Agent #001 · Research & Market Intelligence
Target Account Scanner: governance & failure modes
What this agent is built not to do, and the specific ways it can fail if misconfigured.
Governance notes
- Suppression runs before scoring reaches CRM: accounts already tagged customer or with an open deal are never written, regardless of score (gate_result: skipped_customer / skipped_open_deal).
- Writes land only in agent-owned HubSpot properties — never an existing lead-scoring or lifecycle-stage field.
- Every scored record carries a model/config version and a scan date for auditability.
- The workflow does not create CRM contacts. Contact-level sourcing is a separate, reviewed process (Agent #005).
Failure modes
- Missing firmographic data: nulls are scored neutral where the config defines a neutral default (e.g. unknown revenue), not zero — this avoids penalizing gaps in enrichment as if they were poor fit.
- Stale or missing intent: no intent data scores as "none," not as a false positive.
- Duplicate accounts: rows are deduplicated by domain before scoring.
- Runaway Target band: if more than roughly a quarter of the scanned universe lands in Target, that signals the config is too loose and should be retuned before the list is trusted for outbound.