Horizontal or vertical: it comes down to who decides

Between 1986 and 1999, large US companies stripped out layers. Raghuram Rajan and Julie Wulf tracked more than 300 of them and published the result in the Review of Economics and Statistics in 2006. The number of people reporting directly to the CEO went up, the number of levels between division heads and the CEO went down. The org chart flattened.

Maria Guadalupe, Hongyi Li and Julie Wulf then looked at who filled those new seats at the top. Their 2012 paper is called The Flattened Firm: Not as Advertised, and the title says it all. The seats gained around the CEO went to functional heads, finance, HR, IT, marketing. The matching decisions moved up to group level. The org chart got flatter and the decisions got more centralized.

Counting layers is therefore not enough. Vertical and horizontal management differ on where decisions get made and on which way information travels. You can delete three levels and stay deeply vertical. You can keep a six-level pyramid and let teams settle what matters.

What the two words actually cover

Under vertical management, authority follows the reporting line. A decision that exceeds one job's scope goes up a notch, then another, until it reaches someone entitled to settle it. Information comes back down the same channel, as targets and instructions.

Under horizontal management, authority follows the role. Whoever holds a role decides within its domain without asking for prior sign-off, answers to the roles concerned, and information travels sideways between the people who need it. A reporting line often survives, and it serves a purpose other than approving work.

Question	Vertical management	Horizontal management
Who settles it	The manager, or the manager's manager	The role whose domain it is
How information travels	Up the line, filtered at every level	Straight between the roles concerned
What the manager does	Approves, arbitrates, checks	The job splits across several roles
Where it hurts	Waiting, and decisions made far from the ground	Grey zones wherever a domain stays vague

How many levels a company has is a separate question, the shape of the org chart, which we cover in flat or hierarchical structure.

When vertical wins

Tom Burns and George Stalker set the frame back in 1961 in The Management of Innovation, based on fieldwork in Scottish electronics firms. They contrast the mechanistic system, with its chain of command, its specialization and its written procedures, against the organic system, where tasks get redefined as problems come up. Each one suits a specific context. Mechanistic fits stable environments and routine tasks, organic fits markets and technologies that keep moving.

Sixty-five years later, the literature still backs that split. Three situations hand the advantage to vertical.

Urgency, when a call has to be made in ten seconds

Katherine Klein spent ten months observing the trauma resuscitation teams at the Shock Trauma Center in Baltimore, with Jonathan Ziegert, Andrew Knight and Yan Xiao. Their paper appeared in Administrative Science Quarterly in 2006. These teams handle life-threatening emergencies, with membership churning constantly and novices to train live.

They keep an explicit hierarchy. The senior physician hands the active leadership role to a more junior one, then takes it back, sometimes several times during a single resuscitation. The authors call it dynamic delegation. When a mistake kills and the call has to be made in ten seconds, knowing who is in charge right now beats the best consensus.

The lopsided cost of a mistake

Raaj Sah and Joseph Stiglitz formalized that trade-off in the American Economic Review in 1986. A hierarchy pushes every project through several successive filters and rejects a larger share of the good ones. A polyarchy decides in parallel and lets through a larger share of the bad ones. The choice is about which kind of error you would rather make, as we set out in the problems with hierarchy.

A pharmaceutical cleanroom, an aircraft maintenance shop and a nuclear plant are right to filter. A good project killed costs them far less than a bad project approved. Add the traceability a regulator demands, a named chain of responsibility on every decision, and the reporting line becomes the simplest tool for the job.

Expertise that is very unevenly spread

Horizontal assumes the person closest to the problem knows enough to settle it. In a trade where ten years of practice separate a senior from a beginner, that assumption breaks. An accounting firm that lets an intern sign off a tax filing has confused autonomy with an empty chair.

One more parameter sits mostly outside a company's control. Nicholas Bloom, Raffaella Sadun and John Van Reenen measured the delegation of investment, hiring, production and sales decisions across nearly 4,000 firms in North America, Europe and Asia, for the Quarterly Journal of Economics in 2012. Firms headquartered in high-trust regions delegate significantly more. Together with the rule of law, product market competition and how hierarchical the dominant religion is, these factors account for four fifths of the cross-country variation in decentralization. Delegating costs more where trust is thin, and a single company shifts that dial slowly.

When horizontal wins

Philippe Aghion, Nicholas Bloom, Brian Lucking, Raffaella Sadun and John Van Reenen combined two large datasets on delegation, one covering ten OECD countries, the other US establishments, then looked at what happened during the 2008 crisis. Their paper appeared in the American Economic Journal: Applied Economics in 2021. Firms that had pushed power down to their plant managers before the crisis outperformed centralized ones, and the gap concentrates in the sectors hit hardest.

