How to transition to horizontal management

Middle managers made up 29% of US layoffs in 2024, against roughly 20% a year between 2018 and 2022, according to job tracker Live Data Technologies. A good share of those cuts were announced as a move toward flatter, faster, more autonomous teams. Most were cost reductions wearing a nicer word, and everyone in the affected teams worked that out within a week.

Genuine horizontal transitions look different, and they are harder. A manager does work you barely see while they are still there. They settle trade-offs, they order priorities, they carry information between floors. Deleting the position leaves all of that work intact. When nobody picks it up on purpose, it goes to whoever talks loudest, and the organization becomes harder to read than it was before.

Every horizontal organization that lasts has replaced its management line with something written: named roles, one decision rule per kind of question, and a record of how the structure changed. This article covers that swap, with the cases that document it and the ones that failed.

What "horizontal" actually means

A horizontal organization cuts the number of levels between the person doing the work and the person deciding. It keeps authority and slices it differently. In a pyramid, authority sits on a position and covers everything below it. In a horizontal organization, authority sits on a role and covers one written domain.

Take three ordinary decisions: switching coffee suppliers, publishing a press release, hiring an intern. With a management line, each one travels upward until it finds the level allowed to make it. With roles, each one already belongs to someone who makes it alone.

	Management line	Horizontal organization
Authority	Sits on a position, covers everything below	Sits on a role, bounded by a written domain
Information	Travels up and down, filtered at each floor	Goes straight to whoever needs it
Everyday decision	Signed off one level up	Made by the role that owns the domain
Decision spanning several roles	Settled by the shared manager	Handled by a named rule, consent or majority
Structural change	Annual reorg	Regular governance meeting
What it costs	Management salaries, approval delays	Meeting time and the work of writing roles

The horizontal versus vertical question rarely gets settled in one go. Most organizations keep a management line over legal, safety and payroll, and open up the rest.

Three questions before removing a single layer

A culture audit costs a lot and tends to confirm whatever the person who commissioned it already believed. Three measurements tell you where you stand, and they double as the baseline you compare against a year later.

How long between a problem being raised and the decision that handles it? Take the last ten issues a team escalated. Date the moment someone said this is broken, then the moment something changed. The median gives you the real pace of the organization, whatever the org chart claims.

Who actually decides? Ask five people who signed off the last expense over 5,000 euros and the last hire. Where the answers diverge is exactly where the official org chart stopped describing the work.

What happens when a manager is away for three weeks? If the team keeps going and nothing piles up, that manager's work was already distributed. If everything waits for their return, the transition starts with whatever they hold alone.

Those three numbers put the problems a heavy hierarchy creates where they actually hurt rather than where people assume they are.

What the organizations that last actually do

Buurtzorg, twelve nurses and no boss

Jos de Blok founded Buurtzorg in 2006 with four nurses. The Dutch home care organization now runs more than 10,000 caregivers in over 750 teams, with no middle managers at all. Each team is ten to twelve nurses covering around fifty patients in one neighbourhood. The team hires, schedules, holds its budget and picks its own premises.

Head office is about fifty administrators. What replaces middle management is eighteen regional coaches for the whole network. A coach decides nothing and shows up when a team asks.

An Ernst & Young study published in 2009 found that Buurtzorg uses roughly 40% fewer care hours per client than other Dutch home care providers, with patients staying in care about half as long.

Morning Star, a colleague contract instead of a job description

Morning Star processes tomatoes in California, employs around 400 permanent colleagues and 2,400 seasonal ones, and runs with no titles, no promotions and no managers. Gary Hamel described how it works in the December 2011 Harvard Business Review, under the title "First, Let's Fire All the Managers".

The central mechanism is the CLOU, short for Colleague Letter of Understanding. Every year each person writes their personal mission, the accountabilities they take on, and the metrics they agree to be judged by. They then negotiate that document with the colleagues their work directly affects, which takes hours of back and forth. A colleague can commit spending with no sign-off, provided they consulted the people concerned.

