The Silo Paradox

Skip to main content

Imagine living in an old town. Beautiful coffee shops, fancy restaurants, hand-made craft stands. The city is wealthy, and so are the people who run its local businesses, yet they rarely need to talk to each other. They prefer to work independently, make their own decisions, and rely on their own supply sources. That’s enough to succeed. The city functions well without any urge for close business ties.[1] Nobody feels they’re missing anything.

Until an emergency happens that requires all of them to sit down together.

Organisations evolve in much the same way. Initially, knowledge is spread everywhere. It’s not that hard to understand the business as a whole, and if you don’t, you know exactly whom to ask for an immediate answer.[2]

Then people realise this isn’t efficient.[3] To move things forward, they need to collaborate. A lot. So they start organising themselves into structures scoped to their own area of responsibility. If they don’t know something, they figure it out on their own to stay efficient and avoid coordination overhead. Why shouldn’t they? Effectiveness is what the business rewards, and unnecessary meetings only slow things down. Building relationships takes effort that could otherwise go into development.

This is how boundaries emerge, naturally.[4] Seen from inside the story, it’s entirely logical. Yet over time, quietly, it erodes the company’s ability to share knowledge at scale. As a consequence, leaders start talking about silos and stress the need to share knowledge more widely. When that doesn’t work, they find a culprit: culture.[5] But is it really a cultural problem?

Every new boundary makes local thinking easier, which is expected behaviour from a team’s perspective.[6] Whenever a system is decomposed into independent parts, those parts will optimise for themselves unless interdependence is created. This way, teams can stay flexible and avoid cross-team battles, even though that goal is rarely stated outright. It also helps them handle complexity: each team can focus on tending its own garden rather than spreading its attention across multiple contexts.

Now scale this up to the level of a whole company: every team ends up with its own vector, not necessarily correlated with what the organisation actually needs. It’s hard to call this balanced when there are no balancing loops in place to correct it.[7]

Once we understand that this is predominantly a structural problem, we can apply structural changes at the organisational level, instead of cultural ones. Bento, Tagliabue and Lorenzo’s research[4:1] proposes targeted bridges between clusters in the form of team rotations, shared goals, and joint metrics. Star and Griesemer’s boundary objects[8] describe the same idea from another angle. Shared artefacts, reliable enough to hold a common identity across teams yet flexible enough to fit each team’s local needs, make a dependency between teams explicit. The reverse Conway maneuver works in the opposite direction: by deliberately restructuring teams first, you can reshape a system’s architecture over time, using the structure itself to streamline the flow.[9] All of this comes down to engineering a bit of friction that makes collaboration structurally necessary, not optional.

Going back to the city: after the emergency, the merchants finally build a guild, a town council, and shared infrastructure to make collaboration possible, because the crisis forced new structures into place. Wherever a system allows its parts to succeed independently, local optimisation will emerge naturally. When you change the structure, flow changes with it.

Deliberately designed networks and bridges, literally and metaphorically, are how organisations create cohesion.

Theory of the firm (Coase, R.H., 1937): argues that a firm’s boundaries are set by weighing the cost of coordinating through the market against the cost of coordinating internally, so independence holds only for as long as that trade-off favours it. ↩︎

Dunbar’s number (Dunbar, R.I.M., 1992): proposes a cognitive limit of roughly 150 stable relationships a person can track, below which a group can still function informally because everyone can keep tabs on who knows what. ↩︎

Communication overhead / Brooks’s Law (Brooks, F.P., 1975): shows that the number of communication channels in a group grows as n(n-1)/2, so adding people to a team increases coordination cost quadratically rather than linearly. ↩︎

Organizational silos as a network phenomenon (Bento, F., Tagliabue, M. & Lorenzo, F., 2020): scoping review finding that silos typically emerge on their own, from ordinary patterns of interaction and reinforcement, not because anyone deliberately designs them that way. The resulting fragmentation is still structural, a matter of how information flows through the organisation, not of people’s attitudes. The same review found that targeted interventions (e.g. team rotations, shared goals) can...