Signal Scored Account Prioritization for GTM Agencies
Built a live scoring system that reads funding, hiring, and product signals to tell reps which accounts are in a buying window right now.
situation
Reps at a GTM agency were working a flat, unprioritized list, so outreach timing was essentially random and accounts got hit before they were in a buying window. The thesis: agencies want companies big enough to afford a retainer but small enough to move fast, roughly 50 to 1,000 employees. But headcount is a bracket, not a signal. The question was what separates an account ready to buy now from one that just fits the size filter.
result
Built a system that scores accounts on live buying signals instead of static firmographics. Recent funding, weighted by recency. Job postings for GTM engineers, RevOps analysts, SDRs, and BDRs, weighted by how directly each role signals GTM investment. And a custom changelog signal that reads a company's public product updates and flags anything relevant to outbound. Scores sort into three tiers, with a no tier bucket excluded from outreach.
The changelog signal took most of the build time. Changelogs publish constantly and almost none of it matters to a GTM agency. I meta prompted until the AI could reliably answer three things: is there a changelog, does it contain a relevant change, and what specifically changed. When it hits, it hits hard. A changelog reading "now supports Shopify integration" published last week becomes an opener about slow partner adoption in the first months without a dedicated outbound motion targeting Shopify merchants. That is not a cold email. That is timing plus context.
impact / results
Replaces a rep's morning of scrolling a flat list with a queue already ranked by who is actually in a buying window. The changelog signal alone turns a cold email into a warm one grounded in something the company did last week, the kind of specificity that lifts reply rates and shortens the number of touches it takes to book a meeting.
tech stack
Clay (enrichment, scoring, tiering), Claygent (changelog reading), meta prompting (the relevance filter), Find Jobs (hiring signals).
Web scraping for changelog backfill, Crunchbase or PitchBook for cleaner funding data, Slack alerts on Tier 1 entry, CRM sync to push tiered accounts into a rep's live queue.
process map
reflections
The lesson was about process, not tooling. You can plan the logic before you build, and you should, but the actual differentiator only showed up after a lot of trial and error and sitting with the problem longer than felt comfortable. The eureka does not arrive on schedule. Sometimes the alpha is there and you find it. Sometimes it is not and you ship with what you have. The job is doing the reps that make the eureka possible, not assuming it turns up on the first attempt.