Lift

Lift is the percentage improvement in a metric (like conversion rate) that a test variant achieves compared to the control, measuring the real impact of a change.

Lift is the percentage difference in performance between a test variant and the control. It's the primary way teams quantify the impact of an experiment — answering the question "how much better (or worse) did this change make things?"

Lift = ((Variant Metric - Control Metric) / Control Metric) x 100

For example, if the control converts at 4.0% and the variant converts at 4.6%, the lift is +15%.

Positive vs. Negative Lift

  • Positive lift — The variant outperformed the control. The change improved the metric.
  • Negative lift — The variant underperformed. The change hurt the metric.
  • Zero (or near-zero) lift — No meaningful difference. The change had no detectable impact.

Negative lift is still a useful result — it tells you what doesn't work and prevents you from shipping a change that would hurt performance.

Relative Lift vs. Absolute Lift

  • Relative lift — The percentage change (e.g., "+15% lift"). This is what most people mean when they say "lift."
  • Absolute lift — The raw difference in metric values (e.g., "0.6 percentage points"). More useful when comparing across tests with different baseline rates.

A 15% relative lift sounds impressive, but if the baseline is 0.5%, the absolute change is only 0.075 percentage points. Context matters.

Why Lift Alone Isn't Enough

Lift needs to be paired with statistical significance and confidence intervals:

  • Lift without significance — The improvement might be due to random chance
  • Lift with a wide confidence interval — The true improvement could be anywhere in a large range
  • Lift with narrow confidence and high significance — You can confidently ship the variant

Lift in Practice

Most CRO programs target lifts of 5–20% per winning test. Smaller lifts (1–3%) are common for mature, well-optimized pages. Larger lifts (20%+) are possible on pages that haven't been tested before or have obvious UX problems.

The real gains come from compounding many small lifts over time through continuous testing.