CRO Prospecting OS — An Agent Pipeline for Evidence-Backed Outreach
Finding potential clients for conversion rate optimization services requires more than collecting store URLs or identifying brands running ads. A useful prospect needs to match the target market, have an active acquisition channel, and present a credible opportunity for a relevant sales conversation. I built CRO Prospecting OS to organize that process from initial discovery to a personalized outreach package. The system uses seven specialized Claude Code agents to research advertisers in the Meta Ad Library, evaluate business fit, review mobile shopping experiences, verify supporting evidence, identify potential contacts, and prepare client-facing reports. Each prospect moves through a defined sequence of stages, with Python-based controls determining when it's allowed to progress. Instead of generating generic CRO recommendations, the system examines the journey from a live advertisement to its landing page or product page. Potential issues are documented with screenshots and reviewed independently before being included in a report. Qualified prospects can receive a PDF review and a personalized outreach draft, but nothing is sent automatically. Every package requires human review, and all communication is handled manually. The workspace has processed a real batch of 56 prospects. The experience also exposed weaknesses in earlier quality checks, leading to a stricter approval gate that prevents packages from advancing without the required validation.
Finding the right CRO prospects was only half the work. Researching them properly took even longer.
Selling conversion rate optimization services requires a reason to start a conversation. Simply finding an e-commerce brand and sending a message about improving conversions isn't enough. The brand needs to be a reasonable fit, and the outreach should demonstrate that someone has actually looked at its business. For skincare brands running paid advertising, that means checking whether they're actively advertising, understanding their products and positioning, and reviewing what happens when someone clicks an ad and reaches the store. Doing all of this manually for every prospect created several problems: - Finding and evaluating suitable brands involved repetitive research across multiple websites. - Reviewing ad-to-landing-page journeys required switching between ads, mobile pages, product pages, and supporting evidence. - Preparing personalized reports and messages took additional time after the research was complete. - Research quality could vary between prospects, especially when findings were not documented consistently. Using AI without clear restrictions created a different risk: unsupported CRO claims, assumed performance problems, and findings that hadn't been verified against the actual page. I needed a workflow that could handle repetitive research without sacrificing accuracy or control.
AI could handle the research, but important decisions needed rules it couldn't skip.
The first challenge was separating two questions that are often mixed together. Is this brand a good potential client? And is there enough verified evidence to make a useful, specific observation about its website? A company can be a good commercial fit without having an obvious CRO issue worth discussing. Likewise, finding a website issue doesn't automatically make a brand a suitable prospect. I designed a 100-point qualification score focused only on business fit, with unknown information explicitly marked as unknown rather than guessed. Separately, the system evaluates the strength of the available outreach angle based on actual findings. I also separated prospect discovery from detailed CRO reviews, allowing the workflow to qualify brands before committing time to more thorough analysis. The next decision was to stop relying on AI instructions alone for critical approval steps. Independent QA agents verify findings and contact information, while Python code enforces state transitions, approval requirements, and file integrity checks. That combination gives AI room to research and draft, without allowing it to decide on its own that incomplete or unsupported work is ready to use.
Seven specialized AI agents, one controlled pipeline, and evidence checks before every outreach package.
Find better prospects. Start conversations with real evidence.
We build AI-powered prospecting workflows that help B2B teams research potential clients, identify relevant opportunities, and prepare personalized outreach without relying on generic messages or unsupported claims.
Discuss Your Project