A single-dimensional score — firmographic fit, or behavioral engagement, or intent data alone — misses the compound signal that separates in-market accounts from vaguely interested ones. The agent configures scoring models that weight multiple signal categories: firmographic fit (industry, company size, geography against ICP), technographic profile (stack signals indicating category readiness or competitive displacement opportunity), behavioral engagement (email open sequences, website pages visited and in what order, webinar attendance, content downloads), and intent signals from third-party providers (keyword cluster activity, review site visits, competitor comparison behavior). Weights are set from historical win/loss data rather than intuition — the model reflects what actually predicted a closed deal, not what someone guessed would matter.

A static score assigned at the moment of first contact becomes wrong within days — the lead's behavior changes, their intent signals evolve, their company posts a funding announcement, and none of it updates the number sitting in the CRM field. The agent maintains scores dynamically: recalculating each lead and account's score as new behavioral signals arrive (email click, new page visit, demo request), as enrichment data refreshes (new technographic signal detected, job change at the account), and as third-party intent data updates. Scores decay when engagement goes cold, preventing stale high scores from clogging the top of sales queues with leads that last showed interest six months ago. Every score reflects the current state of the lead, not their state at first contact.

A scoring model is only credible if it's validated against what actually happened. The agent runs ongoing calibration analyses comparing model predictions against closed-won and closed-lost outcomes: are the accounts that converted to revenue scoring above threshold at the right rates? Are high scores correlating with pipeline conversion? Is a particular signal category over-weighted (lots of points for intent but no conversion lift from intent signals in isolation)? Calibration reports surface model drift — when the correlation between score and conversion starts to weaken — and recommend weight adjustments with supporting data. The model evolves with the business rather than calcifying at the configuration it had when someone first set it up.

A score that lives in a CRM field but doesn't trigger action is just a number. The agent connects scoring to routing and alerting logic: when a lead crosses the MQL threshold, it routes to the assigned SDR queue with the score breakdown showing which signals triggered the threshold. When a dormant account's score spikes — intent surge or a new contact behavior pattern — an alert fires to the AE or CSM with context on what changed. When a whole territory sees an unusual score spike pattern, a manager alert surfaces the trend before the quarterly review. The scoring model becomes an operational system, not a reporting artifact.