A briefing for Disney Experiences
How Disney Streaming runs Databricks as the foundational data layer — with Snowflake as a first-class analyst surface and AI/ML built on top.
Begin →One foundation. Two consumption surfaces. AI on top.
Powering Disney Streaming's full AI/ML estate — model serving, agentic workflows, fraud detection, and the security capabilities behind them.
Snowflake stays first-class. Same 360 Gold data, no copy, no ETL — read directly via Iceberg.
Databricks owns the catalog of record. One copy of every 360 table. Governed end-to-end.
A Databricks proposal is underway to unify data and business logic within Disney Streaming — Experiences in early would shape Wave 1 scope.
Physical layer
360 Gold data stored once, readable from Databricks and Snowflake. No copies, no 7 PB/month round-trips.
Meaning layer
One YAML repo defines every metric. CI/CD fans out to UC Metric Views, Snowflake Semantic Views, Looker, Tableau.
Why it matters for Experiences: subscriber churn, LTV, DAU, attendance, dwell — defined once, not five different ways. Finance, BI, and AI agents all read the same number.
Iceberg and a unified semantic layer power the complete ecosystem — no lock-in, no rewrites, every tool a first-class consumer.
Last 12 months on the hybrid pattern.
Active and in-flight Databricks AI/ML workloads at Disney Streaming, organized by business outcome.
Identity & Customer 360 · Household & Device Graph · Audience Profiling · Audience Segmentation
Campaign Planning & Optimization · Media Mix Modeling · Market Research & Insights · Lookalike Audience Creation · Subscriber Lifecycle (churn, win-backs) · Lifecycle Marketing · Next Best Action
Search & Discovery · Content Recommendation · Real-time Personalization · Audience Activation · Engagement Tracking
Subscription Pricing & Packaging · Upsell / Cross-sell & Bundles · Subscriber Acquisition Cost & Playback · Subscriber LTV & Margin Optimization
Ad Sales Pipeline & Deal Management · Inventory Forecasting & Packaging · Rate Cards, Discounts & Deal Governance · Direct / Programmatic Channel Mix · AI Agent-Assisted Media Selling
Pricing & Yield Management · Ad Decisioning & Pacing · Contextual & Audience Targeting · AI Ad Creative Generation · Ad Experience Personalization · Measurement & Attribution · Incrementality Testing & Lift Measurement · Brand Safety & Suitability Controls
Real-time inference. Governed feature store. On-data AI.
Streaming inference flags chargeback risk in-flight. Cut fraud rate from 18 bps to 10 bps — well below industry average.
Custom models developed on Databricks powering personalized search with relevant context and recommendations.
ML-routed retry paths on payment failures. Reduces involuntary churn at the subscription gateway.
Custom models surface guest-specific details at every customer service touchpoint.
Unified subscriber data powers churn, LTV, and segmentation models — lakehouse-native, no out-of-band copies.
Recommendation and ranking models trained on Databricks serverless GPU, directly on lakehouse data — no extra ETL to a separate training cluster.
Topics that map directly to what your team has already flagged.
Next step — a working session to map one specific Experiences use case onto this foundation.
Let's talk →