Lift (also called uplift or conversion lift) is the percentage change in a metric between a test variant and its control: (variant rate − control rate) ÷ control rate × 100. If a control converts at 2.0% and the variant at 2.4%, the lift is +20%.
A negative lift means the variant did worse.
Lift is almost always reported as a relative number.
The raw gap between the two rates, here 0.4 percentage points, is the absolute difference, and mixing the two up is one of the most common ways test results get misread.
What uplift means
"Uplift" and "lift" mean the same thing in CRO: how much better (or worse) the new version performed than the old one, as a share of the old one.
"Conversion lift" is the same idea applied to conversion rate, but you can measure lift on any metric, such as revenue per visitor, add-to-cart rate or average order value.
| Measure | Formula | Example (2.0% to 2.4%) |
|---|---|---|
| Relative lift | (B − A) ÷ A × 100 | +20% |
| Absolute difference | B − A | +0.4 percentage points |
Testing tools usually show relative lift by default.
Optimizely, for example, displays the relative improvement over the baseline for A/B tests, but switches to absolute improvement for bandit tests and tests with Stats Accelerator turned on.
Check which one your tool shows before you compare numbers across tests.
Lift in A/B testing
A test on a whisky subscription homepage run by Convertibles shows how the two numbers relate.
Adding a sticky call to action moved conversion from 1.135% to 1.366%, across 25,935 visitors with a 95.1% probability to beat control.
- Absolute difference: 1.366 − 1.135 = 0.231 percentage points.
- Relative lift: 0.231 ÷ 1.135 = 0.2035, or +20.4%.
A "+20% lift" sounds large, yet it came from less than a quarter of a percentage point.
That is normal: on low baseline rates, small absolute changes produce big relative lifts.
On a higher baseline, the same absolute gain gives a smaller lift.
In a wellness product page test by Oliver Kenyon, purchase conversion went from 5.27% to 5.74%, a 0.47-point gain that works out to +8.93%.
Three points to keep in mind when you read a lift figure:
- Lift is an estimate, not a fact. The reported number is a single point from a range of plausible values. A test with few conversions can show +20% when the true effect is close to zero.
- Lift depends on the metric. A sports nutrition test by Blend Commerce that renamed "autoship" to "subscribe" gave +100.43% subscription orders per visitor, while add-to-cart rate fell 11.89%. The same change had a positive lift on one metric and a negative lift on another.
- Winning tests overstate lift. Tests that are stopped because they look good tend to report effects larger than what shows up after launch, a pattern explained in our page on the winner's curse.
Lift is also the unit you plan with.
The minimum lift a test is set up to detect is its minimum detectable effect, and smaller lifts need far more traffic to detect.
@oliverkenyon
Blend Commerce
ConvertiblesRelated terms
- Baseline conversion rate: the control's rate, the denominator in every lift calculation.
- Control and variant: the two versions whose rates you compare.
- Minimum detectable effect: the smallest lift a test is powered to find.
For how large lifts usually are across real tests, see our research on average A/B test lift.
You can browse every result by page and element in the test database, or return to the CRO glossary.