The Hallway Was Load Bearing

Yang and colleagues did something most studies of this cannot. They watched the collaboration network of a very large firm, roughly 61,000 people, before and after it went firm-wide remote, and measured the shape of the network rather than asking anyone how they felt about it.

The network became more static and more siloed. Fewer ties bridged distinct parts of the organization, and the ties that did exist were less likely to change. Communication shifted toward asynchronous channels. Their conclusion was that this would make it harder for employees to acquire and share new information across the organization.

Gibbs and colleagues found the same shape independently, reporting that people networked with fewer individuals and fewer business units. Two different firms, two different methods, one finding. The inside of each group held. The space between groups emptied out.

Your teams did not stop talking. They stopped talking to anyone they were not already talking to.

Why the thin ties are the ones that mattered

This is where a fifty year old sociology paper earns its keep. Granovetter's argument was that the people you are closest to are embedded in the same cluster you are, so their information overlaps heavily with yours. The connections that bring genuinely new information are the weaker ones, because those reach into parts of the network your own circle does not touch.

For decades that was a beautiful idea with correlational evidence behind it. Then Rajkumar and colleagues ran the causal test at a scale nobody else could. Using randomized experiments on LinkedIn's connection recommendations across more than twenty million people over five years, two billion new ties and six hundred thousand new jobs, they confirmed the effect and corrected it. The relationship is an inverted U. Moderately weak ties are the most productive channel, and the returns fall off again as ties get weaker still.

Which is a precise description of the relationship you had with the person two teams over whose name you knew, whose project you vaguely understood, and who you bumped into near the printer about once a fortnight. Not a friend. Not a stranger. Exactly the band the evidence says does the most work.

That relationship had no owner, no budget line and no place in the operating model. It was produced by a floor plan.

The part the office camp should read before the next memo

Thomas Allen spent more than a decade measuring how engineering organizations actually communicate, and produced the finding that still bears his name. The probability that two people communicate regularly about technical work falls off sharply with the distance between their desks, and is largely gone once that distance passes about fifty meters.

Fifty meters. Not one building. Not one campus. Roughly one floor of one wing, and that is the generous reading.

So the office was genuinely an alignment machine, and it was a local one. It manufactured weak ties beautifully among people who were already near each other, and did approximately nothing for the sales lead on the third floor and the architect in the annex. Most organizations then spent twenty years reorganizing by function, which seats the people who most need a shared picture as far apart as the building allows.

Bringing everyone back restores the ties that a floor plan can produce. If the gap you are trying to close is between two floors, the building was never closing it.

What actually degrades, in the language of the work

Alignment is not agreement. Two people can nod at the same words in the same meeting and walk out holding different pictures of what was decided, and nothing in the meeting reveals it. What closes that gap is a shared mental model, meaning a common representation of the task, the equipment and each other that lets people act independently and still act compatibly.

Mathieu and colleagues measured exactly that and found that greater convergence in those models produced better team processes, and that better processes produced better performance. The models are the mechanism. The performance is downstream.

Now consider how a shared mental model is normally maintained. Not in the quarterly review. In hundreds of tiny corrections, most of which nobody would call a conversation. Somebody overhears a wrong assumption and fixes it in four words. Somebody sees a diagram on a whiteboard and updates their picture without mentioning it. Those are the interactions that the network study found disappearing, and they are the ones nobody thought to replace, because nobody had ever noticed them happening.

Heath and Staudenmayer gave this failure a name two decades ago. They called it coordination neglect, and their argument is that organizations are much better at dividing a task than at integrating the pieces afterwards, because dividing is the part that feels like organizing. Distributed work did not create coordination neglect. It removed the informal integration that was hiding it.

The quieter finding, which is about people rather than networks

Bartel, Wrzesniewski and Wiesenfeld studied virtual employees and found something that sits underneath all of this. The more physically isolated somebody was, the less respected they perceived themselves to be within the organization, and that perception explained why more isolated people identified less with the organization.

That matters for alignment in a way that is easy to miss. A person who feels peripheral asks fewer questions. Asking a question is an admission of not knowing, delivered to an audience, and the calculation is different when you are not sure you are a full member of the group. So the alignment gap widens, and it widens fastest for exactly the people whose picture was least complete to begin with.

What to do about it, in ascending order of effort

• Run the two person test. Take two people on different teams who depend on each other. Ask each one separately what the current priority is and what the other team is working on. Compare. The distance between the answers is your alignment gap, expressed in the only units that matter.
• Schedule the weak ties, because they will not happen otherwise. The band that does the most work is the acquaintance, so build the thing that produces acquaintances. Rotating pairs across team boundaries, a genuinely cross-functional review, a standing slot where one team explains its current problem to another. It feels artificial because it is. The floor plan was artificial too, it was just already paid for.
• Make the shared picture an artifact. If the model lives only in people's heads, distance erodes it and nobody can see the erosion. Written decisions, a visible current priority, a one page picture of what each team is trying to do this quarter. The test is whether a new person could reconstruct the picture without asking anyone.
• If you do bring people in, be specific about who and with whom. Given the fifty meter finding, three days a week with your own function seated together is a different intervention from one day a week with the function you most misunderstand. Only one of those closes the gap that the research actually identified.