations did not create centralised governance out of stupidity. They created it because decentralised operations in the 1990s and 2000s produced real failures: inconsistent customer experiences, regulatory violations, security breaches caused by teams deploying without coordinated controls. Centralisation was the rational response, and in many organisations, centralisation worked. In some it still works. A global bank whose central risk function prevented a 200 million exposure during the 2022 rate shock did so precisely because the function had cross-cutting authority that no autonomous unit would have possessed. That is not busy work. That is governance doing what governance is supposed to do.
The honest version of the argument is not “all centralisation is theatre.” It is that the mechanism for achieving coherence has changed, and that many organisations are paying the full cost of human coordination when a substantial share of it could be mediated by structural contracts and machine verification. Firms that cannot tell the difference between coordination that protects and coordination that persists for its own sake will not benefit from this analysis.
The cost of centralisation has changed because the technology for coordination has changed. When the only mechanism for cross-unit coherence was a human being in a meeting, centralisation was the cheapest option. When machine-readable contracts and continuous AI reconciliation can mediate interactions at negligible cost, the coordination overhead of centralisation becomes the most expensive option. Whether that mechanism change is sufficient for a given organisation, at a given scale, is an empirical question rather than an ideological one.
A Nordic retailer created an “autonomous unit” around fulfilment. The unit received a charter, a process definition, and a budget envelope. The unit was not given the data, the deployment authority, or the hiring authority. Within six months, the unit’s lead was spending more time defending the charter than operating the process. The board concluded that “the autonomous unit model doesn’t work for us.” The model had not been tried.
The political failures are more common and more subtle:
Enthusiastic launch, quiet withdrawal. The CEO’s attention moves to an acquisition. The CTO carries the political weight alone. Central functions reassert themselves through “enterprise standards.” Nobody cancels the transition. The transition stops being protected.
Support that sounds like support. “We fully support the model, and we think it would benefit from a lightweight governance overlay.” Within three months, the unit operates under the same coordination layer the charter removed, now called “governance support.”
Measurement capture. The board asks for evidence. The evidence is defined by the functions the model displaced: programme metrics, alignment scores, stakeholder satisfaction. The unit’s actual operating improvements are classified as “not yet sufficient to offset the governance risk.”
Talent poaching. The unit’s best people are “temporarily seconded” to an urgent enterprise initiative. The unit’s performance declines. The decline is cited as evidence that the model underperforms.
Each mechanism is individually rational and collectively lethal. The defence is authority sustained long enough for the evidence to appear. That window is approximately twelve months.
Failed reform immunises the organisation against the real thing.
See also: All articles · Illusions in the Boardroom · Illusions of Work