AI-Powered Personalized Content Recommendations for Sports and Entertainment Venue

golden_state_warriors · Media & Entertainment · 2023 · source Agentic: no · promotional bias: low · source-backed · extracted with prompt 0.5.4 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: Sports and entertainment fan personalization

The Golden State Warriors needed to deliver the right game, event, retail, and venue content to each fan across mobile, web, CRM, and future arena surfaces despite a large content fire hose and a lean non-technology organization.

Constraints:

Architecture

The Warriors and SADA built a Google Cloud recommendation system using BigQuery, Dataflow, a discovery/recommendation engine, content-management feeds, and a GSW API abstraction layer. Raw user events flow into BigQuery, Dataflow transforms them into the discovery engine, content generated in CMS pipelines also feeds the discovery engine, and the API exposes recommendations to mobile, web, CRM/email, and potential arena surfaces such as screens or scoreboards. The build-once/deploy-everywhere API lets a lean sports organization focus on customer experience while Google Cloud and SADA operate the heavy personalization infrastructure.

Diagram confidence: explicit (topology described by the source).

flowchart LR
  fan-event-analytics[fan-event-analytics] --> bigquery-warehouse[bigquery-warehouse]
  bigquery-warehouse --> dataflow-transform[dataflow-transform]
  dataflow-transform --> discovery-engine[discovery-engine]
  content-management-pipeline[content-management-pipeline] --> discovery-engine
  discovery-engine --> gsw-api[gsw-api]
  gsw-api --> mobile-web-crm-surfaces[mobile-web-crm-surfaces]
Drag to pan · scroll or use +/− to zoom · reconstructed from the source's described component topology

Components

Component Role Interacts with
fan-event-analytics Raw user events and behavior from digital fan interactions. bigquery-warehouse
bigquery-warehouse Stores current and historical analytics data for recommendation inputs. fan-event-analytics, dataflow-transform
dataflow-transform Transforms captured events into discovery-engine-ready data. bigquery-warehouse, discovery-engine
content-management-pipeline Moves previews, recaps, galleries, videos, event, and retail content toward the discovery engine. discovery-engine
discovery-engine Recommendation engine powered by Google Cloud/Vertex AI capabilities. dataflow-transform, content-management-pipeline, gsw-api
gsw-api Build-once abstraction layer that exposes recommendations to downstream products. discovery-engine, mobile-web-crm-surfaces
mobile-web-crm-surfaces Warriors app, ChaseCenter.com, email/CRM, and possible arena screens where recommendations appear. gsw-api

Data flow

Fan behavior and raw user events are captured in BigQuery. Dataflow transforms those events into the discovery engine. Content from the Warriors' content-management systems also moves through a content pipeline into the discovery engine. The GSW API abstracts the discovery engine output and serves personalized recommendations to mobile, web, CRM/email, and potentially venue screens, scoreboards, or other downstream fan-experience surfaces.

4 supporting quotes
“through big query all the raw user events that we talked about are being captured”transcript, stated
“we have the data flow which is taking all the events uh and then transforming all the events into the discovery engine”transcript, stated
“the content which is being uh generated through our content Management Systems”transcript, stated
“this API is what enables us to take the hard work that the the discovery engine is doing and then deploy that to many different products”transcript, stated

Why this architecture

The Warriors chose managed Google Cloud primitives and an API abstraction because they needed personalization at global-brand scale without building a large internal infrastructure team. BigQuery gave them historical behavioral data, Dataflow moved events into the discovery engine, and the GSW API let the same recommendation work appear across mobile, web, CRM, email, and future venue surfaces.

3 supporting quotes
“we are a global brand uh people know the Golden State Warriors all over the world but the reality is that we're a little less than 600 people”transcript, stated
“Big query has been a massive piece to the success of this project because one we've used it for so long”transcript, stated
“one of the big things that we stress uh quite a bit is that this build once deploy everywhere approach”transcript, stated

Models used: Vertex AI · Google Cloud Discovery Engine · BigQuery · Dataflow

Scale

Alternatives considered (roads not taken)

technology for technology's sake

Rejected because: The Warriors wanted technology to enable business and fan-experience goals, not chase buzzwords.

“using technology as an enabler for the our business not the reason why”transcript, stated
“it's just easy to fall into of a trap a little bit and just do all these different things”transcript, stated

separate recommendation integrations per channel

Rejected because: The API abstraction supports a build-once/deploy-everywhere strategy across many products.

