Flatter alone is not healthier.
At a glance
- Challenge
- A 90,000-person company had eleven layers between its chief executive and its front line, some sixteen thousand leaders, and about one in eight of them leading a team of one. It did not feel like a structure problem. It felt like slow decisions.
- Andrew’s role
- Group Head of Organisational Capability and Transformation. I created the Space to Lead concept and mobilised my team to design it and run it with the executive committee, from the top down.
- Scale / context
- The whole enterprise: every function, every geography, and a set of rules that had to work for a research laboratory and a manufacturing site at the same time.
- What changed
- Fewer, better-supported leaders; spans and layers held to published rules rather than to a target; and the measures handed to the line on the enterprise dashboard rather than kept by a programme.
Every large company eventually reaches for the same instrument: take out layers, count what it saved, call the result flatter. I created Space to Lead to get further than that, by writing rules that differentiated by the nature of the work rather than by the shape of the chart, and then showing every leader in the company exactly where they stood against them.
The blunt instrument
Every few years the same tool comes off the shelf, and this time it has a name. The Great Flattening has household names cutting management layers by a third, removing thousands of management roles at a stroke and rebuilding around self-directed teams, and AI has handed the reflex a fresh justification. It is asking the managers who remain to carry far more people.
The spans bear that out. Gallup’s span-of-control research, published in January 2026, puts the average manager’s team at 12.1 people, up from 10.9 a year earlier and close to half again as large as when it started measuring in 2013.
Cutting layers is easy to announce, easy to count and easy to do badly. It removes cost without removing work, so the work lands on whoever is still there, and the organisation discovers a year later that it has the same problems and fewer people to solve them. What actually correlates with performance is organisational health: McKinsey’s index puts top-quartile organisations at around three times the total shareholder returns of bottom-quartile ones. Flatter is not the same as healthier. It is not reliably even a step towards it.
What the organisation actually felt
Nobody complains about a layer count. They complain about waiting.
Eleven layers between the chief executive and the front line, and roughly one in five people sitting nine or more layers down. Sixteen thousand leaders, about one in eight of them leading a team of one. A team of one is not a team. It is a reporting line with a job title attached, and it costs the organisation a layer, a set of meetings, a performance conversation and a career path that leads nowhere in particular.
What that produced was not an org chart complaint. It was slow decisions, unclear decision rights, and an administrative load that fell hardest on the people closest to the work. That is the thing worth fixing, and a layer count is only the symptom you can see from the outside.
Rules, not a mandate
The temptation at that point is to issue a number. Every leader will have at least seven people. It is simple, it is auditable, and it is wrong, because it assumes that all work is the same work.
So we wrote golden rules instead. A ceiling on layers: eight, or nine where operations are people-intensive. An enterprise ambition for spans of about six and a quarter. A floor beneath which a team is not a team. And, most importantly, a deliberate range around that ambition set by the nature of the work rather than by the shape of the chart.
That last part is the whole idea. Where work is complex, interdependent and knowledge-heavy, in research or in legal, a leader spends real time interlocking their people’s work, and a small span is correct. Where work is well defined and the person doing it is largely self-sufficient, on a sales team or a manufacturing site, a leader can carry many more, and a small span is waste. Spans across the company ranged from a handful to more than twenty, and they were supposed to. A single number applied to both ends damages the knowledge-heavy half of the business in order to flatter the average.
Rules travel where a mandate does not. A mandate has to be enforced by whoever wrote it. A rule can be handed to a designer, written into a playbook, and used by someone who has never met you.
Rules
- A ceiling of eight layers, or nine where operations are people-intensive.
- A floor, below which a team is not a team.
Spans
An enterprise ambition of about six and a quarter, as an average across all three bands
A small span
A leader spends real time interlocking their people’s work, so fewer of them is correct.
A middle band
Some of the work interlocks and some of it does not; the ambition sits here.
A large span
The work is well defined and largely self-sufficient, so a small span is waste.
Complex, interdependent, knowledge-heavy work
Well-defined work, where the person doing it is largely self-sufficient
Then hold up the mirror
A rule nobody can see themselves against is a suggestion, so two things ran alongside the rules.
The first was saying plainly which was which. Some of the golden rules were guidelines, to be argued with where the work justified it. Others were firm. Blurring those two is how a rulebook loses its authority: if everything is negotiable then nothing is, and if nothing is negotiable you have written a mandate after all.
