Build an AI Lead Research and Prospecting Agent
Build a capped evidence-first prospecting agent that researches approved public signals, preserves raw records, explains uncertainty, remembers prior events, and drafts without sending.
Define the Offer and Research Decision
Choose one offer, customer, geography, exclusion set, and decision the research should improve.
Research and Select Observable Signals
Compare source-backed event hypotheses and choose one or two signals with date and evidence rules.
Approve Sources, Caps, Cost, and Privacy
Allowlist sources and define canary size, full-run limits, cost ceiling, timeout, retries, privacy, and secrets.
Preserve Raw Evidence and Normalize Records
Create a run manifest, save immutable raw records, assign stable IDs, and preserve field-level evidence.
Deduplicate Entities, Events, and Repeat Runs
Use stable keys, retain duplicate links, separate company and event identity, and preserve review state across runs.
Classify Evidence Without Inventing Intent
Separate facts, derived values, hypotheses, recommendations, unknowns, and visible classification rules.
Build the Review Report and Grounded Drafts
Render scope, reconciliation, evidence, unknowns, safe links, exports, and editable drafts from one normalized run.
Run the Canary, Repeat, Recover, and Hand Off
Test success, stale, duplicate, uncertain, hostile, empty, partial, and failed-source behaviour before any future automation.