Skip to content

CRO Delivery

Owner: Bryan ยท Scope: how we run CRO rounds on a client hub. The method itself is public and canonical; this page is the internal delivery mechanics.

The method is public

Do not duplicate the methodology here. Read and link to the canonical, client-facing versions:

  • CRO Methodology: the operating model (investigation, research, optimization, evaluation).
  • CRO Process: evidence to shipped test to measured decision.

Client hubs link clients to these same public pages (from each hub's Resources / Links page). If you change the method, change it there.

Running a round on a client hub

Each round lives in the client repo under docs/cro/round-<n>/:

  1. Create docs/cro/round-<n>/ with an index page (schedule + hypotheses) and, for multi-surface rounds, one page per surface (Homepage Banner, PDP Redesign, etc.).
  2. Add the round to the nav in mkdocs.yml and a linked row in cro/index.md.
  3. Keep results tables as TBD until the test has actually run. Do not declare a winner from a short sale window, a campaign traffic shift, or a partial week unless the caveat is explicit.

Test vs direct rollout

  • A/B test when there is enough traffic, a clear control, a reversible change, and uncertain impact.
  • Direct rollout when the fix is required, traffic is too low to test, or an approved brand update needs to ship. Record it with the Direct Rollout Measurement Plan (Templates).
  • Research only when evidence is weak or the cost is not yet justified.

CRO design assets

  • Raw and reference files (screenshots, heatmaps, test exports) live in the client's team-data/ (private repo source).
  • If a screenshot needs to appear on the site, copy it into the relevant docs/cro/round-<n>/assets/ folder with a clean filename, compress to WebP, max 1600px, then link it. Keep originals in team-data/.
  • The .ab-compare / .ab-shot / .ab-mobile-shot CSS for side-by-side comparisons exists in every client repo and the template.

Recording a result

Use the Test Summary template. Record: dates, page/flow, primary metric, guard-rail metrics, control vs variant, result, decision, next action, and any external noise (sale, spike, outage, campaign change, stock issue).