What minimum detectable effect measures
Every test has a resolution limit, in the same way a bathroom scale does. Below that limit, a real difference and ordinary day-to-day variation look identical, and no amount of staring at the dashboard will separate them. The minimum detectable effect, usually shortened to MDE, is that limit stated in advance: the smallest change the test is big enough to notice.
It is set by three things you choose before launch. More traffic lowers it, because larger samples smooth out the noise. A longer run lowers it, for the same reason. A base conversion rate that is already high lowers it, because each visitor carries more information than on a page where conversions are rare. Fix any two and the third follows.
Read the other way round, MDE tells you what a test cannot do. If a page receives modest traffic and you can only run for a few weeks, the test will only ever be able to detect a large swing. Small refinements will come back inconclusive no matter how carefully they are built.
Why minimum detectable effect matters
It decides which ideas deserve a test at all. Rewriting a headline, replacing a hero image and cutting a form to its essentials are big enough changes to move a page noticeably. Nudging a button a shade darker usually is not, and testing it on a modest audience burns weeks to produce a shrug.
It also protects the budget conversation. Agreeing the MDE up front turns a vague request to test everything into a short list of changes worth the traffic they will consume. For sites in Nepal and other smaller markets, where a single landing page may see a fraction of the visitors a comparable UK or Australian site sees, this is not a technicality — it is often the difference between testing and guessing.
Common mistakes with minimum detectable effect
The usual mistake is calculating it after the fact, when a test has already failed to reach a verdict. By then the choice has been made for you. The second is confusing the smallest effect the test can see with the smallest effect worth having: those are different numbers, and the useful test is one where the first sits below the second.
The third is quietly shrinking the target mid-test because the result is not arriving. Once you lower the bar to fit the data you have, you are no longer running a test, only justifying a preference. Set the threshold before launch, alongside the sample size it implies, and write both down.
How to act on it
Start from the business side rather than the statistics. Ask what improvement would actually change a decision — what lift would justify redeveloping the checkout, or funding a new landing page — and treat that as the effect you need to detect. Then work backwards to the traffic and the weeks required.
If the arithmetic says the test would take longer than the plan allows, you have three honest options: test a bolder change, combine low-traffic pages into one test, or skip testing and make the change on judgement while watching the funnel. Pick one deliberately. Sustained conversion rate optimisation on a small site depends far more on choosing testable questions than on running more tests.