The observation is agricultural and very old. Work a field harder and the yield rises, then rises more slowly, then stops rising, then falls. Turgot stated it in the eighteenth century and Ricardo built the theory of rent on it.

The precise form matters: holding everything else fixed, each additional unit of one input adds less than the one before it. The condition is doing the work. Returns diminish because the other inputs did not scale — the same land, the same equipment, the same management attention.

The curve has three regions, not two

Most uses of the phrase collapse it into "more stops helping". There are actually three parts and the distinction is where the decisions are.

Increasing returns. Early on, each addition adds more than the last. A second person on a task often more than doubles it, because one of them can specialize. Networks and platforms can stay in this region for a long time.

Diminishing returns. Each addition still helps and helps less. This is the normal operating region, and it is where you want to be — it means adding is still profitable, just less so.

Negative returns. Each addition subtracts. The project with too many people where coordination now costs more than the capacity added. This region is real and companies enter it regularly without noticing, because the input is visible and the coordination cost is not.

Where the bend is

The practical question is never whether returns diminish. It is where you sit, and the answer differs by input.

Headcount on one project. Bends early and hard, because communication paths grow faster than people do. The classic case is adding people to a late project, which makes it later.

Hours worked. Bends and goes negative faster than almost anyone plans for. Output per hour falls as hours rise, and the errors that arrive at the end of a long stretch cost more than the hours bought.

Spending on a channel. Bends as the reachable audience saturates. The first amount finds the people already looking; later amounts pay more per person for people less interested.

Polish on a deliverable. Bends sharply, and this is the one the Pareto principle is usually invoked about.

Skill. Bends much later than people expect, which is why deliberate practice keeps paying long after it feels like it should have stopped. The bend is far enough out that most people quit before reaching it.

The move the curve implies

When one input hits diminishing returns, the returns to a different input have usually gone up.

The field responds less to labor because it is short of water. The team responds less to headcount because it is short of clear decisions. The channel responds less to spend because the offer is wrong. That is why the law is a diagnostic and not a stopping rule: a bend is a signal that the constraint has moved somewhere else, which is the whole argument of the bottleneck.

The expensive mistake is to keep feeding the input that used to work. It is easy to keep doing, because it worked and because there is a visible lever attached to it, and it is the shape of most plateaus.