Don't test everythingDon't test everything
, but you must test. If you knew exactly what to do you would just do it (and so would your competitors)If you knew exactly what to do you would just do it (and so would your competitors)
.
But "a test every month or two" is simply too slow to succeed.
Suggetions:
- Focus on systems, not hacksFocus on systems, not hacks
It takes robust processes to Scale wins across the entire organization.
- Decouple experimentation from product developmentDecouple experimentation from product development
Your product development process is optimized for shipping a reliable, polished experience to users.
But experimentation requires you to Explore before you refine and Take big swings early o... - Run only as many variants as the hypothesis requiresRun only as many variants as the hypothesis requires
If it's possible to test a hypothesis with a single variant, don't test additional variants.
Running fewer variants lowers your Minimum Detectable Effect and simplifies implementation and QA.
...