The minimum detectable effect turns a vague wish to measure a change into a concrete design target.
Without it, there is no way to decide how many users a test needs.
It links the statistics to the business, because the effect worth detecting depends on what the change costs.
It is stated either as a relative lift, such as 5% more sign-ups, or as an absolute one, such as 1 percentage point.
The difference matters: a 5% relative lift on a 10% rate is half a percentage point.
Together with the threshold and the power, it sets the sample size.
Halving the minimum detectable effect roughly quadruples the sample needed.
Teams with little traffic must accept larger detectable effects, or their tests would never finish.
Changes to mature products often have small effects, so many teams plan for lifts of a few percent.
Setting it too large produces quick tests that miss real but modest wins.
Setting it too small produces tests that need more users than the site has.
Effects smaller than the target can still come out significant, but the test is not built to catch them reliably.
In practice it is chosen from past tests, the cost of the change and the traffic available.
Writing it in the test plan lets readers interpret a result of no difference correctly afterwards.
The usual misreading is treating the minimum detectable effect as a forecast of the effect, when it is only a design choice.