Optimizing add-to-cart (a proxy) instead of revenue produces directionally wrong winners in ~1 of 5 tests.

Optimizing add-to-cart (a proxy) instead of revenue produces directionally wrong winners in ~1 of 5 tests.
From a backtest where both metrics were tracked; 19% sign-flip between proxy and revenue.
Proxy metrics are higher-volume, so they hit significance faster — tempting. But a variant can lift add-to-cart while dropping average order value, netting negative.
— Proxy ▮▮▮▮▮▮ speed, ▮▮ correctness
— Revenue ▮▮ speed, ▮▮▮▮▮▮ correctness
Use proxies to guardrail, never as the primary decision metric on monetized funnels.
Read the number, not the story. [backtest · 19% flip]