Case Study: T‑Mobile achieves real-time, scalable ad-fraud detection and 95% faster code performance with Databricks

Protecting mobile users from fraud with behavioral AI

T‑Mobile’s Marketing Solutions team faced rising advertising fraud across its mobile network and needed an end‑to‑end way to detect diverse fraud types in real time. The challenge was to analyze billions of ad transactions and up to ~10 TB of network data daily, but an initial on‑prem Apache Spark build proved complex, costly to scale, and distracted the team with heavy DevOps work.

T‑Mobile migrated to Azure Databricks, using Spark, MLflow and a proprietary Normalized Entropy algorithm to score behavioral anomalies, plus a streamlined dashboard for investigation and reporting. The cloud platform eliminated infrastructure overhead, cut a routine job from eight minutes to 23 seconds (≈95% faster), enabled real‑time fraud detection at scale, and gave marketing teams actionable risk insights to improve ad placement and mitigation.

T-Mobile

Eric Yatskowitz

Data Scientist