Same-Day Vendor Sweep
148 vendors became a segmented TheoPay rollout dataset in one workday, because the schema was written by the implementation lead, not for the implementation lead.
A 148-vendor onboarding list landed for a newly acquired Idaho customer's TheoPay implementation: two columns, vendor name and URL. The boss-side estimate for enriching it (cart type, product mix, Mastercard acceptance) was days to a week of manual sweep, and that estimate was correct for the traditional motion. The trap is what 'enriched' means. Raw vendor facts are not a sequencing decision, and an implementation team cannot prioritize 148 vendors alphabetically. Crawl for surface facts and hand off raw data, and you have moved the bottleneck from lookup to translation instead of removing it.
The move was to treat the schema as the load-bearing piece, not the crawl. Defined a Firecrawl MCP schema written by someone who would have to sequence the rollout: cart structure, product mix taxonomy, Mastercard acceptance as a discrete field (because that is how the eligibility-logic team makes the call), free-text catalog notes for what will not fit a controlled vocabulary. Shared the CSV into Claude Code as orchestration surface, ran the schema across 148 sites, and wrote seven fields per row back to the same CSV in a format the implementation team could act on directly. Did not stop at 'data is enriched.' Translated the output into a segmented TheoPay rollout plan: integration-complexity buckets, product-mix splits for the eligibility-logic team, Mastercard fanout for targeted follow-up. Anyone with Claude Code and Firecrawl MCP can crawl 148 sites; what makes this artifact different is that the fields were chosen by someone who has to sequence the rollout.
A 148-row enriched CSV in a 7-field schema spanning cart structure, product mix taxonomy, Mastercard acceptance, and free-text catalog notes, with a segmented TheoPay rollout plan layered on top. Cart-structure split: 98 vendors fit standard cart-detection default integration, 50 need rework (22 inquiry/booking, 22 subscription signup, 6 external link). Product-mix split: 39 physical only, 42 service only, 67 both. Mastercard fanout: 49 verified, 4 likely, 95 unverified, which converts a 148-vendor sweep into a 95-vendor targeted follow-up at the login-gated-checkout boundary. The free-text product_mix_notes column carries the vendor-specific context that would otherwise have been collected by clicking through 148 sites by hand.
Without it, an implementation team prioritizing 148 vendors alphabetically because the integration-complexity signal cost a week to collect. With it, a 95-vendor targeted Mastercard follow-up surfaced before integration begins, sequencing driven by integration complexity, and the same schema reusable for the next state customer's vendor list.
The schema is the work. The crawl is the easy part.