The Conversion Ratio Optimization Agent autonomously tests landing page variants, identifies which combinations of copy, layout, and calls-to-action convert best, and ships the winning variant. It runs continuous experiments against live traffic, evaluates statistical significance before declaring a winner, and documents what it learned — turning your marketing pages into compounding growth loops rather than static assets.

Every experiment is logged with its hypothesis, traffic split, conversion delta, and confidence score, giving your team a searchable record of what works and why.

Clone the repo, open it in Warp, and ask the agent to run the setup skill:

Clone https://github.com/warpdotdev/cro-agent-oss.git, open it in Warp,
and run the setup skill at .warp/skills/setup/SKILL.md

It walks through every integration interactively — collects your values, replaces all placeholders, writes .env, seeds the owner mapping, and verifies each integration before finishing.

Or set it up manually:

1. Configure secrets

Copy .env.example to .env — or configure the same values as Oz team secrets — and fill in your Notion, BigQuery, Metabase, Slack, Google Ads, and Sanity credentials.

2. Replace placeholders

Every identifier in the repo ships as a placeholder. Find what still needs swapping for your own project IDs, datasets, channels, and hosts:

grep -rl 'PLACEHOLDER' . --exclude-dir=.git

3. Set up Notion

The skills expect specific Notion databases and fields — Page Registry, Roadmap & Backlog, Knowledge Base, and Analysis Log. The full schema lives in .warp/skills/cro/references/config.md.

4. Wire Oz

Create two Oz environments to keep Slack credentials isolated — cro-silent runs research and registry sync without the Slack token, while the main environment posts to your channels. Then add the three cron schedules (daily page-registry sync, twice-daily weekday monitors, Friday recap) and the four Notion webhooks that drive the experiment loop. The full wiring guide lives in .warp/skills/cro/README.md.

Any page that drives a conversion event is a candidate for optimization. The agent works well on product landing pages, pricing pages, feature announcement pages, and campaign-specific landing pages.

Teams using the agent alongside intent-driven traffic sources — paid search, product-led referrals, or email campaigns — see compounding returns as each winning variant raises the baseline for the next round of experiments.