Uplift

Uplift is the relative difference of a metric between the treatment and control groups; it says by what percentage the change made that metric better or worse.

Uplift is the final number of any experiment, but on its own it is misleading because it states only the size of the effect, not the confidence in it. A 20% uplift on a small sample can be zero tomorrow, while a 3% uplift on a large sample is stable; this is why uplift must always be read beside a confidence interval. The difference between relative and absolute uplift matters too: a conversion rate going from 2% to 2.4% is a 0.4-point absolute uplift but a 20% relative uplift, and reporting the bigger number without saying which one inflates the picture. A negative uplift is also a valid result, not a failed experiment.