Reorganisations promise that the boxes and lines on a chart will fix a problem that is usually about something else, strategy, trust, or who gets to decide. They rarely do. But the shape of an organisation is a real lever: it decides what is easy and what is a fight, who talks to whom by default, and which goals quietly win when two of them collide. Knowing the four basic shapes, and what each one costs, is how you stop reorganising by instinct and start choosing on purpose.

The quick version

  • Functional groups people by what they do (engineering, sales, finance). It builds deep expertise and is efficient, but slows down when work has to cross those silos.
  • Divisional groups people by what they serve (a product, region, or customer type). Each division owns its result, at the cost of duplicating functions and losing scale.
  • Matrix overlays two of those at once, so people report to a function and a product/project. It shares scarce experts across work, but gives everyone two bosses, and has well-documented failure modes.
  • Network keeps a small core and coordinates the rest through partners, contractors and alliances. It is flexible and light, but trades control and institutional memory for that flexibility.

The idea in depth: structure follows strategy

The foundational point is older than most management fashions. Business historian Alfred Chandler, studying how DuPont, General Motors, Standard Oil and Sears grew through the early twentieth century in Strategy and Structure (1962), drew a conclusion now compressed to a slogan: structure follows strategy. When these firms diversified their products and markets, the old centralised, function-by-function setup buckled, so they invented the multi-divisional (M-form) structure, breaking the company into semi-autonomous units each running its own business, steered from the centre by financial targets.

The lesson is not that the divisional form is best. It is that the right structure depends on what you are trying to do. A company competing on operational efficiency in one product line wants different wiring from one juggling a dozen businesses across thirty countries. So the move is to start every structure conversation from strategy, not from the org chart: what must this organisation be good at, and which shape makes that the path of least resistance? If you cannot name the strategic reason for a reorg, you are rearranging furniture.

Henry Mintzberg sharpened this into a useful catalogue. In his work on organisational configurations, he argued there is no single best structure, each viable shape is a coherent bundle of choices (who holds power, how formal the rules are, what skills dominate) that fits particular conditions: how stable the environment is, how big the organisation is, how much the work depends on coordination, and how fast it must innovate. Get the fit wrong and you get predictable failure: a rigid bureaucracy in a fast-moving market is too slow; a free-form structure at scale, with no guardrails, descends into chaos. The structure is not good or bad in the abstract; it is well-fitted or badly fitted to the job.

flowchart TD
  S(["Strategy: what must we
be best at?"]) --> Q{"What varies most,
the work, or the market?"}
  Q -->|"One product, scale &
deep craft matter"| F(["Functional
group by discipline"])
  Q -->|"Many products/regions,
each needs its own focus"| D(["Divisional
group by output"])
  Q -->|"Both at once, experts
are scarce & shared"| M(["Matrix
two-axis reporting"])
  Q -->|"Speed & flexibility beat
owning everything"| N(["Network
small core + partners"])
Structure follows strategy: the dominant source of variation points you toward a shape. Leaders Loop

The four shapes, and what each one costs

Functional. Everyone who does the same kind of work sits together: all the engineers under engineering, all the marketers under marketing. This is the default for younger and single-product companies, and for good reason, it concentrates expertise, makes career paths and standards clear, and avoids duplicating roles. Its cost shows up the moment a goal needs several functions to move together: a product launch becomes a relay race across silos, each optimising its own leg, with no one owning the finish line. The functional form is efficient at doing things and clumsy at delivering outcomes.

Divisional. Here the company splits into units built around an output, a product line, a customer segment, a geography, and each division contains the functions it needs. This is Chandler's M-form. Its great virtue is accountability: a division has its own P&L and an obvious owner, so it can move fast and stay close to its market. The bill arrives as duplication (five divisions, five marketing teams) and lost scale, plus the risk that divisions compete instead of cooperate and that good ideas never cross from one to another. So the move, if you go divisional, is to be deliberate about what stays shared at the centre, and to expect to pay for the rest.

Every org structure buys you one strength and bills you for it somewhere else. The question is never "which is best" but "which bill can we afford".

Matrix. A matrix refuses to choose between function and output, and runs both at once: an engineer reports up a functional line (to the head of engineering, for standards and career) and across to a product or project lead (for what to build this quarter). The appeal is real, it lets a company share scarce specialists across many efforts and balance two priorities that a single-axis structure would force you to rank. It is also the structure most likely to go wrong, which is the subject of the next section.

Network. The network form keeps a lean core and coordinates much of the actual work through outside parties, contractors, suppliers, alliance partners, platforms. The sociologist Walter Powell described this in "Neither Market nor Hierarchy" (1990) as a genuinely distinct way of organising, not a loose market of one-off transactions, and not a top-down hierarchy, but a web held together by relationships, reciprocity and trust, especially where knowledge is tacit and learning matters more than control. The upside is flexibility and focus: you own what you are best at and rent the rest. The cost is exactly the control and memory you gave away, coordination across organisational boundaries is hard, and capability you outsource is capability you no longer build.

