score-inbound-leads
Score inbound leads
The insight: a score nobody can explain is a score nobody trusts — and a score on a contact who isn't reachable is worse than no score, because it spends a rep's morning on a ghost. The naive version multiplies magic constants into a number; a rep sees "72", disagrees once, and the model is dead. Scoring here is four commitments: every point traces to a quoted field value (evidence or it didn't happen); missing data scores as UNKNOWN with a stated policy — never a silent zero, because "we couldn't see their headcount" and "they're the wrong size" must never look the same; weights are a config the user approves, not numbers you invented; and validity gates the score — a tier is surfaced only for a contact who is demonstrably reachable and current. Scoring is deterministic arithmetic — code, never an LLM (an LLM adds variance to math a formula does for free), and the same input must produce the same tiers on every re-run.