Their explanation rests on two effects pulling against each other. When everything moves, local information gains value and speed matters, which favors decentralized firms. Painful calls, closing a site, cutting deep into costs, come easier from the top, which favors centralized firms. Over the period studied, the first effect wins.

The same logic holds outside a recession. The faster a team's environment changes, the further from the top the person who spots the problem sits, and the more the trip upward costs in delay and distortion. Knowledge work meets those conditions all the time.

The fake flat

Valve, the studio behind Steam, saw its new employee handbook leak in 2012. It describes a company without managers, where everyone picks their own projects and wheels their desk over to whichever team looks interesting. The document travelled widely. In 2013, Jeri Ellsworth, let go after a year and a half at Valve, called it a pseudo-flat structure and described a hidden layer of powerful management, complete with popular kids and troublemakers. Other former employees have since told similar stories.

Dropping titles without moving decision rights makes power invisible. It re-forms around seniority, closeness to the founders and ease at winning a room, and nobody can challenge it because officially it no longer exists. Jo Freeman described that mechanism back in 1972 in The Tyranny of Structurelessness, writing about American feminist collectives. It is also what ran a good share of liberated companies into the ground.

Horizontal organizations that last carry more written structure than average. Mário Kaphan, co-founder of Vagas.com, described in March 2014 on the Management Innovation eXchange how his Brazilian online recruitment company runs. It employed 160 people at the time, with no titles assigned to anyone. Go-to people emerge because colleagues treat them as the reference on a topic, decisions get built by seeking consensus rather than settled by majority vote, and each area gathers its full team for two hours twice a month. Kaphan reports growth never below 24% a year since the company started in 1999. That figure comes from the founder, so treat it with care. The machinery he describes teaches more. Horizontal holds together thanks to regular meetings and an explicit decision rule.

Morning Star, the world's largest tomato processor, has run since 1990 with no manager positions, on two written rules: commitments get honoured, interactions stay voluntary.

How to settle it for your team

The question works decision by decision rather than company-wide. List the ten decisions that genuinely matter where you are: hiring, setting a price, shipping a feature, committing spend, changing a process, killing a project. For each one, write down who settles it today, then who should.

Three criteria steer each line.

The cost of a mistake weighs heaviest. A reversible, cheap mistake belongs as close to the ground as possible, an irreversible one earns a filter.

The speed of change comes next. A price reviewed once a year handles central sign-off perfectly well. A product call that comes up three times a week gets expensive in waiting time.

Then there is the question of who holds the information. When the person facing the customer knows things the exec team does not, sending the decision upward means deciding with less information.

Most organizations that work end up mixed. They stay vertical on safety, compliance and capital allocation, and go horizontal on product, customer service and how daily work gets organized. That mix survives only when it is written down somewhere, otherwise everyone runs their own version of it.

Writing down who decides what

Rolebase is where that list lives. Every role on the org chart carries its purpose, its decision domain and its accountabilities, which turns "who settles this?" into something you look up instead of something you negotiate. Decisions stay attached to the role that made them, with their date and their context.

A governance mode sets who may change the structure itself. Three settings exist, from Free mode where anyone edits anything to Strict mode where every structural change goes through a proposal decided together. Meetings, tasks and discussions live in the same place as the roles, so nobody rebuilds the context from scratch each time.

See the features, or create your organization free for up to five active members.

What is the difference between horizontal and vertical management?

Under vertical management, authority follows the reporting line. A decision that exceeds one job's scope goes up a notch until it reaches someone entitled to settle it. Under horizontal management, authority follows the role. Whoever holds a role decides within its domain and answers to the roles concerned. How many levels the org chart has is a separate question.

Is vertical management ever the better choice?

Yes, in three situations. It wins when a call has to be made in seconds and a mistake costs lives, as in the trauma teams studied by Klein, Ziegert, Knight and Xiao in 2006. It wins when a bad project approved costs far more than a good project killed, which Sah and Stiglitz formalized in 1986. It wins when expertise is very unevenly spread between the person who spots the problem and the person who can handle it.

Is flattening the org chart enough to go horizontal?

No. Guadalupe, Li and Wulf showed in 2012 that large US firms had cut levels while pulling functional decisions closer to the CEO. A flatter org chart can come with more centralized decisions. What changes the model is moving the decision rights.

Why do some horizontal organizations fail?

Because they remove titles without writing down who decides what. Power then re-forms around seniority and closeness to the founders, with no way for anyone to challenge it. Jo Freeman described that mechanism in 1972 in The Tyranny of Structurelessness, and former Valve employees have reported the same dynamic since 2013.