How Fanatics Betting and Gaming built a multi-agent customer support system

Fanatics Betting and Gaming (FBG) created a multi-agent customer support system on AWS to handle the unique challenges of the sports betting industry. Customers expect immediate, precise answers, particularly during live events when every second counts. Queries span account issues, deposit limits, state-specific regulations, and responsible gaming resources, with rules varying across every licensed jurisdiction.

Traditional chatbot solutions based on decision trees are inadequate for this complexity, leading to customer frustration and increased costs due to growing human agent queues. FBG, a part of the Fanatics brand, operates in multiple U.S. states and requires 24/7 support, especially during high-traffic events like NFL playoffs and the Super Bowl.

To address rapid growth in support volume, FBG's engineering team developed a multi-agent AI system on AWS that resolves customer issues more quickly, accurately, and cost-effectively than a human-only support model. This article outlines the system's architecture, the AWS services used, and the patterns for designing similar multi-agent customer support solutions.

As FBG scaled, human support demands increased proportionally with the customer base, raising operational costs. The team identified an opportunity to enhance customer experience while preparing for future growth. U.S. state regulations for payments, deposit limits, withdrawals, and responsible gaming differ, making it challenging to provide accurate responses.

During major sporting events, support requests can surge to over 40 inquiries every two minutes, necessitating instantaneous scalability without compromising response quality. Queries range from transaction history and account settings to betting rules and self-exclusion options, requiring a nuanced understanding of conversational context to identify and respond to signs of problem gambling.

FBG needed an autonomous system capable of handling this complexity while ensuring compliance with responsible gaming and regulatory requirements. The solution involves a multi-agent system, each specializing in different aspects of customer interactions. Leveraging their expertise in Amazon Elastic Kubernetes Service (Amazon EKS), FBG deployed, scaled, and iterated on each agent independently.

Amazon Bedrock was chosen for its model-agnostic access to multiple foundation models via a single API, allowing the team to match tasks to the best model and swap models as better options become available. The architecture employs an orchestrator pattern, with a primary orchestrator agent receiving customer messages, coordinating with specialized tools and sub-agents, and generating a unified response.

This design enables the addition of new capabilities, such as new tools, knowledge domains, and business units, without modifying the core system.