A written commitment replaces an instruction received. It is dated, signed and revisited every year.

Bayer, the same idea across 100,000 people

Buurtzorg and Morning Star were born flat. Bayer is converting a 160-year-old group. The programme launched in July 2023 brings management layers down from thirteen to six or seven depending on the division, and organizes work in teams of six to ten people running on 90-day cycles.

The public figures give the scale. The group cut around 11,000 positions between 2023 and early 2025, close to 7,000 of them managerial according to its 2024 annual report, and targets 2 billion euros of annual savings by the end of 2026. Over the same stretch the share price lost more than 40% of its value in 2024, largely for reasons that predate the reorg, starting with the glyphosate litigation inherited from Monsanto.

A plan that removes 7,000 management positions does look a lot like a cost plan. It is also the best documented large-scale horizontal transition anyone has, and it will take another two or three years to know what it does to decision speed.

What makes a transition fail

The catalogue of failures teaches more than the catalogue of successes, because it repeats.

Adopting a complete system in one go gets expensive fast. Zappos adopted holacracy in January 2014. In March 2015, Tony Hsieh offered severance to anyone who preferred not to follow, and 18% of staff left. The company gradually brought managers back afterwards. Medium dropped the same model in March 2016, and Andy Doyle wrote in his public post that the system was levying "a small but persistent tax" on the team's effectiveness and on its sense of connection.

Removing the official structure grows an unofficial one. Jo Freeman described this in 1972 in The Tyranny of Structurelessness: the informal structure runs on seniority, friendship and comfort speaking up, it appears in no document, and nobody can contest it. The piece on the liberated company traces the same mechanism through French research.

A transparency ritual closes as soon as its conditions change. Google gathered staff every Friday for TGIF, where anyone could question leadership. In November 2019, Sundar Pichai announced a shift to monthly and a refocus on strategy, after repeated leaks. Twenty years of open weekly questions closed in one memo.

Dressing a headcount reduction as autonomy costs trust for years. A team that watches its manager leave, absorbs the workload and gains none of the matching decision rights draws the obvious conclusion, and it is right.

Trust comes out of the structure

In 1996, Amy Edmondson studied nursing units in two American teaching hospitals, starting from an ordinary hypothesis: teams that work well make fewer medication errors. The data said the opposite. The best-rated units reported the most errors, because they were saying so instead of covering up. The paper came out in the Journal of Applied Behavioral Science and produced the concept of psychological safety, which Edmondson formalised three years later from 51 teams in a manufacturing company. Her definition fits in one line: the shared belief that you can take an interpersonal risk without being punished or humiliated.

Project Aristotle points the same way and gets quoted badly. Google analysed 180 internal teams, 115 in engineering and 65 in sales, and published five factors in 2015. Psychological safety comes first, dependability second, then structure and clarity, which Google defines as each person's understanding of what is expected of them and of the process for meeting it. Three of those five factors describe how the work is organised rather than how people feel about each other. Of course the study reports correlations without settling which way the arrow runs, and a team that succeeds may well manufacture its psychological safety along the way. It remains the largest public sample on the question.

So people extend trust on evidence. I know who took this decision, I know the scope that person decides alone on, I know where I could have objected. Those three pieces of information get written before autonomy is handed out, and they outlive whoever set them up. Asking for trust first, as Isaac Getz does when he makes prior trust the starting point of the liberated company, leaves a gap, and the informal structure Jo Freeman described moves straight into it.

Thierry Weil and Anne-Sophie Dubey measured how far the autonomy actually granted goes, in Au-delà de l'entreprise libérée, published in 2020 by La Fabrique de l'industrie across ten organisations of 50 to 1,300 employees. The autonomy teams obtain almost always covers how the work gets done, rarely the objectives, and almost never governance itself. Teams work it out within weeks, and the distance between the speech and the real scope costs more than saying nothing would have.

