04 · CRO

Can your site run A/B tests?

A site with 100 conversions a month needs about four months to detect a 30% lift, and more than eight for a 20% one. Both exceed the 6-to-8-week maximum a test can run before seasonality and cookie loss contaminate the result. That site cannot run A/B tests — it can do other things, several of them more profitable, but not testing.

No signup · no email · runs in your browser

Your numbers

From GA4: users and conversions over the last 30 days.

Not testable

At this traffic, the test would take 23 weeks

That is 5.3 months running a single test. By the time it ends, the answer no longer applies to the business you had when you launched it. This is not a tool or agency problem: there is no volume.
Visitors per variant
53,208
Conversions per variant
1,596
Duration
23.1 wks
Duration against the 8-week ceiling23.1 wks
08 weeks12+

What you can actually detect with this traffic

The highlighted row is the smallest lift that fits in 8 weeks.

LiftVisitors/variantDurationVerdict
+5%207,93590.3 wksno
+10%53,20823.1 wksno
+15%24,19010.5 wksborderline
+20%13,9116.0 wksfits
+30%6,4522.8 wksfits
+50%3,3342.0 wksfits
+75%3,3342.0 wksfits

At this volume, here is what does work

  1. Friction audit. Heatmaps and session recordings — at least 100 sessions before concluding anything. You find what is broken, not what is “improvable”.
  2. Fixes that need no test. Forms failing on mobile, invisible CTAs, broken links, duplicated checkout steps. Just fix them.
  3. Step-by-step funnel analysis. A step losing 60% does not need statistical significance to justify touching it.
  4. Qualitative research. Five interviews with recent customers beat a test that will never close.

Why it exists

The two limits almost no calculator applies

A floor of 100 conversions per variant. Below that, a winner is noise even if the math says the sample is reached. When this floor overrides the formula, the tool says so on screen instead of hiding it.

A ceiling of 8 weeks. Beyond that, seasonality and cookie loss contaminate the result. And the minimum is 2 weeks: less does not capture a full weekly cycle.

When the verdict is red, the tool proposes what does work at that volume: friction audits with session recordings, fixes that need no test, step-by-step funnel analysis and interviews. Same useful work, correct label.

Questions

What people usually ask

The standard two-tailed two-proportion test, the same one used by the public calculators from Evan Miller, Optimizely and VWO. With a 3% base rate, α = 0.05 two-sided and 80% power, it returns 53,208 visitors per variant to detect a 10% lift.

Did you get a number you did not expect?

Send it over and I will tell you what I would do with that case in the first 30 days, with nothing to hire. If the honest answer is that you do not need an agency yet, I will tell you that too.