CRO is the discipline, not the metric. The metric is conversion rate (CVR); CRO is the work an operator does to move it. The loop is consistent across teams: instrument the funnel, identify where sessions are leaking, form a hypothesis about why, test a change against a control, and ship or kill based on the read. The substrate on a DTC storefront is the canonical funnel — landing → PDP → cart → checkout → confirmation — with a stage CVR computed at each transition so the leak is locatable before any change is tested.
Quantitative inputs (funnel reports, segment breakdowns) tell an operator where sessions drop; qualitative inputs (session replay, on-site surveys, heatmaps) suggest why. Both feed the hypothesis. Tests are judged against statistical significance rather than eyeballed — the mechanics of significance, minimum detectable effect, and peeking belong to the A/B test discipline. Common levers cluster in three places: PDP (layout, imagery, social proof and reviews placement), cart and checkout friction (guest checkout, address autocomplete, Apple Pay and Shop Pay), and post-purchase (thank-you-page upsell, subscription opt-ins).
Two pitfalls worth naming. A stage-CVR lift does not guarantee a revenue lift: a checkout-friction removal can disproportionately convert low-basket sessions and quietly regress AOV, so the primary metric is usually revenue per session rather than stage CVR alone. And sample pollution from concurrent promos, paid-traffic mix shifts, or sale weekends compromises the read — tests should run outside those windows unless they are the test condition. Positioned correctly, CRO is complementary to acquisition: lifting CVR on the same paid traffic lowers effective CAC arithmetically, without buying another impression.