Of course visibility can turn against the team. James Barker followed an American manufacturer that had replaced its supervisors with self-managing teams, and his 1993 Administrative Science Quarterly article, "Tightening the Iron Cage", describes what the teams built instead: value-based normative rules, enforced on one another with a rigour the old supervisors never reached. Weil and Dubey found the same drift in France. Peer control is harder to challenge than a boss, because it has no name and no procedure applies to it.

The dividing line holds up well in practice. Make the roles and their domains visible, along with governance decisions and their reasoning, the minutes, and the history of org chart changes. Keep individual response times, who read what and per-person dashboards out of view. The first list makes power contestable, the second moves it to the group without making it any easier to contest.

The sequence that works

1. Write the roles before you remove the positions. A role carries a purpose, a domain its holder decides on alone, and accountabilities the others expect. One person holds several. The exercise surfaces the accountabilities nobody owns and the topics three people each believe they run. See how to define roles and how a role differs from a job description.

2. Name what stays closed. Safety, client commitments, payroll and legal obligations stay decided at the top in most organizations. Saying so plainly is what makes everything you open next credible.

3. Set one decision rule per kind of question. A decision inside a domain gets made alone. A decision touching several roles goes through consent decision-making, which asks for the absence of a reasoned objection rather than everyone's enthusiasm. Structural choices keep an advice process with the people affected.

4. Open a monthly governance meeting. That is where roles get created, amended and retired. The governance meeting handles the structure, the operational meeting handles the work in flight. Without that space, the org chart freezes within three months and the gap with real work widens.

5. Keep the record of changes. A horizontal organization moves often. With no history, nobody can find who created a role, when, or off the back of which decision, and the structure becomes unreadable to a newcomer within a year.

6. Start with one team. Research from La Fabrique de l'industrie puts the working size for self-organization between 5 and 40 people. A pilot team gives you numbers to compare against the three baseline measurements, and an internal story that convinces better than a leadership memo.

7. Support the middle managers first. They lose the most footing, and they decide whether the transformation survives. A manager whose position is removed without another one offered will defend the old one, which is rational.

8. Treat every role as a hypothesis. A role written on the day of a governance meeting describes what the team believes is useful that day, and it is often wrong on the first try. The Plan-Do-Check-Act cycle, formalised by Walter Shewhart and spread by W. Edwards Deming, says what to do about it: hold the role for a quarter, look at what moved on the baseline measurements, then amend it or retire it at the next meeting. Bayer follows the same logic with its 90-day cycles.

That short loop is the difference between a written structure and a living one. An organization that waits for its annual reorg to fix a badly cut scope accumulates twelve months of workarounds, and those workarounds end up being the real structure. The small regular adjustments of the Toyota Production System work here for the same reason they work on a shop floor: one change at a time can be measured, ten at once cancel each other out.

The checklist for setting up a horizontal structure turns this sequence into steps you can tick off.

Tools

A tool creates no practice. It makes visible what the organization already decided, and it keeps the structure out of a spreadsheet that three people maintain by hand.

Four needs come up during every transition: seeing who holds which role without asking anyone, deciding together with a trace left behind, running meetings whose agenda follows from the roles in the room, and finding out why the structure changed.

Rolebase covers those four. The org chart is interactive, roles nest and move by drag and drop, and every role carries its purpose, domain, accountabilities, checklist and indicators. Proposals are voted by consent, unanimity or majority, and an accepted proposal applies the org chart changes it carried. Every edit lands in a history you can undo. Three governance modes set who may change what, from open editing to mandatory approval through a proposal. The product is open source, free up to five active members, then 5 € per user per month.

Rolebase holds the roles, meetings, decisions and role-bound tasks, and syncs with Google Calendar and Office 365. Team chat stays in the tool you already use. Visibility there covers the structure and the decisions, and individual activity stays out of it. The review of tools for horizontal teams places each category on its own ground.

