When does evolutionary change stop working, and an imposed process redesign become the right call?

## What evolutionary change is good at, and what it cannot reach Small reversible changes are the right default because they are cheap to be wrong about, they need nobody's permission, and they accumulate. What they cannot do is change the shape of the thing they are optimising. Every column policy, work-in-progress cap and entry criterion operates **inside** a given division of responsibility. If that division is what produces the delay, no sequence of policies reaches it, and a team can run experiments diligently for two quarters while the same number refuses to move. This is the honest limit, and a lead is expected to know it rather than to keep faith with the method past the point of evidence. The failure of nerve in the other direction is just as common: reorganising because a redesign feels decisive, when a fortnight of measured policy changes would have done it. ## Signals that the structure is the constraint None of these is conclusive alone; together they are a case: - **Flat results across several honest experiments.** Not one disappointing change - a run of them, each with a stated prediction, none of which moved the measure. - **The dominant blocked-days category is owned outside the room.** When most time lost sits with a group that attends none of the team's cadences, the team is not empowered to fix its own biggest cost. - **The same policy keeps getting reverted.** A rule that is agreed, then quietly abandoned every quarter, is usually fighting an incentive that lives elsewhere. - **Every item crosses the same hand-off.** If the division of work guarantees a queue on 100% of items, the queue is structural, not behavioural. - **Improvement depends on one person's presence.** A loop that stops the week its convenor is away was never a system change. Against those, a stage that queues, a class of work that is estimated badly or a review step people avoid are ordinary problems that policy changes handle well. ## What a redesign costs | Cost | What it looks like | |---|---| | Relearning trough | weeks of reduced output while people work out who does what now | | Lost tacit knowledge | the informal routes around old hand-offs vanish with them | | Reset baseline | before-and-after comparison across the change is weak, so the last experiments become unreadable | | Political capital | roughly one such move per year of credibility, whether or not it works | | Concentrated risk | one large bet replaces many small ones, and it cannot be reverted in a stand-up | The reset baseline is the one most often missed. Everything the team learned about its own flow is expressed in a structure that no longer exists, and the new arrangement needs weeks of finished work before it can be judged at all. ## Making the call defensible 1. **Bound it.** Redesign one service or one boundary, not a department. A bounded change keeps a comparison group and limits the trough to the people who must experience it. 2. **Name the constraint out loud** and state what about the redesign removes it. A redesign that cannot answer that question is a reorganisation in search of a rationale, and it usually moves the queue rather than removing it. 3. **Carry one measure across the boundary.** Keep measuring the same thing with the same clock even though the change makes it noisy, so the redesign is answerable to evidence rather than to how it feels. 4. **Set a date to inspect the redesign itself,** far enough out for the trough to pass, and say in advance what would count as it having failed. 5. **Time it.** Do not land a structural change immediately before a two-week holiday shutdown or any other period when the feedback loop cannot run - the trough will be attributed to the change and the evidence will be worthless either way. 6. **Restore the loop afterwards.** The new structure is not a destination; it is a new starting point for the same evolutionary cycle, and a redesign that is followed by no further experiments has simply replaced one frozen arrangement with another. ## The question behind the question An interviewer asking this is not looking for loyalty to a method. They are testing whether you can say what would change your mind - what evidence would make you stop running experiments, and equally what evidence would make you refuse a restructure that someone senior is pushing for. Both refusals are part of the job. The strongest answers name a real occasion of each: a time the data said the structure was the problem, and a time the pressure to reorganise was resisted because a cheaper change had not yet been tried.