Year three is when a center stops improving on its own. The symptoms show up before the numbers move, which makes it diagnosable rather than inevitable.
Years one and two of a center improve almost regardless of what you do. Ramp is happening, people are learning the work, and the baseline was zero, so every quarter looks better than the last. Then somewhere in year three the curve flattens, and it flattens without anything visibly going wrong.
This is a pattern we see often. It is not a law, and I want to be careful about that, because the version of this argument that says every center inevitably plateaus is both unfalsifiable and useless. What I will claim is narrower: it happens often enough to be worth testing for deliberately at the end of year two, and when it happens the cause is usually structural rather than a talent problem. Read what follows as a diagnostic, not a prediction about your center.
Why it happens when it happens
The improvements of years one and two come from ramp. People get faster at the work, the obvious process defects get fixed, and the attrition of the first year settles. None of that requires a change in how the operation is governed, which is why it works without one.
By year three the ramp is done. The next increment of improvement has to come from changing how the work is designed, and that requires authority the center usually does not have. If the operating method belongs to a partner, or the process documentation lives in somebody else’s system, or every material change has to be approved by a function two time zones away that is measured on something else, then the center has run out of the improvements it is permitted to make. That is the plateau. It looks like a performance problem and it is an authority problem.
Symptoms, before the numbers move
These are observable, and they show up a quarter or two before cost or quality metrics change. Test them against your own center rather than reading them as findings about it.
Improvement ideas stop arriving. Not because people stopped having them, but because the last several went nowhere. When a center goes quiet on process suggestions, that is usually the most reliable early signal there is.
Every escalation goes to the same two people. Knowledge concentrated in individuals rather than in documented workflows means the operation cannot absorb their absence, and it means nobody else is being trusted with judgment.
The documentation is stale and everybody knows it. Process documents describing how the work was done eighteen months ago, with the real method living in people’s heads, is the signature of an operation that has been improving informally and recording nothing.
Onboarding time is going up, not down. In a healthy operation the fifteenth hire ramps faster than the fifth. If it is slower, the accumulated knowledge is not in a form that transfers.
Senior people are leaving and junior people are staying. The layer with options is the layer that notices a ceiling first. Retention that looks stable in aggregate can be hiding this entirely.
Reporting has become the deliverable. When a growing share of the center’s senior time goes into reporting on the work rather than improving it, the operation has become something that is managed rather than run.
Four diagnostic questions
Who can change a process here without asking anyone outside this building? If the answer is nobody, the plateau is explained and no amount of hiring will move it.
Where does the operating method actually live? In your documentation, or in a partner’s playbook, or in four people’s experience. This determines which intervention is available to you.
If the top three people left this quarter, what would break? The honest answer is a measure of how much of the operation is in individuals rather than in the system.
What has this center been allowed to decide in the last six months? Not asked to do. Allowed to decide. The list is usually shorter than leadership expects, and its length is the ceiling.
Three interventions, and how to choose
Move the authority. If the diagnosis is that the center cannot change its own processes, the intervention is governance: give it defined decision rights, name what it owns, and change what it is measured on to match. This is the cheapest of the three and the one most often skipped, because it requires a change in the parent organisation rather than in the center.
Move the method. If the operating playbook belongs to a partner, no amount of local authority helps, because the thing to be improved is not yours to change. The intervention is a transition of the method into your own documentation and your own systems, which is slower and more disruptive than it sounds and is the whole substance of what people mean by the Level 2 Trap.
Redesign the work. If the authority and the method are already yours and the curve is still flat, then the operation is running its current design as well as that design allows. The next increment requires a different shape: which work is judgment, which is repeatable, and what changes if the repeatable part becomes governed AI roles with named owners. This is the intervention people reach for first, and it is the one most likely to fail if either of the other two is unresolved, because AI layered onto an operation whose method belongs to somebody else deepens exactly the dependency you were trying to escape.
Choose in that order. It is also the order of increasing cost, which is convenient, and the order of increasing appeal, which is not.
What year three is not
It is not a talent problem by default. Retention and hiring interventions do not fix a governance boundary, and a center that is told its plateau is a people problem will conclude, correctly, that leadership has misread it.
It is also not inevitable. Centers that were designed with real decision rights and their own documentation from the start tend not to hit this, which is the strongest evidence that the cause is structural rather than temporal. Year three is when the design decisions of year zero become visible. If you are at year zero now, that is the useful thing to take from this.