The worker cooperative EVEA shows the need clearly. It grew from 60 to 140 people between 2020 and 2023 across three sites, with more than eight competency hubs. Shared governance was already in place. What was missing was a map anyone could read in real time, once a hand-maintained document stopped keeping up.

Knowing whether it is working

What you measure	How	Cadence
Lag between a problem raised and the decision that handles it	Median over the last ten issues	Quarterly
Decisions escalated for lack of a clear mandate	Counted in the governance meeting	Monthly
Roles vacant or held by default	Org chart review	Monthly
Roles created or amended	Org chart review	Quarterly
Reasoned objections raised in governance meetings	Count	Quarterly
Workload reported by former managers	One-to-one interview	Quarterly
Voluntary departures	Usual HR tracking	Twice a year

The count of amended roles and the count of objections read together. Zero objections over two quarters signals silence rather than agreement, and a structure that has stopped moving measures the gap widening with the real work.

Three of those are enough for the first year. A transition that produces dashboards before it produces fast decisions has picked the wrong target.

Frequently asked questions

How long does a transition to horizontal management take?

Budget six months for a pilot team of 5 to 40 people, long enough to write the roles, set a decision rule and hold a few governance meetings. At group scale the unit is years: Bayer started in July 2023 and was still rolling the model out at the end of 2025. The efforts that flip a whole organization in a few weeks are also the ones that reverse fastest.

Do you have to remove all the managers?

Most horizontal organizations keep a management line over legal, safety, payroll and client commitments. What changes is what the position is for. At Buurtzorg, middle management was replaced by regional coaches who follow forty to fifty teams with no decision power. At Bayer, layers went from thirteen to six or seven rather than to zero.

Does horizontal management work in a large company?

Buurtzorg passed 10,000 caregivers with autonomous teams of twelve and a head office of about fifty, which suggests total size matters less than unit size. The difficulty shows up in coordination between units, and that is exactly what Medium cited in 2016 when it dropped holacracy. A large horizontal organization works as a network of small teams explicitly linked, rather than as one big team.

What is the difference between horizontal management, holacracy and sociocracy?

Horizontal management describes an intention, cutting layers and moving decisions closer to the work. Sociocracy and holacracy offer precise mechanisms to get there: written roles, governance meetings, consent decision-making, double linking between teams. You can adopt one of them whole or borrow the parts that help.

What happens to middle managers?

They hold several roles instead of one position, usually around expertise, facilitation or client relationships. They are the group that loses the most footing during a transition, and the group that gets trained the least. In the US, middle managers accounted for 29% of 2024 layoffs according to Live Data Technologies, against roughly 20% a year before that, which explains how much suspicion the word "horizontal" now carries inside teams.

Does trust come before or after autonomy?

The two reinforce each other, and the practical order runs from structure to trust. A team trusts a decision when it knows who took it, the scope that person decided alone on, and where an objection could have landed. Those three pieces of information get written before autonomy is handed out. Asking for trust first, with nothing written, leaves a gap that informal power occupies, as Jo Freeman described in 1972.

Can transparency harm a team?

It harms when it targets people instead of decisions. James Barker described in 1993, in Administrative Science Quarterly, a manufacturer where self-managing teams produced a system of norms enforced between peers more rigorously than the supervision they replaced. Weil and Dubey observe the same drift in France. The workable rule is to make roles, decisions and minutes public, and to leave response times, activity and per-person dashboards out of view.

What it really takes

A horizontal transition asks for less conviction and more writing than most people expect. Buurtzorg has held for twenty years with teams of twelve and a head office of about fifty. Morning Star has held longer still, on a contract renegotiated between colleagues every year. Neither runs on trust alone. Both run on documents anyone can read.

Start with one team, write its roles, set its decision rule, measure decision lag before and after. The rest of the organization will follow or wait, and you will have at least one number to put against the objections.

See what Rolebase does, or create your organization for free up to five active members.