Definition
Minimum detectable effect (MDE) is the smallest improvement an A/B test is designed to reliably detect. It is decided before the test runs, and it drives how much traffic the test needs. Expressed as a relative percentage, an MDE of 10% on a 3% baseline conversion rate means the test is built to catch a move to 3.3% or better.
MDE is where most testing mistakes begin, because teams skip it. Without a stated MDE, you cannot know how much traffic a test needs, so you either run underpowered tests that never reach a conclusion, or you stop early and mistake noise for a result.
How MDE relates to sample size
MDE and sample size trade off against each other:
- A smaller MDE (detecting tiny effects) requires a much larger sample, because tiny differences are hard to distinguish from noise.
- A larger MDE (only catching big effects) needs less traffic, but risks missing real, smaller improvements.
Detecting a 5% relative lift can require several times the traffic of detecting a 10% lift. So the practical question is not "what is the smallest effect possible?" but "what is the smallest effect worth my time to detect?" Answer that first, then size the test with a sample size calculator, and confirm the result with an A/B test significance check.
Why MDE matters for revenue
MDE forces an honest decision about what a test is for. If a change would only be worth shipping at a 10% lift, there is no point powering the test to detect a 2% one and paying for many times the traffic. Setting the MDE aligns the test's cost with the value of the outcome.
It also prevents two opposite failures. Too small an MDE burns traffic and time chasing effects too minor to matter. Too large an MDE quietly misses real wins because the test was never powerful enough to see them. The right MDE sits at the smallest lift that would actually change your decision.
From a Revenue Intelligence perspective, MDE is part of measuring a fix honestly. When you close a revenue leak and test the result, the MDE defines what "worked" means in advance, so you are not tempted to declare victory on a difference too small to be real. The Revenue Intelligence framework treats this discipline, deciding what counts as success before you look, as central to trustworthy measurement, whether you are testing a checkout change or any other fix.
Related metrics
Minimum detectable effect works with statistical significance, which tells you whether the effect you set out to detect is real, and it directly determines the sample size a test needs. Set the MDE, size the test, then run to significance.
In short
Minimum detectable effect is the smallest lift a test is built to catch. It is decided up front, it drives sample size, and choosing it well aligns a test's cost with the value of what you are trying to learn. Set it before you run, not after.