Revolutionizing Sports Betting with FanDuel's AI Chatbot

Overview

• FanDuel, America's #1 sportsbook, partnered with AWS to develop an AI-powered betting assistant called ACAI (Autonomous Betting Assistant) to enhance the customer experience.
• The presentation covers the journey of ACAI from inception to production rollout, including the challenges faced and the technical architecture.

From Hackathon to Prototype

Identifying the Opportunity

• FanDuel held a global hackathon to explore opportunities for applying generative AI to improve customer experience and internal operations.
• The team landed on the idea of an autonomous betting assistant to create a more seamless experience for customers.

Building the Prototype

• FanDuel reached out to the AWS Generative AI Innovation Center for support, starting with a single engineer and a few data scientists.
• The team went through a discovery workshop to understand the customer, define success metrics, and identify challenges.
• The prototype focused on creating a single in-app flow for the customer to research, discover, and execute bets, without leaving the FanDuel app.

Key Prototype Findings

• Customers often leave the FanDuel app to search for information, leading to a disjointed experience and expired bets.
• The assistant was designed to be assistive, not fully autonomous, allowing customers to maintain control over the betting journey.

From Prototype to Production

Rollout Strategy

• FanDuel's rollout strategy focused on three key principles: ensuring accurate information, learning from the experience, and iterating gradually.
• The team started with a small alpha pilot, then expanded to a beta release, gradually increasing the scope and customer base.

Technical Architecture

• The solution leverages AWS services, including API Gateway, Lambda, Amazon RDS, Amazon ElastiCache, and the AWS Bedrock generative AI platform.
• Bedrock provided a consistent API for experimenting with different language models, allowing rapid prototyping and deployment.

Overcoming Challenges

1. Context Awareness: Ensuring the language model maintains context and understands pronoun references across multiple prompts.
   • Solution: Integrating DynamoDB to store and provide context to the language model.
2. Latency: Addressing the high latency (up to 12 seconds) of the initial language model responses.
   • Solution: Caching data in Amazon ElastiCache and fine-tuning the language models to improve response times.
3. Testing and Evaluation: Adapting testing approaches to handle the unpredictable nature of language models.
   • Solution: Implementing a comprehensive test suite with defined thresholds and continuous evaluation of customer feedback.

Metrics and Measurement

• FanDuel focused on measuring accuracy as the primary metric for evaluating the ACAI solution.
• The team implemented a robust testing framework, including static and dynamic tests, to validate model performance and track changes.
• Continuous customer feedback and evaluation loops were used to understand evolving user needs and adjust the solution accordingly.

Business Impact and Future Outlook

• ACAI has significantly improved the customer experience by providing a seamless, in-app flow for researching, discovering, and executing bets.
• The solution has reduced the time it takes customers to construct complex bets, such as parlays, from hours to seconds.
• FanDuel is actively exploring the use of AWS Bedrock Agent Core and API Gateway streaming capabilities to further enhance the ACAI solution.
• The success of ACAI has led to a growing partnership between FanDuel and AWS, with plans for 20-30 additional generative AI projects in the coming year.

Key Takeaways

1. Generative AI can revolutionize customer experiences in highly regulated industries like sports betting by automating repetitive tasks and providing personalized assistance.
2. Successful deployment of generative AI solutions requires a balanced approach, maintaining human control and oversight while leveraging the technology's capabilities.
3. Iterative development, continuous testing, and customer feedback are crucial for adapting generative AI solutions to evolving user needs and maintaining high accuracy.
4. Leveraging a comprehensive platform like AWS Bedrock can significantly accelerate the development and deployment of generative AI applications, enabling rapid prototyping and seamless scaling.
5. Generative AI can drive significant business impact by improving operational efficiency, enhancing customer experiences, and unlocking new opportunities for growth.