Back to Insights

A/B Testing · 6 min read

How long should an A/B test actually run?

AB

The most common mistake we see in existing testing programmes isn't a bad hypothesis. It's stopping a test the moment it looks like a winner. A variant showing a promising lift after three days feels like a result. Statistically, it's usually noise.

Why early results mislead

Traffic and conversion behaviour fluctuate by day of week, by traffic source mix, and by factors entirely unrelated to the test — a payday, a competitor promotion, an unrelated marketing push. A test needs enough time and enough conversions in each variant to average that noise out.

As a working principle, we run tests for a minimum of two full business cycles (typically two weeks) and until each variant has reached a pre-calculated sample size — never until a dashboard simply looks favourable.

The discipline that protects the result

This means committing to a test duration before the test starts, not deciding when to stop based on how the data is trending. It's a small discipline that prevents a large and common error: rolling out a change that was never actually proven to work.

It's slower than watching a live dashboard and calling it when the line moves. It's also the only way to know that a result is real before you build a roadmap on top of it.

Want this thinking applied to your own store, not just written about?

Book Your Free Strategy Call

Start here

Your next customer is already visiting your website.

Let’s make sure they don’t leave without buying.