Most of your tests should be cheap, fast, and small

Most of your tests should be cheap, fast, and small

There is a romance to the big swing, the full redesign, the new pricing model, the channel nobody has tried. Big swings have their place, but a growth programme built on them is slow, expensive, and fragile. The engine that compounds is a high volume of small, cheap, fast tests, because volume is what turns experimentation from luck into a rate of learning.

The maths is simple and unforgiving. If your hit rate is one winner in four tests, then the team running twenty tests a quarter finds five winners and the team running four finds one. The difference is not cleverness, it is throughput. So the discipline is to drive the cost of a single experiment down until you can run many, then keep the queue full. Every hour you shave off building a test is an hour you spend running another one.

Cheap and fast also changes how it feels to be wrong. When an experiment costs a day, a losing result is a lesson. When it costs a month, a losing result is a wound, and wounded teams stop testing the risky things that produce the biggest wins. Low cost per test is what keeps the appetite for being wrong alive, and being wrong cheaply is the entire game.

Small does not mean trivial. A small test is one with a tight scope and a clean read, not one that changes nothing that matters. The skill is to find the smallest version of the idea that still produces a real signal. You do not need to rebuild the onboarding to test whether a shorter first step lifts activation, you need to test that one step. Shrink the test, not the question.

Reserve the big swings for when the small tests have pointed the way. The pattern that works is to probe cheaply, find where the signal is strong, and only then commit a large build to exploit it. Big bets earned by evidence are good bets. Big bets placed on a hunch are how teams burn a quarter and learn nothing. Let the cheap tests buy the conviction for the expensive ones.

INTERVIEW EWOUD: Give an example where running many small cheap tests beat one big bet, for your own venture or a client. Roughly how many tests, what did they cost to run, and what did the volume surface that a single big test would have missed?