Real-time Lead Scoring for CRM

Gradient boosted models scoring leads in real time; uplift +18% in SQLs using CRM events and marketing touchpoints.

View Code

Overview

Built an end-to-end real-time lead scoring system that processes CRM events and marketing touchpoints to predict lead quality. The system uses XGBoost models with feature engineering pipelines that capture behavioral patterns, engagement metrics, and demographic signals. Deployed with streaming architecture for sub-100ms inference latency.

Code Highlight

Real-time Feature Engineering Pipeline

import pandas as pd
from kafka import KafkaConsumer
import xgboost as xgb
from redis import Redis

class LeadScoringPipeline:
    def __init__(self):
        self.model = xgb.XGBClassifier()
        self.redis_client = Redis(host='localhost', port=6379)
        self.consumer = KafkaConsumer('lead_events')

    def engineer_features(self, lead_data):
        """Extract behavioral and engagement features"""
        features = {
            'page_views_last_7d': self.get_pageviews(lead_data['email']),
            'email_opens_last_30d': self.get_email_engagement(lead_data['email']),
            'company_size_score': self.company_size_mapping(lead_data['company_size']),
            'behavioral_score': self.calculate_behavior_score(lead_data),
        }
        return pd.DataFrame([features])

    def predict_lead_score(self, lead_data):
        features = self.engineer_features(lead_data)
        score = self.model.predict_proba(features)[0][1]

        # Cache result for 1 hour
        self.redis_client.setex(
            f"lead_score:{lead_data['email']}",
            3600,
            score
        )
        return score

Key Results

+18% increase in SQL conversion rate

Sub-100ms inference latency

99.9% uptime in production

Reduced manual lead qualification by 60%

Technologies Used

Python

XGBoost

Apache Kafka

Redis

PostgreSQL

FastAPI

Project Category

ai automation

Repository

View on GitHub