How to Calculate Minimum Detectable Effect (MDE)

MDE in plain language

If a page converts at 4%, an MDE of 10% relative lift means you design the test to detect a change of 0.4 percentage points: from 4.0% to 4.4%. In absolute terms, the MDE is 0.4 percentage points; in relative terms, it is 10%. State both so stakeholders do not mistake one for the other.

A smaller MDE takes more visitors to detect because it is harder to separate from random variation. A larger MDE produces a shorter test, but may miss improvements that are meaningful at scale. The right number is based on the value of the decision, not on the lift you hope to see.

A practical way to choose it

1. Start with baseline performance. Use the eligible audience’s recent conversion rate, not a sitewide average.
2. Translate impact into value. Estimate what an absolute conversion change is worth in incremental orders, qualified leads, or revenue.
3. Set the smallest change worth implementing. Include engineering, design, operational, and customer-trust costs.
4. Check feasibility. Use the baseline, MDE, confidence, and power to estimate required sample and duration.
5. Adjust the experiment, not the math. If it is infeasible, test a larger intervention, higher-frequency metric, or higher-traffic surface.

For example, a subscription page with 20,000 eligible monthly visitors may be able to detect a 10% relative improvement, while a low-traffic enterprise demo page may need a larger change or a different research method.

Common MDE mistakes

Do not choose an unrealistically tiny MDE just to sound rigorous; it can create an experiment that would take months. Do not choose a huge MDE only to get a quick answer; a no-result would tell you little about a change that still may have paid for itself.

Also avoid using a post-hoc observed lift as the MDE. Decide the threshold before traffic flows, then interpret the full effect estimate and uncertainty at the end. A test can miss its preplanned threshold and still provide useful directional evidence, but it should not be relabeled a win after the fact.

Put MDE into the launch plan

Enter your assumptions in our A/B test sample-size calculator, then plan enough time to cover a representative business cycle. Record the absolute and relative MDE in the experiment brief, alongside the primary metric and guardrails.

MDE works with either frequentist or Bayesian reporting, but it does not replace a decision rule. Our frequentist vs. Bayesian guide explains why the analysis method and the operating process must line up.