Key results
model accuracy in identifying customers at risk of churn
drop in churn among mid- and high-value accounts within 6 months
ROI from data-driven retention campaigns
increase in sales team productivity through prioritized outreach
About the Client
A U.S.-based national distributor of rigid packaging solutions operating through a network of regional brands. The company serves clients across industrial, life sciences, food & beverage, and consumer goods sectors. Following rapid expansion through acquisitions, the client faced challenges with fragmented customer data and limited visibility into churn patterns.
Key Challenges
Fragmented customer data across merged ERP and CRM systems
Inability to detect early warning signs of customer disengagement
Revenue leakage from mid-value client attrition going unnoticed
Sales reps lacked clear prioritization to intervene with at-risk accounts
Our Approach
Centralizing Customer Data
- Integrated data from ERP, CRM, order management, and support systems into Azure
- Used Azure Data Lake Storage for raw data staging and Snowflake for data warehousing
- Created customer-level data marts using dbt
Feature Engineering & Churn Definition
- Defined churn as no orders in the last 120 days
- Engineered features including RFM scores, support interactions, and order velocity
- Segmented customers by geography, product line, and account tier
Predictive Churn Modeling
- Built an XGBoost model in Python, hosted on Azure Virtual Machines
- Evaluated model with precision, recall, and AUC to ensure business relevance
- Automated weekly churn scoring using Azure Functions
Business Enablement
- Developed Tableau dashboards tailored for sales and regional teams
- Embedded churn scores and risk flags into CRM workflows
- Trained account managers to act on churn indicators effectively
Implemented Solution
Azure-Based Data Platform
- Azure Data Lake Storage for raw data landing and backup
- Azure Data Factory (ADF) for schema standardization and ETL
- Snowflake as the central data warehouse
Data Transformation & Modeling
- Modular transformations using dbt for auditability
- Built customer history snapshots for churn trend analysis
- Created analytics-ready data marts for ML and reporting
Python-Based ML Pipeline
- Churn model developed using XGBoost, scikit-learn, and pandas
- Hosted on Azure Virtual Machines, with weekly scoring via Azure Functions
- Scores stored in Snowflake for downstream consumption
Dashboarding & Alerts
- Tableau dashboards connected to Snowflake via an extract connection
- Views segmented by region, rep, and product category
- Churn alerts pushed to CRM and email to enable proactive outreach
Technologies Used
