Fanatics Betting and Gaming deployed a multi-agent AI system on AWS for customer support
Fanatics Betting and Gaming deployed a multi-agent AI system on AWS for customer support in sports betting. The system adapts to the rules of individual US states, recognizes signs of problem gambling and escalates complex cases to a human agent.
Fanatics Betting and Gaming (FBG), a sports betting operator active in multiple US states, deployed a multi-agent AI system for customer support on AWS infrastructure. The goal was to handle a growing volume of inquiries — more than 40 inquiries arrive every two minutes during major sporting events — while also dealing with rules for payment methods, deposit limits, withdrawals and responsible gambling that vary from state to state (for example, Indiana versus New Jersey). According to the company, traditional chatbots based on decision trees could not handle this complexity, leading to customer frustration and growing queues for human agents.
The system uses an orchestrator-based architecture. A customer message from the FBG mobile app passes through Salesforce Einstein and a Spring AI service running on Amazon EKS, then through Amazon Bedrock Guardrails (prompt injection detection) and a responsible gambling classifier built on the Amazon Nova 2 Lite model before reaching the so-called Supervisor Agent. This agent runs on the Claude model from Anthropic within Amazon Bedrock and calls specialized tools as needed — including a RAG pipeline for information retrieval and MCP servers for accounts and transactions.
The responsible gambling classifier evaluates each message against an approved framework and, when a classification indicates high severity, automatically transfers the conversation to a human agent along with the full context. According to the company, the modular architecture, in which each agent has a clearly defined responsibility, allows the team to add new tools, knowledge domains and business units without changing the rest of the system, and the solution is intended to handle requests faster, more accurately and at lower cost than support provided solely by humans.
The source does not describe the rest of the flow, in which the supervisor synthesizes responses from the individual tools. You can find details in the source article.
Why it matters
The case shows how companies in regulated industries, such as betting or finance, can use an architecture with an orchestrator and specialized agents on Amazon Bedrock to handle peak customer support demand, comply with regulations that differ from state to state and reduce the cost of human support at the same time. A key element is automatic transfer to a human in sensitive cases, for example when there are signs of problem gambling.
Two audiences, two different impacts
What this means
For individuals
Developers designing similar customer support systems can use the described architecture — an orchestrator, specialized agents, MCP tools and Bedrock Guardrails — as a reference pattern for their own multi-agent solution.
For a business
Based on the case described, companies in regulated industries can use such an architecture to reduce customer support operating costs and scale faster during peaks without compromising regulatory compliance.
DevelopmentCheck the original
Event sources
only one source so far · 1 publisher, 0 independent. We count feeds from the same owner only once.