Left: clicks & budget before a kill is statistically safe. Right: feed live numbers for a KILL / SCALE / KEEP TESTING verdict.
| Conf. | Clicks | Budget |
|---|
Beta(conv + 0.5, clicks − conv + 0.5)) and works out how sure it is which side of break-even you're really on.
Your observed CVR can already sit under break-even while the sample is still too thin to trust — the confidence number is how sure the math is it'll stay there, which is what you actually act on.
≥ your confidence it's a loser → KILL; ≥ your confidence it's a winner → SCALE; neither → KEEP TESTING.