I had a conversation this spring with an athlete who badly wanted the 100m to be his race.
He was training hard. He was not skipping sessions. But he was not following the sprint program properly, and over a few months he had effectively conditioned himself like a 400m runner.
His 400m time improved significantly. His 100m time did not.
I explained that the workouts were not suggestions. Sprint training has science behind it. Rest intervals matter. Distances matter. Change them and eventually you are training for something else.
That is the cleanest description I know of a problem I have watched inside growth organizations more than once.
The team is working. Effort is not the issue. Activity is high and rising. But the inputs being optimized are no longer the inputs that produce the outcome anyone is being measured on, and because everything is busy, it takes a long time for that to surface.
It shows up as a set of metrics that all improve while the number that matters does not. Response rates up, cost per lead down, campaign volume up, pipeline flat. Every one of those is a real improvement. None of them is the event.
The question that usually breaks it open is not whether the team is working hard enough. It is this.
What are these activities training us to be good at, and is that the thing we said we needed?
Sometimes the honest answer is that the program drifted, quietly, one reasonable adjustment at a time, until it was producing a different athlete.
Drift of this kind has a recognisable mechanism. It is almost never one bad decision. It is a sequence of individually sensible adjustments, each made under pressure, each solving a real problem, and none of them evaluated against the outcome they were collectively supposed to serve.
A channel underperforms, so budget moves to the one converting fastest. That channel converts fastest because it captures buyers who are already in market. Pipeline holds up, so the decision looks correct, while new demand quietly stops being created. Nobody chose to stop creating demand. Every step was defensible.
Six months later the pipeline is roughly the same size and a completely different shape, and the shape was the part nobody was measuring.
Catching it early is unglamorous. Name, at the start, the two or three inputs that actually produce the outcome. Then check whether those specific inputs are still the ones being optimized. Not whether the metrics improved, because metrics almost always improve. Whether the ones improving are still the ones that matter.
He is fast. He will get the 100m time. But it meant going back to the program he had been slowly editing without noticing.