The second was making every leader’s own position visible. The measures went to the executive committee on a regular cycle, and they also went into plai, the company’s own decision-intelligence app, where any leader could look up their own spans, their own layers, their own teams of one, their own same-grade reporting lines. The hallmarks of organisational health, for their own organisation, on demand.
That changes who does the work. Nobody had to be told to act. A leader could see where they stood against something they had already agreed to; their peers could see the same thing; and so could the chief executive. Sometimes the job is to be Switzerland, report the news accurately, and let other people act on it.
That belief is wider than this programme. Data about structure, talent or culture tells its own story, and shared transparently across a peer group it moves more than any instruction does. You are not telling people what to do. You are showing them where they are against what they said they wanted.
Step change, or every time somebody leaves
There are only two ways a large organisation ever changes shape, and Space to Lead used both.
The first is step change. When a business is being redesigned anyway, the rules become the guardrails of the design: you are drawing a new organisation, so you draw it inside the ceiling and the spans from the first sketch rather than auditing it afterwards. Sanofi’s manufacturing and supply organisation ran it that way as a standalone exercise, and the enterprise transformation that followed carried the rules into its own blueprints and playbooks.
The second is organic, and it is the one most companies dismiss. Every time somebody leaves, a question gets asked before the requisition goes up: does this role still exist in this form, and could these two small teams sit under one leader rather than two? On its own that sounds glacial.
It is not. At an average tenure of around three years, an organisation turns over the equivalent of its entire population every three years. You do not need a transformation to reshape an organisation. You need to stop rebuilding the old one every time somebody resigns.
What it delivered
The pilots moved first. Spans widened towards the ambition, layers came down, engagement improved, and the large majority of leaders reported that decisions were getting faster, which was the point rather than a side effect.
Within a year of the rules being set, a substantial part of the enterprise had been assessed and was implementing. The deepest layers were materially thinner. Around a hundred and eighty people had been trained to run the assessment themselves, which mattered more than the number suggests: it meant the next redesign did not have to queue for a central team. That is the whole arc in one programme: diagnose the shape the organisation is actually in, mobilise against rules rather than a target, then sustain it by handing the measures to the people who own the work.
Then it stopped being a programme, which is the outcome I care about most. The measures went onto the enterprise people dashboard, where leaders are held to them alongside everything else they are held to. The rules were written into the design playbooks of the transformation that followed. The thing carried on working after the programme that made it had gone.
What I learned
Guiding principles
Flatter alone is not healthier. A layer count is a symptom you can see from outside, not a diagnosis, and taking out structure without taking out work moves the load rather than the problem. What makes an organisation faster is fewer but better managers: fewer layers, and alongside them leaders with the skill, the span and the decision rights to actually decide.
Write golden rules, not a mandate. A single span target is auditable and wrong. Differentiate by the nature of the work, say plainly which rules are advice and which are firm, and then publish where everyone stands against them. You are not telling people what to do; you are showing them where they are against what they agreed to, and common accountability moves further than instruction.
A team of one is not a team. It is a reporting line with a title attached, and it costs a layer, a meeting series, and a career path that leads nowhere in particular.
Incremental health does not always need a transformation. At three years’ average tenure an organisation turns over the equivalent of its whole population every three years. The moment somebody resigns is the cheapest design lever anyone has, and almost nobody uses it.
Hand the measures to the line, then keep holding the mirror. A programme that keeps its own scorecard keeps it forever, so put the measures on the dashboard the business already reads. But handing them over is not walking away. The reporting has to keep coming, accurate and public, or the organisation quietly returns to the shape it had.
Why it matters now
AI is creating the next wave of flattening, and a great deal of it will be done with the same blunt instrument, this time with the technology as the excuse. The reasoning will sound like productivity and arrive as a layer count.
The more interesting move is underneath it. AI is arriving inside the manager’s own job: the approvals, the paperwork, the reporting, the chasing, the parts of a manager’s week that consume the time without needing the person. Take that friction out of the manager’s journey and you have not made a case for fewer managers. You have made room for the part of the job only a person can do: judgement, coaching, setting direction, building a team that works, and noticing what is happening to engagement and morale before a survey tells you.
So the question was never how many managers there are. It is whether the ones you keep have the room, the span, the decision rights and the time to lead. That is a design question, and it is the same design question whatever is driving it.