“this build once deploy everywhere approach and so this API is what enables us”transcript, stated

building deep cloud infrastructure expertise in-house

Rejected because: The Warriors identify as a basketball team with limited resources and partnered with Google Cloud/SADA for infrastructure expertise.

“we're also not a tech company we're a basketball team like at the end of the day our Core Business is being a basketball team”transcript, stated
“we are smart enough to know that we are the experts in a lot of areas”transcript, stated

Patterns

retrieval-ranking (retrieval)

The recommendation engine ranks content and events against fan behavior and interests.

“can we do a good enough job of recommending ing what you think you may be interested next”transcript, stated

batch-offline-processing (data-pipeline)

BigQuery and Dataflow move raw event analytics into the discovery engine for recommendations.

“we have the data flow which is taking all the events uh and then transforming all the events into the discovery engine”transcript, stated

tool-use-orchestration (integration)

The GSW API abstracts discovery-engine outputs so multiple products can consume recommendation data.

“the API is really doing a lot of the hard work of enabling us to then take this data that's coming out of the discovery engine”transcript, stated

Tradeoffs

Use a build-once API abstraction.

Chose GSW API over direct per-product discovery integrations over separate integrations for mobile, web, CRM, and arena surfaces.

Rationale: One API lets the Warriors reuse recommendation outputs wherever the fan experience needs them.

“this build once deploy everywhere approach and so this API is what enables us to take the hard work”transcript, stated
“mobile web and CRM being the three that we've highlighted here but this could be many more in the future”transcript, stated

Partner for cloud infrastructure.

Chose Google Cloud and SADA expertise over building every cloud capability internally.

Rationale: The Warriors wanted to focus limited staff on fan experience rather than cloud infrastructure.

“we are not experts in is cloud infrastructure right and compute and all these different things like that's not what we are good at”transcript, stated
“we needed to partner with someone um who could help us do that and help us build it from scratch”transcript, stated

Gotchas (what bit them)

A sports organization can drown in its own content fire hose.

Root cause: Every game and event creates previews, recaps, galleries, videos, and related content.

Resolution: Use behavior and content signals to recommend the right content at the right time and device.

“there's a lot of content fire hose exactly that times 82 plus playoffs plus preseason it's a lot of content that gets created”transcript, stated
“getting content to the right people at the right time on the right device in a way that makes sense for them”transcript, stated

Small teams cannot afford one-off infrastructure for each fan surface.

Root cause: The Warriors have a global brand but fewer than 600 employees and limited technology resources.

Resolution: Use managed cloud services and a build-once API layer for reuse across surfaces.

“we have to do a lot with a little you got to be Scrappy”transcript, stated
“the technology has been so easy to integrate with the limited resources that we have”transcript, stated

Recommendation systems must serve more than the team's game schedule.

Root cause: Most Chase Center events are not Warriors games, and fans have varied interests across concerts, family shows, retail, and community events.

Resolution: Recommend events and content based on expressed interests and historical behavior.

“most of the events in the arena are actually going to be non-w Warriors games”transcript, stated
“based off what you're interested in we're showing you artists that are based on that interest”transcript, stated

Evaluation

The talk does not report formal offline recommendation metrics. Evaluation is framed around whether recommendations improve fan journeys across channels, make content more relevant, and let a lean team deliver personalized experiences.

Metrics: pilot built over six months earlier · channel deployment breadth across mobile, web, and CRM · personalized event/content relevance

3 supporting quotes
“we built a pilot uh probably a little over six months ago now”transcript, stated
“what events are people clicking on what articles are people clicking on”transcript, stated
“so I think everybody wants to have have a more natural navigation experience wants to get to the interesting information as quickly as possible”transcript, stated

Outcomes

Scale is described through content and venue scope: 82 regular season games plus playoffs/preseason, over 150 Chase Center events per year, about 50 Warriors games, a global fan brand, and fewer than 600 employees.

3 supporting quotes
“recommendation engine powered by now vertex AI”transcript, stated
“focus on uh the the customer experience of of how fans are actually going to experience this”transcript, stated
“approach and so this API is what enables us to take the hard work that the the discovery engine is doing and then deploy that to many different products and so you're seeing mobile web and CRM being the three that we've highlighted here but this could be many more in the future this could be uh scr”transcript, stated