Predictive Account Scoring
Produces a hybrid account score combining intent signals, firmographic fit, and LLM reasoning.
What it is
High fit and high intent rarely peak at the same time. This agent fuses both into a single calibrated score by combining a trained gradient boosting model over firmographic fit with the composite intent vector from the aggregator.
How it works
Enriched account records and intent vectors are passed to an XGBoost classifier trained on historical CRM data. The model produces a fit-probability score, which is fused with the intent score using a configurable weighting formula. An LLM reasoning layer reviews edge cases near scoring thresholds and applies qualitative adjustments the statistical model cannot capture. The final score is normalized, confidence-bounded, and written to the account record.
How it fits into the pipeline
Decision engine of the prioritization layer. Consumes firmographic records and intent scores, and produces the ranked account list that feeds the Ranking and Explainability Layer and the Account Prioritization Agent.