The lift you ship is systematically smaller than the lift you measured — expect ~20-30% shrinkage.

The lift you ship is systematically smaller than the lift you measured — expect ~20-30% shrinkage.
Winner's curse: you only roll out tests that cleared significance, and that selection biases the kept estimate upward.
— So portfolio-wide, your annual "sum of wins" overstates real gains.
— Fix: apply shrinkage to reported effects, validate big winners with a holdback, and forecast roadmap value at the discounted estimate.
Read the number, not the story. [shipped lift −20-30% vs measured]