Org design for a 20-to-50 person company

Some scale context, because most organisational-design writing is calibrated for companies an order of magnitude larger. In the Census Bureau's 2022 employer-firm tables there were 6,395,635 employer firms in total, of which roughly 1.31 million sat in the entire ten-to-499 employee range and 656,917 in the ten-to-nineteen band. Those figures were released in April 2025 and the band splits between them are computed from the published table rather than published directly, which is worth stating whenever they are used. The point is that a twenty-to-fifty person company is not a small version of a large one; it is the modal serious business, and it is under-served by the literature.

The first structural crisis in this band is the first manager layer. Before it, everyone reports to the owner and coordination happens by proximity. After it, there is a person between the owner and the work, and the owner's job changes from doing to deciding. Practitioners locate this transition somewhere in the ten-to-twenty-five headcount range; that placement comes from practice rather than from a source we can cite, and it is offered as observation rather than as a finding.

The symptoms are more reliable than the headcount. You are past the inflection when routine approvals queue at your desk, when two people give a customer different answers because there is no owner of that answer, when new employees are onboarded by whoever is free, and when you cannot take a week off without deferring decisions rather than delegating them. Those are observable in your own week, which is why this page describes symptoms instead of naming a number.

An organisational chart shows people and reporting lines. An accountability chart shows seats and the functions each seat owns. They are different objects and the second is the one that does work in a company this size, because the interesting question is not who reports to whom but which functions currently have no owner. Draw the seats the business must fill — sales, delivery or operations, finance, people, and whatever your model demands — and write two or three accountabilities under each before putting a name anywhere.

Doing it in that order surfaces the two conditions that actually cause trouble: a function with no owner, and a person occupying three seats that pull in different directions. Both are invisible on a conventional chart and obvious on an accountability chart, which is why it is worth drawing even by operators with no interest in the framework it comes from.

For structure, a functional shape is right for almost every business in this band. Divisional shapes make sense once you genuinely run separate businesses — different customers, different economics, different physical locations with their own profit and loss. A matrix below a hundred people is usually a mistake: it doubles the coordination cost of a company that does not yet have the management capacity to pay it once.

Titles in this band are unreliable and collide constantly, particularly in the trades, where operations manager, service manager and dispatcher describe overlapping work in different companies. Define the seat by its accountabilities rather than by its title, and expect to have to explain the distinction to your own team, because they inherited the old titles too.

Span of control — how many people report to one manager — is real, consequential, and surrounded by numbers that cannot be sourced. The most widely repeated figure in management writing is attributed to a specific article whose citation now returns a 404, and the large-consultancy archetype tables that would be the modern substitute were not retrievable in this pass. The figure may well be right; it is not verifiable, so it does not appear here presented as evidence.

What is publishable is the dependence. Span is a function of work variability, not of management theory: a supervisor overseeing highly standardised, co-located work can carry many more people than one overseeing dispersed, variable, judgement-heavy work. That means the honest answer to 'how many reports is too many' is derived from your own work, not looked up. Watch for the observable failure signs instead — a manager who never has time for one-to-ones, escalations that skip a layer as a matter of routine, and quality that varies by who happened to supervise the job.

The de-layering method from large-organisation practice does transfer, even where its reference tables do not. Map every manager, count their direct reports and the layers between the front line and you, and look for layers that exist to carry one person or to preserve a title. In a fifty-person company more than three layers between the work and the owner is almost always one too many.

Underneath structure sits the thing that actually changes your week: written decision rights. A responsibility mapping — who is responsible, who is accountable, who is consulted, who is informed — is a heavier instrument than most companies this size need in full, but the accountable column is not optional. Pair it with an explicit delegation ladder, because 'delegate more' is useless advice and 'decide, act, then tell me' versus 'recommend, then wait' is a workable instruction.

The most common people problem in an acquired business is the best technician who is a poor manager. Promoting them removes your strongest practitioner and creates your weakest supervisor, and the damage runs in both directions at once. The alternative is not to block their progression but to build a technical ladder that pays for excellence at the work — lead technician, principal estimator, senior engineer — so that management is a different path rather than the only promotion available.

Promoting from within during the first months is a genuinely hard call. It buys enormous goodwill from an inherited team and it is the fastest way to install a manager who understands the business, and it also locks you into an assessment made before you understood the person or the role. The safer version is an explicit interim appointment with a stated review date and written outcomes, which is honest with everyone and reversible without humiliation.

Whether you promote or hire, write the seat's outcomes first. A job scorecard states what must have been achieved by when, the competencies that make it achievable, and how each will be measured. It is the same discipline as the accountability chart applied to one seat, and it converts a hiring decision from an impression into a comparison.

Finally, design against key-person concentration rather than only against workload. Ask which single people, if they left in month two, break the business — then treat the answer as a structural problem rather than a retention problem. Bench depth behind each critical seat is what turns a business that depends on individuals into one that depends on roles, and it is also the honest form of succession planning inside a company you bought.