Do You Need a COO Yet? What Breaks at 20, 50, and 80 People
Article
Founders ask this question at the wrong moment, usually right after something breaks. Here’s what breaks at each stage, which of it is actually a COO problem, and why the answer arrives earlier than it used to.
“Do we need a COO yet?” is a question founders ask reactively. Something went wrong, a client got three conflicting answers, a hire fell through, the finance close took three weeks, and, in the post-mortem, someone says the words.
That’s a fine time to ask. It’s just late. The better version is to know in advance what breaks at each stage of growth, so that when it happens you recognize it instead of being surprised by it.
So here is what I’ve watched break, over and over, at three headcounts. Not every company hits these at exactly 20, 50, and 80 — revenue model, industry, and how much of the work is automated all move the numbers. But the sequence is remarkably consistent, and the sequence is what matters.
Around 20 people: the founder becomes the router
At twenty people the company still fits in the founder’s head. Everyone knows everyone, most decisions get made in a hallway, and the founder is the person every question routes through. This works. It works right up until the moment it doesn’t, and the moment usually looks like this:
The founder’s calendar is full of things that don’t need them. Not big decisions — small ones. Which vendor. Whether to approve the expense. How to handle the customer who wants an exception. Each one takes five minutes and there are forty of them a day, and the founder is now a router, not a leader.
What this is: a founder-capacity problem. It’s not a COO problem yet, it’s the moment for an Executive Assistant, and soon after, a Chief of Staff. Someone who takes the logistics, then someone who takes the coordination.
Except for one thing. Twenty-person companies in 2026 are not twenty-person companies in 2023. A team this size can be running a dozen AI agents across support, reporting, recruiting, and finance — adopted one at a time by whoever needed them, with nobody deciding who is accountable when one of them is quietly wrong for three weeks. That isn’t a coordination problem. It’s an ownership problem, and ownership is a COO’s job.
So the honest answer at twenty is: probably not yet, unless nobody can name the person who owns the AI. If that’s you, the COO-shaped hole has arrived early, and it doesn’t care what your headcount is.
Around 50 people: functions exist but nobody owns them
Fifty is the stage where the company stops being a group of people and starts being a set of functions. There is a sales team now. There is someone doing finance. Someone is “doing HR,” which usually means the office manager and a payroll login. The functions exist.
What doesn’t exist is ownership of them. The person doing finance is doing it in addition to something else. The sales team reports to the founder, who is not a sales manager. Three departments are making commitments to the same client, and none of them checked with the others because there’s no one whose job it is to make them check.
Here’s what breaks: handoffs. Sales to delivery. Delivery to finance. Hiring to onboarding. Every place where one function ends and another begins, there’s a gap, and things fall through it. The founder finds out from the customer.
What this is: a structure problem, and structure is the COO’s job. This is the classic moment. Nobody owns the functions, and the founder can’t own them all, and the Chief of Staff (if there is one) can coordinate around the gaps but has no authority to close them. The question at fifty isn’t whether you need an COO. It’s whether you need one full-time yet and at fifty, honestly, most companies don’t. This is where a fractional COO earns their keep: build the structure, put owners on the functions, and hand over a working operation to whoever runs it next.
Around 80 people: the company runs on people who remember
Eighty is the stage where structure that was never written down stops working, because the people who held it in their heads are outnumbered by the people who joined last year.
Here’s the tell. Something that always worked stops working, and when you dig in, it turns out it only ever worked because Maria knew to do it. Maria was promoted, or left, or is just too busy now, and the thing she used to do wasn’t a process. It was Maria.
Multiply that by every function. At eighty, the company is running on institutional memory, and institutional memory doesn’t scale, doesn’t onboard, and doesn’t survive turnover. The operating cadence, who meets, about what, with what numbers in front of them, must become real, because the hallway version has stopped reaching everyone.
What this is: an operating-system problem. The company needs someone whose full-time job is to make it run, to own the cadence, the metrics, the cross-functional decisions, and the managers of the managers. At eighty, the answer to “do we need a COO” is yes, and the only question left is whether the founder hires one now or discovers it at a hundred and twenty.
| ~20 people | ~50 people | ~80 people | |
|---|---|---|---|
| What breaks | The founder becomes the router for every small decision | Handoffs between functions; nobody owns the gaps | Things that only worked because someone remembered |
| The founder finds out from | Their own calendar | The customer | Turnover |
| What kind of problem | Founder capacity | Structure | Operating system |
| Usual answer | EA, then Chief of Staff | Owners on the functions — often a fractional COO | A full-time COO |
| The AI exception | If nobody owns the AI, the COO question arrives here | AI agents inside handoffs make the gaps worse, faster | Governance is now a function; someone has to run it |
Why the threshold moved
Three years ago I would have written this article with the first stage at thirty and the COO conversation starting at fifty. The reason it now starts at twenty is the same reason I keep coming back to: AI has moved the point at which a company has more operational surface than its founder can hold.
An agent inside a workflow is a decision being made. A dozen of them across a company is a dozen decision-makers, none of whom sit in the leadership meeting. That governance surface used to appear at eighty people, because it took eighty people to have that many decisions in flight. Now it appears at twenty, and it lands on whoever notices first which is usually nobody, until something is wrong.
I’ve written elsewhere about why the COO is the natural owner of that surface. The short version is that it isn’t technical work. Decision rights, escalation paths, accountability for output — those are operations, and they need an owner, not a policy.
The question to ask this week
Whatever your headcount, here’s the diagnostic in one line: when something operational breaks, who finds out, and how?
If you find out from your calendar, you’re at stage one. If you find out from the customer, you’re at stage two. If you find out because someone left, you’re at stage three.
And if the answer is “nobody finds out, because it’s an agent and it’s been wrong for a while” — that’s the fourth answer, and it means the question you came here to ask is already answered.
