Organizational Design: Choosing What Will Be Easy to Build

An organisation chart looks like a diagram of authority — who reports to whom, who decides what.

It functions as something else: a map of which conversations are cheap and which are expensive.

Two people on the same team, sharing a manager and a set of objectives, coordinate by talking. Two people in different divisions coordinate through meetings, agreements, escalation, and the schedules of people who have other priorities.

That difference is the entire practical content of organisational structure, and it is why Conway's Law holds: the systems an organisation builds end up shaped like its communication paths, because designing across an expensive boundary is harder than defining an interface and working around it.

Which means choosing a structure is choosing what will be easy to build.

The trade nobody escapes

The two basic options both work and neither is free.

Group by function — all the engineers together, all the designers, all the salespeople. Expertise deepens because specialists learn from each other, standards stay consistent, and career paths are clear. The cost is that delivering anything crosses boundaries, so shipping requires negotiation between groups with different priorities and separate schedules.

Group by product or customer — cross-functional teams owning an outcome end to end. Delivery is fast because everyone needed is in one place. The cost is duplication and drift: five teams each solve the same problem differently, specialists have nobody to learn from, and standards diverge.

Matrix structures attempt both by giving people two reporting lines. This genuinely addresses the trade-off and introduces a new problem — when the two lines disagree, someone in the middle absorbs the conflict, and it is usually the person with least power to resolve it.

There is no arrangement without a cost. The useful question is not which structure is correct but which failure mode you can better tolerate right now, which changes as an organisation grows and as its constraints move.

What growth does

Coordination costs rise faster than headcount, for the reason described in Brooks's Law: the number of possible pairs grows with the square of the team size.

Below a certain scale this is invisible. Everyone knows what everyone else is doing, decisions happen in conversation, and process would be overhead. Organisations at this size are often unusually effective and frequently attribute it to culture when much of it is arithmetic.

Past that point the informal approach fails, usually not gradually. Something is missed, a decision gets made twice, two teams build the same thing. The typical response is to add process — and the typical error is to add too much, too late, in reaction to a specific failure rather than as a considered design.

The pattern recurs at each growth stage, which is why organisations often oscillate between over- and under-structure rather than settling. That oscillation is a delayed feedback loop, and the delay is why it overshoots in both directions.

Two things help. Adding structure slightly before it is needed rather than after a visible failure, which is unpopular because the need is not yet obvious. And removing structure that outlived its purpose, which almost never happens, because repealing a process requires someone to accept responsibility if the original problem recurs.

What reliably goes wrong

Reorganising instead of fixing the actual problem. Restructuring is visible, feels decisive, and is frequently a substitute for addressing something specific. It also has real costs — relationships broken, context lost, months of reduced output — that are rarely counted against the expected benefit.

Optimising parts rather than flow. Making each department efficient does not make the organisation efficient; work sitting in queues between departments is invisible on any single team's metrics. This is the theory of constraints applied to an org chart, and it is why cross-boundary work is where most delay actually accumulates.

Measuring what is easy to measure. Team-level metrics encourage local optimisation at the expense of the whole, which is Goodhart's Law with a reporting line attached.

Confusing structure with capability. No arrangement compensates for the wrong people in key roles, and no reorganisation fixes a strategy problem.

Assuming a structure that worked elsewhere will transfer. Published models come from organisations with particular histories, sizes, and products. Copying the visible structure without the underlying conditions usually reproduces the form and not the result.

The most useful diagnostic is a single question: which conversations are currently expensive, and are those the ones that matter most? If the answer is yes, the structure is fighting the work — and that is worth changing. If the expensive conversations are ones nobody needs to have often, the structure is doing its job, whatever it looks like on paper.