Lead Score ML Model
A production machine learning system that ingests CRM data and applies a gradient boosting model to assign conversion probability scores to every sales lead.
The Challenge
Our client had a pipeline of 3,000+ leads and no systematic way to prioritize. Sales reps spent equal time on low-quality inbound leads and genuine high-intent prospects, resulting in inconsistent win rates.
Core Offerings
A complete ML pipeline covering data ingestion, model training, deployment, and real-time inference.
Feature Engineering
Extracted 40+ behavioral, firmographic, and engagement features from HubSpot and Clearbit.
Model Calibration
Tuned an XGBoost model optimized via Optuna with SMOTE oversampling for class imbalances.
Real-time Inference API
Serverless FastAPI deployment for live scoring, alongside nightly pipeline bulk re-scoring.
Sales Dashboard
Custom React dashboard surfacing probabilities, ranked lists, and per-lead SHAP explanations.
Technical Architecture
The Impact
"Achieved a 32% improvement in win rate for highly scored leads and an 18% increase in average deal size by focusing reps on better-fit accounts."