What is the minimum detectable effect (MDE)
The minimum detectable effect (MDE) is the smallest lift an A/B test can reliably detect with the traffic you have. You choose it before launching the test, and the smaller it is, the more visitors you will need.
By Fernando Blanco Dosil · Updated November 2, 2026
What it means in practice
Say your store converts at 2%. Setting a 10% MDE means: I want my test to be able to see whether the variant lifts conversion to 2.2% or more. If the real lift is smaller, say 3%, the test will probably miss it, even though it exists.
That is why the MDE is not what you expect to win, but the bar below which you accept being blind. It is a business decision: which improvement is worth detecting.
Relative or absolute
It can be expressed in two ways, and mixing them up is a very common mistake with calculators.
- Relative: a percentage of your current conversion. From 2% to 2.2% is a 10% relative MDE.
- Absolute: the difference in points. From 2% to 2.2% is an absolute MDE of 0.2 points.
- An absolute MDE of 1 point on a 2% baseline is a 50% relative lift, something almost no change achieves. Always check what the calculator is asking for.
Why small effects are so expensive
Visitors per version for a 2% baseline conversion rate, at 95% confidence and 80% power.
| Relative MDE | Variant conversion | Visitors per version |
|---|---|---|
| 5% | 2.1% | 315,210 |
| 10% | 2.2% | 80,683 |
| 20% | 2.4% | 21,109 |
| 30% | 2.6% | 9,798 |
Calculated with the classic two-proportion formula. Halving the MDE roughly quadruples the sample.
How to choose your MDE
Start with the business: which improvement is worth the effort. If 5% more sales means 200 euros a month, it may not be worth six months of testing. If it means 20,000, it is.
Then look at traffic. Work out how long the MDE you would like takes and, if the answer is absurd, raise the MDE or change what you test. A three-week test that concludes beats a six-month one nobody finishes.
How Bixlow handles it
The MDE is a planning concept: it helps you decide before you start whether a test is viable. Bixlow does not show an MDE in its results. What it shows is the probability that each version is the best, the estimated lift with its credible interval and whether the minimum sample and duration have been met.