A full-funnel suite of propensity models for a healthcare staffing marketplace — scoring B2B facility prospects, nurse leads, and shift-level matches so sales, marketing, and ops focus on the highest-probability actions.
A healthcare staffing marketplace has to grow on three fronts at once — sign new facilities (B2B sales), convert nurse signups into active workers, and fill every posted shift. Each is a different "who is most likely to act next?" question.
Sales chased facilities by gut feel, marketing treated every nurse lead the same, and ops had no signal for which nurses would actually claim a given shift — so effort scattered across low-probability targets.
Ranked not-yet-onboarded facilities into priority quintiles from public CMS quality signals, nearby nurse supply, and cluster modeling; results auto-sync to Salesforce monthly.
Predicted which nurse signups become active users within 60 days using XGBoost, Optuna tuning, and class-imbalance handling to focus onboarding and marketing spend.
Ranked the nurses most likely to apply for each open shift using 50+ engagement, shift, and nurse-facility fit features with a strict posting-time train/test split to prevent leakage.
Served ONNX models off BigQuery feature tables, with DVC pipelines for reproducibility and MLflow for model tracking.
Client anonymized; metrics are from production models.