Big experiments take forever and teach almost nothing. I've watched teams burn months on "MVPs" that answered no real question. Small experiments, run properly, hand you a clear signal fast.
The problem with big bets
The pattern I see again and again: a team commits to a six-month rebuild, ships it, and the metric doesn't move. Now what? Half a year gone and nothing you can point to.
Big experiments tangle too many variables together. Was it the feature? The positioning? The timing? You can't say. The test was too big to teach you anything.
A framework for fast learning
Run experiments that isolate one variable and pay off in days, not months:
- Start with a real hypothesis: "Cut the signup form from 8 fields to 4 and completion goes up 20%." Not "let's improve onboarding."
- Pick the metric up front: decide what you'll measure before you touch anything. One primary number. The rest is noise until that one moves.
- Run the smallest version: a landing page instead of a feature. A concierge instead of automation. Five users instead of five hundred.
- Set a deadline to decide: two weeks, max. If the signal isn't clear by then, the experiment was too big.
What to actually measure
Watch behaviour, not opinions:
- Leading indicators: activation, time-to-value, adoption. They predict retention better than any survey.
- Real signals: did they come back, invite someone, pay? Behaviour beats stated intent every time.
- Drop-off: where people leave is just your next experiment, pre-written.
What's the smallest test that could give us a clear answer? If you can't say, you're not ready to run it.
Small experiments stack up. Run one every two weeks and in three months you've got six clear answers — more than most teams collect in a year.