Where the matrix breaks (and how to use it anyway)

The matrix deserves its own section because it is both the most tempting structure for a complex business and the most reliably mishandled. The definitive warning is old and still accurate: Stanley Davis and Paul Lawrence, in "Problems of Matrix Organizations" (Harvard Business Review, 1978), catalogued the pathologies the form is prone to, among them power struggles (the two bosses fight for control rather than balance it), anarchy and the mistaken belief that matrix means decision-by-committee, groupitis (confusing matrix management with endless group meetings), decision strangulation (everything needs two sign-offs, so nothing moves), and collapse during a downturn, when the dual structure is the first thing a frightened company tears out.

flowchart TB
  CEO(["Leadership"]) --> EngF(["Head of Engineering
(functional axis)"])
  CEO --> ProdA(["Product A lead
(output axis)"])
  CEO --> ProdB(["Product B lead
(output axis)"])
  EngF --> E1(["Engineer 1"])
  EngF --> E2(["Engineer 2"])
  ProdA -.-> E1
  ProdB -.-> E2
The matrix's defining feature, and its risk: each person answers to a functional line (solid) and a product line (dotted). Leaders Loop

So the move, if you adopt a matrix, is to treat its failure modes as the design brief rather than as surprises. The recurring fix in the literature is to refuse the fiction that a matrix means two equal bosses for every decision: make decision rights explicit, name, in advance, who decides what, and where one axis leads and the other advises. Davis and Lawrence's own framing is that a matrix is as much a state of mind as a chart; it works only when senior leaders genuinely hold the tension between the two axes instead of letting it harden into a turf war beneath them.

An honest limitation. These are tendencies, not laws. Plenty of matrix organisations run well, and plenty of "clean" functional or divisional ones are dysfunctional for reasons that have nothing to do with the chart. Structure shapes behaviour; it does not determine it. A good structure with bad managers and murky decision rights will still fail, and a clear, trusted leadership team can make a theoretically awkward structure work. Treat any structure as a lens for asking "what will this make easy, and what will it make a fight?", not as a guarantee of either.

A worked example

Take a software company, call it Meridian, that has grown from one product to three, now selling in both Europe and Asia. (Illustrative throughout; this is a teaching example, not a real company.) It started, sensibly, functional: one engineering team, one sales team, one design team. That worked beautifully at one product. At three, it jams: the single engineering queue means every product line waits behind the others, and no one person can be held to a product's number because the people who build it answer to a function, not to the product.

Start from strategy. Meridian's edge now is moving fast in three distinct markets, so the dominant source of variation is the output, not the craft. That argues for a divisional shape: three product units, each with its own engineering, design and sales, each owning its P&L. The cost is plain and worth naming aloud, three smaller engineering teams instead of one big one, some duplicated tooling, and the danger that the units stop sharing what they learn. Meridian decides that the accountability is worth the duplication, and keeps a few things deliberately central (security, brand, the data platform) so it does not pay the full duplication bill.

flowchart LR
  A(["1 product:
functional works"]) --> B(["3 products, 2 regions:
functional queue jams"])
  B --> C{"What varies most
now, craft or output?"}
  C -->|"Output: 3 distinct markets"| D(["Go divisional:
3 product units, own P&L"])
  D --> E(["Keep security, brand &
data platform central"])
Meridian's path: the structure that fit at one product stopped fitting at three, so strategy, not preference, picked the next shape. Leaders Loop

Notice what Meridian did not do: jump straight to a matrix because it sounds sophisticated. A matrix would have made sense only if its specialists were so scarce that sharing them across products beat giving each product its own, and even then, only with explicit decision rights to avoid the pathologies above. The structure changed because the strategy did. That is the whole discipline.

Frequently asked questions

Which org structure is best?

None of them, in the abstract, and any source that tells you otherwise is selling something. The right structure depends on your strategy and conditions: how many distinct products or markets you serve, how scarce your key skills are, how fast the environment moves, and how much the work depends on coordination. Chandler's point holds: decide the strategy first, then choose the shape that makes it easiest to execute.

What's the difference between functional and divisional?

Functional groups people by the work they do (all engineers together); divisional groups them by the output they serve (everything for Product A together, including its own engineers). Functional builds deeper expertise and avoids duplication but struggles to deliver cross-cutting outcomes; divisional gives clear ownership and speed per unit but duplicates functions and can lose scale and shared learning.

Why do matrix organisations get a bad reputation?

Because the thing that makes a matrix powerful, two reporting lines, is also what makes it fail. Davis and Lawrence (1978) catalogued the pattern: power struggles between the two bosses, decision strangulation when everything needs two sign-offs, and a tendency to collapse under cost pressure. Matrices can work, but only when leaders hold the tension deliberately and decision rights are spelled out, rather than left for people to fight over.

Is a "flat" or "network" structure better for startups?

Often, early on, a small core that buys in specialist work stays light and flexible, which suits a company still finding its market. But the network form trades control and institutional memory for that flexibility (Powell, 1990): capability you outsource is capability you do not build, and coordinating across organisational boundaries is genuinely hard. It is a deliberate trade, not a free upgrade.

How do I know it's time to restructure?

When the current shape makes your strategic priority a constant fight, when outcomes that should be straightforward require heroics to cross silos, or no one can be held accountable for a result because the people who deliver it report elsewhere. The signal is recurring friction on the work that matters most, not a quiet preference for tidier boxes. And remember a reorg is expensive and disruptive; be sure the problem is structural before you reach for a structural fix.

Related in the Toolkit

An org structure is only the skeleton; how the work actually flows through it is set by your operating model, and the pathologies of the matrix are really failures of unclear decision rights, which is why those two pieces matter as much as the chart itself.

Where to go next