Predictive Churn & Retention Dashboard
For my Data Science Master's capstone, my team partnered with Ora Fashion to turn churn probabilities into four region-specific retention playbooks for the European market.
Ora Fashion, an online retail brand, was losing customers and didn't know who was leaving, when the drop-off happened, or how to prevent it — a classic case of rising acquisition costs with no visibility into customer lifetime value. The mission: move the company from retroactive reporting to proactive customer lifecycle management.
Explore the Power BI dashboard
Having trouble viewing it? Open the dashboard directly.
From churn probabilities to a marketing playbook
Framing it as CRISP-DM
Anchored the project in the CRISP-DM methodology to keep every technical step tied to the CMO's actual objectives, rather than turning it into an academic exercise.
Building the data foundation
Built an ETL pipeline in Power Query to clean and merge three raw sources — orders, customers, and a field dictionary — then architected a star schema in Power BI: a central sales fact table, customer and product dimension tables, and a dynamic calendar for time-intelligence. My teammate used this same foundation to train the churn-probability model in Python (logistic regression and XGBoost).
Engineering dynamic RFM segmentation
Built Recency, Frequency, and Monetary scoring directly in DAX — scanning the sales fact table to score every customer without hardcoding thresholds — so the marketing team could instantly separate "Champions" from customers genuinely at risk of churning.
Mapping the drop-off curve
Built a cohort analysis matrix tracking return rate by acquisition month. It confirmed that the first two to three months after a purchase are the critical retention window — the mathematical justification the executive team needed to invest in onboarding.
Designing the geo-strategic action plan
Cross-referenced demographics, spending behavior, and geography to split the European customer base into four clusters, each with a tailored retention play rather than one generic campaign.
One size does not fit the European market
- Italy — the high-frequency youth. Younger, smaller basket size, but frequent buyers. Play: gamification, loyalty programs, and exclusive community events.
- UK & Spain — the high-ticket veterans. Older, high spend, low frequency — the "whales." Play: premium offers, targeted up-selling, and win-back campaigns the moment the churn model flags risk.
- France & Germany — the steady core. Moderate spend and frequency, with untapped potential. Play: personalized recommendations, subscription models, and time-limited flash sales.
- Greece & Netherlands — the re-engagement zone. Older, low spend, low frequency, prime churn candidates. Play: reactivation campaigns, discount bundles, and educational content to rebuild trust.
Replacing a one-size-fits-all European retention strategy with four targeted playbooks gave Ora Fashion's CMO a data-backed way to spend the marketing budget instead of burning it on broad campaigns — turning a machine learning model into a scalable revenue-preservation engine.