Case Study: Comcast achieves 10x compute cost reduction and faster AI-powered viewer experiences with Databricks

Transforming home entertainment with voice, data and AI

Comcast, a global media and technology company serving millions of customers, struggled to turn billions of daily events and petabytes of telemetry from 20+ million voice remotes into reliable, real-time insights. Fragile pipelines, small-file ingestion issues, dispersed data scientists using different languages, and the manual effort of developing, training, and deploying hundreds of ML models across cloud, on‑prem and device environments prevented timely innovation in voice-driven viewer experiences.

By adopting the Databricks Lakehouse—Delta Lake for ingest and ETL, MLflow (with Kubeflow) for model lifecycle, automated cluster management, and collaborative notebooks—Comcast modernized pipelines, simplified infrastructure, and sped model deployment. The result: a 10x reduction in compute (640→64 machines), a ~90% drop in DevOps effort for onboarding, model deployments cut from weeks to minutes, improved data science productivity, and an award‑winning, AI-powered voice experience for viewers.

Comcast

Jan Neumann

VP Machine Learning