Test Duration

A credible test plan estimates sample size from the baseline rate, a meaningful effect, an acceptable error risk and expected traffic. The period should allow the needed sample to accumulate and cover relevant traffic patterns without introducing unplanned changes. Running a test for exactly seven days because a template says so can leave results inconclusive when conversion events are rare.

Stopping as soon as one variant briefly looks ahead increases the risk of drawing a false conclusion. Test operators should define stopping rules in advance, monitor instrumentation and track issues such as a campaign launch, pricing change or traffic outage. If an experiment uses a method designed for sequential monitoring, its statistical assumptions differ from those of a fixed-horizon test and must be followed correctly.

Duration is not the same as statistical significance or test sample size, although the three interact. A longer test does not automatically correct biased assignment, tracking failures or seasonality. For a local service site with low weekly lead counts, it may be better to test a larger, meaningful change less frequently than to claim an outcome from a handful of form submissions.