All projects

data

An A/B test you can actually trust

Run an experiment and analyse it without fooling yourself.

intermediatePython~22h

Why it matters

Most reported A/B wins are noise. Being the person who asks about sample size and peeking is immediately valuable to any product team.

What you build

  • An experiment design with a pre-computed sample size
  • Assignment that is stable and unbiased
  • Significance testing done correctly
  • A written result including 'no effect'

If you want more

  • Add sequential testing
  • Add segment-level analysis

Build it 5 steps

  • Deciding after you see the data is how you get a result you like.

  • Assign users randomly and stably

Material for the whole build

Shows statistical literacy applied to a product decision, which is rarer than it should be.

Get a plan built around projects like this