Company signals architecture

The problem

The scoring model leaned on firmographics. Fit tells you who could buy, not who might buy now. We needed timely events, like hires, funding and office moves, to show up in the score fast enough to act on.

What I built

A pipeline that watches news and other sources for eight signal types, including ICP hires, executive hires, job posts, funding, M&A, and office openings and closings.

The flow: news sources, then an n8n workflow, then a Claude evaluation, then a staging sheet, then Clay enrichment, then the CRM. AI-generated account narratives are written back to the CRM so reps see why an account scored.

The combined score runs from -200 to +200 and blends firmographic fit, signals, web and form events, past SQL history and intent data. A single cluster of signals can move an account by 33 points.

Design decisions

• Signals and fit carry equal weight. Neither alone is enough.
• An AI evaluation step filters noisy news before anything touches the score.
• A staging step sits between detection and the CRM, so bad signals can be caught before they change a score.
• Narratives go into the CRM, where reps already work.

Result

The chain runs end to end without manual steps. Signals fire, the score moves, the velocity tracker catches the change, and sales gets an alert. One cluster of an office opening, an ICP hire and funding produces a +33 point swing.