Flattening the org chart moves decisions upward

Between 1986 and 1999, more than 300 large US firms flattened their upper ranks. The average number of levels between division heads and the CEO dropped from 1.6 to 1.2, and the number of people reporting directly to the CEO climbed from 4.5 to almost 7. Julie Wulf, who built that panel with Raghuram Rajan, then looked at where decisions were actually being made. They had moved up. Her 2012 article in the California Management Review is titled The Flattened Firm: Not as Advertised, and its conclusion fits in one sentence: flattened firms concentrate more control and more decision making at the top.

The numbers show how. Of the five positions added to the CEO's direct reports on average over twenty years, four were functional heads such as the CHRO, the CIO or the CMO, who make calls for the whole company. Division heads lost ground, and their pay records it: as soon as functional heads joined the top team, firms in the panel paid division heads less. Interviewed by Wulf, CEOs said they had flattened to "get closer to the businesses" and to get more involved in operations.

The number of boxes on the org chart and the real distribution of decision rights are two different things. A company can flatten and centralize at the same time.

What we are actually comparing

Two numbers describe a structure. The number of levels separates the person doing the work from the person who arbitrates last. The span of control says how many people report to one manager. The two move in opposite directions: at a fixed headcount, widening spans removes levels.

A hierarchical structure stacks levels and keeps spans narrow. A flat structure does the reverse, few levels and wide spans. The word "flat" describes a count of floors. A flat organization can perfectly well have a founder who rules on everything, which is the most concentrated form of hierarchy there is.

Management style sits on a different axis. A pyramid can run with managers who delegate heavily, and a flat organization can run with a leader who keeps every call. That subject has its own article, horizontal versus vertical management. This one is about the shape of the org chart and where decisions land.

The main families of organizational structure

Henry Mintzberg published The Structuring of Organizations in 1979. He describes an organization first by its parts, the strategic apex, the middle line, the operating core, the technostructure and the support staff, then by the mechanism that holds the work together, mutual adjustment between colleagues, direct supervision, or standardization of work processes, outputs and skills. The typologies you meet everywhere carve up that same material. They say how the work is grouped and what coordinates it. On who holds the right to decide, they stay silent.

The functional structure

It groups people by discipline. Production has its own head, so do sales, finance and human resources, and each one concentrates the expertise of its field. A skill gets paid for once for the whole company, and the technical level rises because the specialists work alongside each other.

The bill arrives as soon as a subject crosses two departments. Nobody holds authority over both, so the call goes up a floor. The more different things the company does, the more those escalations clog the top. The four functional heads Wulf watches join the CEO's team come straight out of that logic. This form suits a company selling a coherent range on a stable market, and it matches what Mintzberg calls the machine bureaucracy, coordinated by standardizing work processes.

The divisional structure

It splits the company by product, by market or by geography, and each division carries its own functions with a result to hit. DuPont and General Motors invented it in the early 1920s, when diversification made grouping by discipline unmanageable. Alfred Chandler documented the shift in Strategy and Structure in 1962 and observed that a form which barely existed in 1920 had become, by 1960, the accepted form for large diversified firms.

Headquarters then steers by the numbers, what Mintzberg calls standardizing outputs. Each division gets room to breathe, and the company pays for it in duplicated functions, rivalry between divisions and haggling over transfer prices. This form suits a company serving markets different enough that a single arbiter has nothing useful left to say about each one.

The matrix structure

It ties one person to two lines at once, their discipline and their product or project. It came out of American aerospace in the 1950s and spread widely in the 1970s. The problem it solves is concrete: a complex program needs skills scattered across five departments, and standing up five redundant teams costs too much.

In exchange, the arbitration conflict lands on the person sitting at the crossing, with two bosses and two lists of priorities. Jay Galbraith, who spent his career on this subject, held that organization structures rarely fail by themselves and that management fails at implementing them. His star model asks that strategy, processes, reward systems and people all line up with the structure. A matrix decreed by an org memo, without that tuning, turns every priority into a negotiation.

