← Back to Episodes

You asked

Why do experiment rollouts look better than reality?

Results look strong during rollout but don’t generalize.

Symptom

At 78%, the experiment win rate is well above industry average.

Cause

The mechanism is audience filtering at the exposure stage.

Impact

The economic cost of selection bias accumulates in the gap between what the test predicted and what the rollout delivered.

Full diagnostic context

Results look strong during rollout but don’t generalize.

At 78%, the experiment win rate is well above industry average.

The mechanism is audience filtering at the exposure stage.

The economic cost of selection bias accumulates in the gap between what the test predicted and what the rollout delivered.