Broadly speaking, there are two primitive ways to organize teams. Functional organizations keep people in the same discipline in a single reporting tree, regardless of what products they work on. Divisional organizations keep people working on the same thing in a single reporting tree, regardless of what discipline they come from.

Functional organizations famously struggle to maintain speed at scale because of massive communication overhead. Two people with an intractable disagreement may have to escalate all the way to the top of the organization to get resolution. On the other hand, they’re known for quality and consistency, as people tend to prioritize craft and collaborate across products within their discipline.

Divisional organizations famously struggle to maintain consistency and quality at scale because silos develop. Two different teams facing the same problem may not build common infrastructure or design patterns because there isn’t a strong incentive to collaborate across divisions. On the other hand, they’re known to move very quickly and decisively because people prioritize the problem they have in front of them.

Facebook uses a hybrid of these models. For the most part, we’re organized functionally. But there are a small number of divisional leaders at the top to allow for some decoupling between groups. In my opinion, this has been very successful. We see lots of long-term investment in shared infrastructure, design language, tooling, etc. — and at the same time, individual product teams have been able to move quickly compared to industry peers.

When considering whether to keep a team centralized as opposed to federated into a division I ask whether a team:

  • Serves as fundamental infrastructure or operational capacity that benefits multiple divisions and creates economies of scale
  • Provides product or narrative cohesion across product groups
  • Requires independence is an important aspect of the function
  • Has limited capacity that needs to be centrally prioritized and managed for maximum impact
  • Is primarily composed of specialists whose work is not fungible or widely understood
  • Too small to maintain cultural independence, identity, career progression, and hiring momentum as part of a divisional organization

The more you try to maintain functional and divisional teams in parallel, the deeper you are into what is called a Matrixed organization. With sufficiently strong processes, leadership, and shared vision these can reap the benefits of both functional and divisional structures with fewer downsides. Unfortunately, more often than not they end up reaping all the downsides with few of the upsides over time. I would only indulge in this at full scale for long if I were prepared to build a bureaucracy to maintain it.

However we can harness some of the value of a matrix without deploying it so widely that the overhead gets the best of us. The seniormost leaders in Facebook’s hybrid organization are already there. They are tasked with understanding both each divisional investment as well each function it interacts with. This is a sufficiently focused deployment that we can keep teams aligned through relationship building and simple processes without requiring us to create a large bureaucracy.

For someone so dubious of org charts I am somewhat notorious for rearranging them. Indeed one reason I want people to underindex on them is because it makes it easier to change them. That allows us to react quickly when leadership changes to play to the strengths of the people we have. It gives the agility to adapt when the narrative we want to communicate about how we work has changed.

Perhaps the best insight I have gotten here came from Jocelyn Goldfein who suggested that the likely optimal path is to alternate between functional and divisional every few years. At each switch you quickly harness the benefit of the new structure but only slowly lose the benefit of the previous one as human connections decay slowly. Unfortunately this is very hard to pull off in practice because the transition between the two inevitably creates some larger roles just as it eliminates others so senior leaders would constantly be in the lurch. Perhaps that is why I’ve never seen it attempted.