Distance Does Not Slow Decisions

Here is a thing almost every organization believes about itself and almost none can demonstrate. Ask five leaders to write down, independently, who has the authority to approve a five thousand dollar unplanned spend, who can change a delivery date, and who can kill a piece of work that is no longer worth doing. Then compare the answers.

The results are rarely flattering, and the interesting part is that the organization was functioning anyway. It was functioning because the ambiguity was being resolved continuously, in seconds, by people who could see each other.

The distinction that explains the whole thing

Aghion and Tirole separated two things that organizations routinely confuse. Formal authority is the right to decide, the one on the org chart and in the delegation of authority matrix. Real authority is effective control over what actually happens, and it is determined by the structure of information. Whoever knows enough to make the call tends to make it, whatever the chart says.

Their further point is the one leaders find uncomfortable. Increasing an agent's real authority promotes initiative and costs the principal control, and the two are not separable. You cannot have the initiative without giving up some of the control. That is a trade rather than a problem to solve.

Now apply it to a floor plan. In a co-located organization, real authority routes itself. The engineer who understands the constraint mentions it to the manager walking past, gets a nod, and proceeds. Nobody logged a decision. Nobody consulted a matrix. The information found the shortest path to a call, and the organization moved.

Proximity is not a governance mechanism. It is a very fast workaround for the absence of one, and it works so well that nobody ever writes the real thing.

What happens when the shortcut is removed

Take away the corridor and the organization has to fall back on what it actually wrote down. In most cases that is a reporting line, a spend threshold and a hope. Everything not covered becomes a question, and a question in a distributed organization has exactly one delivery mechanism available, which is somebody's calendar.

This is why the largest measured effect in the literature is a coordination effect. Gibbs, Mengel and Siemroth tracked more than ten thousand IT professionals and found time in coordination and meetings rising while uninterrupted work hours shrank considerably. Yang and colleagues found communication shifting toward asynchronous channels across a firm of roughly 61,000 people. Neither of those is a statement about motivation, discipline or pajamas. They are statements about an organization discovering how many of its decisions had no home.

Heath and Staudenmayer called the underlying habit coordination neglect. Organizations are far better at dividing work than at integrating it, because dividing feels like organizing and integrating feels like overhead. The integration was happening anyway, informally, for free, and it was invisible precisely because nobody had designed it.

The trust finding, which is not the soft one

Bloom, Sadun and Van Reenen collected data on how far decisions about investment, hiring, production and sales were pushed from headquarters to local plant managers in almost four thousand firms across the United States, Europe and Asia. Firms in high-trust regions were meaningfully more decentralized, and decentralized firms earned higher returns from information technology.

Read that last clause twice before the next policy debate. Distributed work is an information-technology-intensive operating model. The evidence says the return on such a model depends on how far authority has been pushed down, and how far authority has been pushed down depends on trust. So two organizations can buy identical tools, adopt identical policies, and get different results, and the difference will be attributed to the policy.

This also explains a pattern that otherwise looks like noise. The randomized trial that found no performance loss, Bloom, Han and Liang's study of 1,612 people at Trip.com, ran inside a company with established roles, review cycles and a functioning performance system. The randomized trial that found an 18 percent loss, Atkin, Schoar and Shinde's study of data entry workers, ran in a setting where the work carried almost no decision content at all and the home environment itself was the variable. Both findings are real, and they are measuring different organizations doing different things.

Why a mandate feels like it works

Bring everyone back and decision speed genuinely improves. Corridors work. A leader who does this will observe a real effect and will reasonably conclude that the policy was correct.

What has actually happened is that the workaround has been reinstated. The decision rights are still undefined, the approval chain is still four people long, and the organization has purchased its speed by requiring everyone to be physically available to everyone else. That is an expensive subscription, it renews annually, and the price includes everyone's commute.

It also has a ceiling. Proximity resolves ambiguity only within the radius where people actually encounter each other, which the engineering communication research puts at about fifty meters. The decisions that cross functions, floors or time zones were never being resolved by the corridor, which is why those are the ones that were slow in 2019 as well.

If bringing people back made your decisions faster, you have not learned that the office works. You have learned exactly how much of your operating model was never written down.

The diagnostic, and it takes about an hour

Authority is the most testable of the four conditions, because decisions leave traces.

• Take five decisions that went up the chain last month. For each one, name what the approver knew that the requester did not. Where the honest answer is nothing, the decision was sitting at the wrong level and distance is not why it was slow.
• Write the three thresholds down and compare. Spend, schedule, and stopping work. Ask five leaders independently. Any spread is a decision your organization has not made, and it is currently being resolved by whoever is most available.
• Count the meetings whose only output is a decision somebody could have made alone. These are the coordination cost made visible. Each one is a decision right that exists but was never stated clearly enough for the holder to use it.
• Ask what happens when somebody decides and is wrong. Aghion and Tirole's trade is real. If the organization cannot tolerate the loss of control, it will not delegate, and it will need the corridor forever.

None of those four questions contains the word office. That is the point. Whatever comes back is the actual constraint, and it will be the same constraint wherever everyone sits.