Flat or tall organizational structure

A flat organizational structure keeps two or three levels between the person doing the work and the person who arbitrates last, with each manager overseeing a wide group. A tall organizational structure stacks more floors and gives each manager a narrow group. That distinction covers something other than the three forms above. Functional, divisional and matrix say by which principle the work is grouped, flat versus tall gives a proportion. The two settings combine freely. A functional company of thirty people fits on two floors, a divisional group of fifty thousand runs ten, and the matrix shows up in both sizes.

None of these labels says who decides. A functional structure can let each department rule alone on its own field, or send the smallest call upward. A divisional structure can hand its division heads a real mandate, or a budget to execute. The grouping principle and the decision rights get chosen separately, and it is the second one that changes how people spend their day.

What hierarchy buys you

The pyramid came out of a coordination problem across distance, on the American railroads of the 1850s. It still solves that problem, and it drags five structural problems along with it. Three real services explain why it has lasted a hundred and seventy years.

It filters. Raaj Sah and Joseph Stiglitz modeled this in 1986 in the American Economic Review. A hierarchy sends every project through several approvals in series, so it rejects a larger share of the good ones. An architecture where several units decide in parallel accepts a larger share of the bad ones. A nuclear operator and a medical device maker prefer the first kind of error, and they are right to.

It gives problems an address. When an issue outgrows a team, everyone knows who it goes to and when it comes back.

It draws a legible path. Levels form a progression people understand, with a title, a scope and a salary at each floor. Flat organizations have to invent something else in its place, and plenty of them fail at it.

Where flat structures break

Coordination cost grows faster than headcount

The number of pairs of people in a team follows n(n-1)/2. At 10 people you get 45 possible links. At 50 you get 1,225. At 150 the count reaches 11,175. A hierarchy cuts that network into subsets and pays for the cut in filtered information. A flat organization keeps the whole network and pays in coordination time.

Medium put a price on that bill. Andy Doyle announced on 4 March 2016 that the publishing company was dropping Holacracy: "we found it difficult to coordinate efforts at scale," and the system had "begun to exert a small but persistent tax on both our effectiveness, and our sense of connection to each other." He added that codifying accountabilities in detail had dampened initiative and the feeling of communal ownership.

W.L. Gore, the maker of Gore-Tex, solved the problem by capping the size of its units. More than 10,000 people work there without a conventional org chart, and each plant is capped at roughly 150 to 200 people. Past that, the company opens another one next door. Gary Hamel, who has studied Gore closely, sums up founder Bill Gore's observation: past a certain headcount, "we decided" turns into "they decided."

GitHub tested the upper limit without that guardrail. The company was founded in 2008 with no managers and no titles. In spring 2014, the engineering department started assigning managers, a few weeks after Chris Wanstrath took over as CEO. Bloomberg told the story in September 2016, when headcount was approaching 600. Julio Avalos, then an executive at GitHub, summed up what had been missing: "Without even a minimal layer of management, it was difficult to have some of those conversations and to get people feeling like they understood what was expected of them."

Informal power fills the space

Jo Freeman wrote the reference text on this in 1972, in The Second Wave, drawing on American feminist collectives. The Tyranny of Structurelessness makes one plain claim: every organization has a structure, and the only question is whether it is written down. Left implicit, it falls back on friendships, seniority, availability and charisma. Freeman defines the elite that emerges as "a small group of people who have power over a larger group of which they are part, usually without direct responsibility to that larger group, and often without their knowledge or consent."

Fifty years later, accounts from inside Valve describe the same mechanism. The studio's onboarding handbook, which advertises a company without bosses, spread widely after it leaked in 2012. Jeri Ellsworth, let go in early 2013, gave the other side: "There is actually a hidden layer of powerful management structure in the company and it felt a lot like high school." Rich Geldreich, a developer who worked at the same studio, wrote about a company he calls "SelfOrganizingCo," with its "barons" close to the executive team and its game of currying favor with sponsors to keep your job.

