Three at the top, one at the edge

A simple allocation rule that survives contact with reality better than most: for every three units of effort spent on the channels carrying your volume, spend one at the edge of what you understand.

The three exist because the channels already working are where marginal improvement compounds, and because operations chronically under-invest in maintaining what works in favour of the next new thing.

The one exists because every channel decays, and the moment you need a new one is the worst possible moment to begin learning it. An operation that discovers its main channel has stopped working, and has no second channel it understands, is in trouble that no budget fixes quickly.

The ratio matters less than the discipline of protecting the smaller number. Under pressure it is always the experiment that gets cancelled, which is precisely backwards: the experiment is the insurance.

Two rules make it hold. The edge work gets a stated question and a date rather than an outcome target, because an experiment judged on return is not an experiment. And it does not get cancelled to fund the three, because the whole point is that it is protected from the pressure that would consume it.