Marketing Personalization · Real-time Analytics
Second Dinner relied on manual, heuristic processes for daily shop personalization, so feature iterations took about a month and recommendations lagged behind fast-changing player interests.
Reported impact
Second Dinner moved from only 10 experiments a year to weekly iterations, sped feature launches 400%, cut model development from months to weeks and now delivers personalized shop offers and bundle recommendations that better match player behavior in real time.
Second Dinner centralized ingest, processing, warehousing, analytics, dashboarding and machine learning on the Databricks Data + AI Platform so a small team could turn manual shop-personalization workflows into continuously updated recommendations. Player engagement and purchase data are ingested through Delta Lake and used to train ML models in MLflow, with Python logic added through PyFunk to speed experimentation across the data team. The first model produced daily population-level recommendations, then evolved into batch precomputed personalization and real-time recommendations delivered through a low-latency model-serving endpoint. Models are uploaded to the registry, retrained weekly and fed with signals about what players see, when they play and what cards they prefer, allowing the Daily Offer Shop and new bundle triggers to react to recent behavior instead of stale aggregate statistics.
Player engagement and purchase signals land in Delta Lake on Databricks, where MLflow-managed models analyze what players see, update or buy. Recommendation outputs are then served back into Marvel Snap through the Daily Offer Shop, precomputed bundles and real-time next-best-action triggers.
Despite utilizing the Databricks Data + AI Platform for various functions — ingesting, processing, warehousing, analytics, dashboarding and ML — they faced challenges in advancing player personalization for their daily shop experience.
Leveraging these new capabilities in Databricks, Second Dinner has shifted from launching only 10 experiments a year to running them weekly.
The manual steps needed to interpret visualizations and continuously push code created numerous opportunities for errors. A single feature iteration could take a month, limiting us to 10 use cases yearly.
The model adapts to player behavior and automatically updates in-game.
The teams build and ingest customer engagement data through Delta Lake to feed their models.
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
Databricks Sample Jul 7, 2026 Bally's Bally's needed a better way to support omnichannel gaming personalization, compliant analytics, and faster market launches because legacy technology fragmented customer and operational data, which slowed… 10x serverless compute Databricks Data + AI Platform · Delta Lake · Lakeflow Jobs Databricks Sample Jul 10, 2025 SRF SRF needed a better way to support audience intelligence, recommendation, and editorial analytics for public media because digital consumption was rising fast while siloed metrics and custom KPI pipelines kept editorial… 1.3M of data ingested and processed each day Databricks Data + AI Platform · Delta Lake · Genie Databricks Sample Jan 14, 2025 Seven West Media Legacy systems made it difficult to access real-time insights, slowing down decision-making and leaving some revenue opportunities on the table. Workflows that once took hours of manual effort are now possible in… $16M Incremental ad revenue from re-engagement campaigns Agent Bricks 