The
Takeaway
Business units lead AI execution without a large central AI organization.
Notable reliance on AWS and a strong preference for proven technology vendors.
Prefer leading closed models, with little interest in training proprietary models.
Early signs of broader adoption of agentic solutions.
Nearly 20% of job openings are AI-related, up 228% since early 2024.
Fanatics places AI in the hands of its business units
Fanatics operates three fundamentally different businesses: online retail, trading-card manufacturing, and regulated sports betting. Each presents a distinct set of AI problems, and Fanatics has structured itself accordingly, managing AI predominantly through a decentralized model with ownership sitting in the business units. While Betting & Gaming has a dedicated Applied AI team, the other divisions embed AI-related roles within their existing organizations.
While a corporate layer exists, its scale remains limited. For instance, Fanatics has employed a company-level Head of AI since June 2024. A few groups operate across the entire portfolio on AI-related matters, such as an AI/ML platform team that builds infrastructure to accelerate AI adoption, and an enterprise data team that supports the overall AI strategy.
Strategic reliance on AWS and a gradual push toward an agentic future
Fanatics has made significant, sequenced investments in AI infrastructure: foundation first, applications after. The company assembled a standardized AI/ML platform for creating, deploying, and monitoring AI models, built largely on managed AWS services (SageMaker, Bedrock). The platform is one piece of a broader bet on AWS that runs through the stack, including storage and data processing.
Fanatics uses AI models from various providers, including OpenAI, Anthropic, and Google. Each of those vendors has carved out a niche within Fanatics.
Fanatics runs three data platforms by design. Snowflake is the fan data and business brain, acting as the engine behind FanGraph, which provides a single view of each fan stitched across commerce, collectibles, and betting. FanGraph also powers audience targeting and advertising. An in-house data lakehouse handles real-time analytics on user activity across Fanatics’ sites and apps, the fast layer the rest of the business runs on. Databricks acts as a secondary analytics platform.
Fanatics is early on agents. The organization has not adopted many of the common agent-building tools yet. However, interest in agentic solutions is building, and it reaches past the tech organization into supply chain, finance, and security.
Controlling the data foundation while staying multi-vendor on models
Fanatics keeps several crucial components in-house. The firm’s shopping platform is a proprietary, custom-built engine with AI woven in via recommendations and search. FanGraph, its unified fan-data layer, was built internally rather than bought. The reasoning is straightforward: these are the systems that make a fan choose Fanatics over a rival, so the company owns them outright.
In a notable architectural decision, Fanatics opted to develop a custom, petabyte-scale data lakehouse leveraging open-source software. Faced with the escalating compute costs and technical limitations of buying into proprietary data warehouses, Fanatics strategically chose to regain control over its infrastructure. By replacing closed vendor ecosystems with open-source foundations, Fanatics eliminated the expensive data duplication caused by vendor silos and solved sluggish 30-second query latencies. This build-centric approach allows them to process a billion daily events with sub-second speeds.
Crucially, owning an open data foundation rather than renting a closed SaaS solution gives Fanatics the operational flexibility to integrate AI on its own terms. The firm enabled the company to integrate Amazon Bedrock so employees can ask business questions in plain English and have AI automatically write database code to generate instant dashboards.
However, many components of the AI value chain are bought, including a whole host of services from AWS. At the model layer, Fanatics is deliberately multi-vendor, selecting the proper model based on the use case.
Interestingly, Fanatics has consistently refrained from making any major moves into training its own AI models. Even in 2023, when many enterprises invested heavily in building proprietary models, Fanatics held back. The organization prefers to treat AI models as interchangeable commodities that can be acquired, evaluated, and replaced as needed.
The pattern flips in business software that employs AI capabilities, where the lock-in is deep. Salesforce runs the customer-service AI, Oracle owns finance with its AI bundled in, and Adobe’s AI tools enable marketing creative.