Moderation

Moderation is the principle that the relationship between a variable and its outcome is often non-monotonic — more isn't always better, and the optimal point usually sits between extremes. In statistics, a moderating variable changes the strength or direction of a relationship between two other variables. In practice, it's the recognition that most inputs have diminishing — and eventually negative — returns past a certain threshold. Work hours, growth speed, feature count, fundraising aggression, risk-taking: each has a range where returns peak and a zone where additional input actively destroys value. The model warns against linear extrapolation and rewards the discipline of finding the inflection point where enough becomes too much.

For every major input — spend, headcount, features, effort — identify where returns peak and where they begin declining. Don't assume that what worked at one level will work at twice the level. Monitor for signs of overshoot: declining marginal returns, rising error rates, team burnout, customer confusion. Build the instinct to ask "are we past the optimal point?" before defaulting to "more."

Moderation recognizes that most inputs have an optimal range and that more eventually becomes worse. Find the inflection point, stop before returns go negative, and resist the assumption that scaling an input linearly will produce linear results.