On a Toyota assembly line there is a cord above the workstation, and any operator who sees something wrong can pull it. Toyota's own description of the system is unglamorous about what happens next: the andon board lights up and the operator calls the person in charge, and the line waits.1 That is the whole grant. The operator does not get to redesign the fixture or reopen the question of whether the model should have been built this way at all. Authority on that line is one decision wide, and it was cut to that width deliberately.

Organizations that admire the cord tend to import it as a mood. We push decisions down. We trust the people closest to the work. That language survives a town hall and almost nothing after it, because it treats delegation as a single setting on a single dial, turned up when leadership wants speed and turned back down after something goes wrong.

The dial by itself does nothing

Allison Koester of Georgetown and Junwei Xia took the question at the level where it can actually be measured, using confidential survey data from manufacturing firms to compare how far decision rights had moved from headquarters out to plant managers against how those firms performed financially. Their working paper, revised earlier this year, reports that a firm's overall level of delegation was unassociated with financial performance.2 Not weakly positive, and not positive with conditions attached either. The total volume of authority pushed downward predicted nothing at all.

Composition was where the relationships lived. Delegating employee hiring, and delegating customer-facing sales and marketing calls, was positively associated with performance, which the authors read as delegation earning its keep where a decision leans on local knowledge that travels badly up a reporting line.2 Delegation of capital investment and new product introduction ran the opposite way and was negatively associated with performance, in the territory where a decision needs coordination across units that only the center can see.2 One dial, and the direction of benefit reverses depending on which decision you happen to be holding when you turn it.

Delegation gets debated as a temperature setting for the whole organization. It behaves like a separate answer to every class of decision you make, and those answers do not all point the same way.

The one they always take back

Earlier this year, in a discussion group I am part of, a manager walked through what her own promotion had actually consisted of. She had the title and the team under her. Her boss kept final say on every hire she wanted to make. Her workaround was to copy him on every decision she made for about three weeks, including the trivial ones, until he stopped wanting to see them and let her staff her own group. She told it as a story about patience wearing someone down.

Read it against the evidence and it becomes something less flattering to her boss. Hiring is the decision class most dependent on knowledge he did not have and could not easily be handed, given where he sat. It is one of the two categories where letting go shows up on the scoreboard. Of everything on her desk he could have reserved for himself, he reached for the one with the best case for staying where it was.

Formal authority and the real thing

Philippe Aghion and Jean Tirole separated these two ideas almost thirty years ago in the Journal of Political Economy, distinguishing formal authority, the right to decide, from real authority, effective control over what actually gets decided. Real authority, in their account, follows the structure of information. They also list the conditions under which a subordinate ends up holding real authority inside a structure that never formally granted it, and the list includes overload of the superior, urgency of the decision, and having several bosses at once.3

Anyone who has staffed a capital committee will recognize that as a description of a normal quarter. The chart says the investment decision belongs to the committee. The committee then sees more proposals in an afternoon than it can properly interrogate, and the one arriving with a clean model and a sponsor who answers questions without hesitating tends to clear. The real decision was made by whoever built the model. Nobody voted to delegate it. It moved on its own, toward the person holding the information, exactly where the theory says it will go.

Which turns the empowerment question inside out. If real authority migrates toward information anyway, then the useful work sits a long way from the announcement. It is going decision class by decision class and asking a narrow question about each one: does this call mostly require knowledge that exists at the edge, or does it mostly require a view across units that exists only at the center? Hiring and local customer judgment answer one way. Capital allocation and the question of what to build next come out somewhere else entirely. An organization can honestly be decentralized in the first and centralized in the second without any inconsistency, and the leaders I find most convincing are the ones who can say out loud which is which and why.

I have argued before, more than once, that the more common failure runs the other direction, that responsibility gets handed down while the authority to act stays put. I still think that is the usual shape of the problem. This research adds a constraint I had not been careful enough about, which is that the repair has to be specific. Pushing everything down is a different mistake wearing better clothes.

So a question for anyone running a leadership team right now. If you listed the six decisions your organization makes most often, could you say for each one where the authority sits today, and could you defend that placement with something other than how it has always been done?