How much traffic do you need before A/B testing is worth it?

As a rough floor, you need around 1,000 conversions a month before classic A/B testing gives reliable answers on small effects. Below that, tests take months to read or never reach significance. That does not mean optimization stops. It means the method changes from testing small tweaks to shipping large, evidence-backed changes.

Why the number is about conversions, not visitors

Sample size is driven by how many outcomes you observe, not how many people arrive. A site with 200,000 sessions and a 0.4% conversion rate has less statistical power than one with 40,000 sessions converting at 4%.

That is why the benchmark matters. Across 1,055 audited tests, ConversionTeam found a median control conversion rate of 4.6%, with ecommerce near 4.7%, subscription at 3.6% and lead generation at 4.5%. Run your own number before assuming you are testable.

The effects you are hunting are small

This is the part that surprises people. In DRIP’s experiment database, winning tests produced a median conversion rate uplift of 1.88% and a median revenue per visitor uplift of 2.77%. A 2% relative lift is a real, valuable result and it is also nearly invisible without serious volume.

Detecting a 2% relative change at conventional confidence needs tens of thousands of conversions per variant. Detecting a 20% change needs a few hundred. So the honest question is not “can I test” but “how big does the change have to be before I can see it”.

Treat those bands as planning guides, not statistics. Run a proper power calculation before any individual test.

What low-traffic brands should do instead

Test bigger things. A new page structure beats a headline swap when you can only see large effects. Whole-page redesigns are testable at low volume precisely because the effect size is large.

Use aggregate metrics. Cart and checkout tests read faster than product page tests because the denominator is already qualified. ConversionTeam’s data shows cart and checkout tests running at 25% to 32% conversion, which is an order of magnitude more signal per visitor.

Use qualitative research harder. Session recordings, customer interviews and support tickets do not need statistical power. At low volume they are more informative than a test that will never conclude.

And accept before-and-after measurement with clear eyes. It is confounded by season, promotion and traffic mix, so it is weaker evidence. It is still better than opinion, and it is honest as long as you say so.

The trap to avoid

Do not run an underpowered test and call it at day seven because the numbers look good. Early peeking is how teams convince themselves of results that reverse. The median test in DRIP’s data ran for 42 days, and published benchmarks collected by roast.page put the median at 23 days. Either way, it is weeks. Decide the runtime and sample size before launch and hold to it.

The other common error at low volume is running two tests on overlapping traffic. Concurrent tests are fine when the samples are large enough to stay independent. Below roughly a thousand monthly conversions they contaminate each other and neither result means much.What this looks like in practice

The largest result we have seen came from a change that needed no statistical subtlety at all. At a DTC supplements brand, a shipping threshold test produced two million dollars in profit. It worked because it changed order composition rather than order frequency, and effects that large are visible at almost any traffic level.

That is the principle for smaller brands. Look for the tests where the mechanism is big and structural, not the ones where you are hoping for a two percent lift you cannot measure.

The starting point

Pull your last ninety days. Count conversions, not sessions. Divide by three. If that number is under a thousand, your first quarter should be research and a small number of large, well-argued changes rather than a test queue.

We publish how the audit and roadmap sequence works at Parah Group, including how the approach changes at lower volumes.