A manager has to answer for a call. Nameless power escapes that question, which makes it far more comfortable for whoever holds it.

The question that actually settles it

The useful criterion is about decision rights. For every type of decision that matters, an organization needs an answer to three questions: who decides, over what scope, and where that answer is written.

A pyramid answers with the job title, which works as long as the work fits inside the job. A flat organization that has written its answers works too. A flat organization that left them implicit watches the vacuum fill itself, and Freeman explains who fills it.

Here is how the question plays out by situation.

Situation	What works
The costly error is approving a bad project (nuclear, healthcare, compliance)	Approvals in series, so levels
The costly error is missing a good project (emerging market, research)	Decisions in parallel, so wide spans
The decisive information sits with the people doing the work	Decision rights attached to the role rather than to the level
The work changes faster than the job descriptions	Revisable roles rather than levels to renegotiate
One person wears several hats	Several roles per person rather than one job title
Headcount passes 100 to 150	A formal place where the structure gets revised on a known rhythm

What replaces the layers

An organization that does without management layers replaces them with written structure. Each role carries a purpose, a domain it decides on alone, and accountabilities others can ask it for. One person holds several roles, and a role changes when the need changes. That is the whole difference between a role and a job description.

Two more pieces go with it. The first is a known rule for anything that outgrows a single role, such as consent decision-making. The second is a recurring slot where the structure gets revised, the governance meeting. Without that slot, the org chart freezes and the gap with the real work widens month after month.

Buurtzorg and Morning Star, the two organizations everyone cites as running without managers, have more written structure than the average company. It is simply distributed instead of stacked.

Rolebase is where that structure gets written. The org chart shows roles with their purpose, domain and accountabilities, decisions stay attached to the role that made them, and a governance mode sets who can change what: Free, where every member edits the whole org chart, Agile, where each role is editable by its leaders, or Strict, where any structural change goes through a proposal. See the features, or create your organization free for up to five active members.

How big can a flat organization get?

No single threshold fits every organization, and the documented cases cluster around 100 to 150 people per autonomous unit. W.L. Gore has capped each plant at 150 to 200 people for decades. GitHub appointed its first managers in spring 2014, on the way to 600 employees. The more useful signal is the moment when nobody can say who decides what any more.

Does a flat structure get rid of bosses?

It gets rid of management floors. Authority moves either to named roles or to informal power that nobody appointed. Jo Freeman described as early as 1972 how the absence of written structure installs an elite with no responsibility to the group.

Does flattening the org chart speed up decisions?

Not on its own. Julie Wulf showed in 2012, across more than 300 large US firms tracked from 1986 to 1999, that flattened firms concentrated more decisions at the top. Speed comes from decision rights that are explicitly delegated, which is a separate choice from the number of levels.

What companies have a flat organizational structure?

W.L. Gore runs more than 10,000 people without a conventional org chart, capping each plant at roughly 150 to 200. Buurtzorg and Morning Star are the two most cited examples of organizations operating without managers. Valve advertises a company without bosses in its onboarding handbook, and accounts from former employees describe an informal hierarchy underneath. GitHub and Medium both started flat and stepped back, GitHub by appointing managers in spring 2014, Medium by dropping Holacracy in March 2016.

What are the disadvantages of a flat structure?

Three costs show up consistently. Coordination grows faster than headcount, since the number of pairs of people follows n(n-1)/2, which is 45 links at 10 people and 11,175 at 150. And power that nobody wrote down reassembles around friendships, seniority and charisma, the mechanism Jo Freeman named the tyranny of structurelessness in 1972. Career paths are the third weak spot, because levels came with a title, a scope and a salary at each floor, and a flat organization has to invent something in their place.

What does a hybrid structure mean in practice?

It means varying the number of levels according to the kind of error you fear. Compliance, safety and finance keep approvals in series. Product and field teams get decision rights attached to their roles. What both regimes share is that each scope is written down and open to revision.