A baseline conversion rate is the conversion rate your page, flow or app converts at today, before you change anything.
In an A/B test it is the control group's rate, and every lift, sample size and minimum detectable effect is measured against it.
What baseline conversion means in A/B testing
A conversion rate is conversions divided by visitors.
The baseline is that number for the current version, measured over a stable period on the same audience and the same goal you plan to test.
It matters in three places:
- Planning. The baseline sets how much traffic a test needs. Low baselines need far more visitors to detect the same relative change.
- During the test. The control keeps measuring the baseline live, so seasonality and traffic mix affect both arms equally.
- Reporting. Lift is the variant's rate compared with the baseline. The same absolute gain means a very different relative lift on a 2% page than on a 40% page.
Define the baseline for one specific metric. A store's sitewide purchase rate, its product-page add-to-cart rate and its mobile checkout completion rate are three different baselines, and a test should use the one that matches its primary metric.
Why the baseline drives sample size
A common approximation for the visitors needed per variant, at 80% power and a two-sided 5% significance level, is:
n per variant ≈ 16 × p × (1 − p) / d²
Here p is the baseline conversion rate and d is the absolute difference you want to detect.
The 16 comes from 2 × (1.96 + 0.84)², the z-values for those two settings.
Our calculation below holds the relative lift fixed at 10%:
| Baseline rate | 10% relative lift | Absolute difference (d) | Visitors per variant |
|---|---|---|---|
| 2% | 2.0% to 2.2% | 0.2 points | about 78,400 |
| 10% | 10% to 11% | 1 point | about 14,400 |
A page converting at 2% needs more than five times the traffic of one at 10% to detect the same relative change.
That is why many teams test on higher-baseline micro-conversions such as add-to-cart clicks, then check purchases as a secondary metric.
The A/B test sample size calculator does this math for you, and the minimum detectable effect guide covers how to pick d.
Examples from the database
Real tests show how much the baseline shapes what a result means:
- On GRENE's desktop homepage, a USP redesign tested by VWO moved ecommerce conversion from 2.64% to 3.18%. That is about half a percentage point in absolute terms, but roughly a 20% relative lift on a low baseline.
- GRENE's desktop mini-cart redesign moved ecommerce conversion from 1.83% to 1.96% over 36 days. By our calculation that is about a 7% relative lift, and at a 1.83% baseline the formula above puts the traffic needed to reliably detect it at roughly 170,000 visitors per variant.
- GRENE's mobile category listing test measured a much higher-baseline metric, product click conversion, which went from 36.81% to 43.03%. A 6-point gain there is about a 17% relative lift.
- Trinity Insight's consolidated buy box for Taylor Gifts moved ecommerce conversion from 1.53% to 3.23%, a doubling from a low baseline. The case study gives no sample size or significance, so treat it as a strong signal rather than a precise estimate.
How to measure your baseline
- Use at least one or two full weeks of data so weekday and weekend behavior are both included.
- Match the audience and device mix you will test. A desktop-only test needs a desktop baseline.
- Skip unusual periods such as a sale, a launch spike or a tracking outage.
- Measure with the same tool and event definition the test will use, so the planned and observed baselines line up.
If the control's rate during the test differs a lot from the baseline you planned with, recheck your sample size.
A lower real baseline means the test is underpowered.
Related terms
- Conversion rate: the metric the baseline is built on.
- Lift: the change measured against the baseline.
- Control and variant: the control arm measures the baseline during a test.
- Sample size: depends directly on the baseline rate.
Browse more terms in the CRO glossary or search every result in the test database.