Agent #002 · Research & Market Intelligence
Buying Committee Mapper: governance & failure modes
What this agent is built not to do, and the specific ways it can fail if misconfigured.
Governance notes
- Champion and Blocker are never confirmed by the workflow itself — only surfaced as candidates for a human to confirm in HubSpot.
- Sourced-but-unconfirmed contacts are never written to CRM contact records directly; they route to a review queue instead.
- Every classification stores a config version and a plain-language basis (which rule or behavioral signal fired) for auditability.
- The agent does not contact anyone directly — it surfaces coverage gaps and lets a human decide how to fill them.
Failure modes
- Sparse contact data: an account with only one or two known contacts can’t be meaningfully mapped — flagged for manual contact sourcing (Agent #005) rather than forced into a classification.
- Rule-only classification without engagement data produces lower-confidence results, written as such.
- Loose config symptom: if influencer classifications exceed roughly 40% of contacts, that indicates the role config is too broad and needs retuning.
- Role drift: classification re-runs on a schedule rather than treating a role as a permanent, one-time label.