TEEING UP A STRATEGY FOR GENERATIVE AI ADOPTION AND ENHANCING BRAND TRUST
Humans have played golf, in one form or another, for hundreds of years. Generative artificial intelligence (GenAI) is just getting started. What role will it play in golf’s future? Given the technology’s vast potential, the United States Golf Association (USGA), a leading voice in golf since 1894, was eager to discover how GenAI could be implemented to enhance how golfers and fans connect to and experience the game.
As a proud sponsor and trusted advisor, Deloitte has been at the forefront with the USGA, helping to build seamless, user-friendly experiences for golfers and fans, including the USGA App built by Deloitte. The organizations hadn't begun leveraging GenAI in the work they were doing together, but the technology’s increased usage in the external marketplace and a deeper understanding of its capabilities sparked an exploration of its potential to drive new digital / data-based experiences for the USGA.
The USGA’s Rules team identified an opportunity to use GenAI to create a scalable way for the organization to convey the official rules of golf to the tens of millions who enjoy the game, and Deloitte shared a similar vision. As the organizations discussed potential solutions, it became clear they could leverage each other’s highly specialized knowledge and unique assets—Deloitte’s in the GenAI space and the USGA’s rules experts coupled with the USGA’s proprietary database of years and years of past rules inquiries from golfers across the country. Together they were a perfect fit and could create a universally accessible way for anyone to ask a rules question and have confidence that they were receiving the correct answer.
TRANSPARENT REASONING CAN IMPROVE GENERATIVE AI’S HANDICAP AND BUILD TRUST
The Rules of Golf was formatted for the eye, with indentations, italics, and other meaningful elements. While these are helpful for a reader to digest quickly, they don’t translate to supporting a data science solution that’s enhanced by information organized in a machine-readable format. The USGA—and only the USGA—had another important asset: thousands of rules queries addressed by its specialists over the years. This proprietary data asset was an integral component for the GenAI solution architecture.
The Deloitte team included avid golfers and people who had never played. Despite varying golf knowledge, they had strong data science skills and an ability—informed by Deloitte IndustryAdvantage™—to implement GenAI at the enterprise level to create effectiveness and efficiency. They also had deep experience helping organizations transform their strategic visions into value. To accelerate impact for the USGA, the Deloitte team's first task was an exploratory analysis of the USGA’s treasure trove of data so the GenAI Rules Engine could learn from it and the database could be integrated into the foundation of the USGA’s growing digital ecosystem and leveraged for new value.
This included careful consideration of the terminology used in The Rules of Golf, as it often differs from user queries, which may be more conversational, or based upon dependencies, such as stroke play versus match play. Working with the raw data holistically helped the team get to the crux of each rule: What does it mean? What constitutes an amendment versus a clarification? And how should the GenAI treat this information?
This process, and the USGA’s long-term vision, informed Deloitte’s creation of a proof of concept for a bespoke query-and-response tool that could satisfy the USGA’s accuracy, scalability, and security requirements and integrate with existing platforms. The USGA aspired for accuracy that could match or exceed the answers the USGA’s own subject matter experts could provide and near real-time response.
To meet those accuracy requirements, Deloitte equipped the USGA’s rules tool with transparent “chain of thought” reasoning that identifies the assumptions guiding the GenAI response. This helps a user recognize that an answer may not be wrong, so much as missing key information, so they can backtrack and provide a sufficiently comprehensive question. The USGA viewed this GenAI version of “showing your work” as a game changer that illuminated a path toward achieving the high standard of accuracy set forth.
Then, Deloitte worked with the USGA to use better prompt engineering, like editor notes, so internal users could understand how to ask better questions and receive better answers in return. For the USGA, the high standard of accuracy wasn’t about advancing technology but about maintaining the trust golfers have in the USGA, the information it provides and the rules it has stewarded since 1894.
Once built by Deloitte, the rules response-generation tool was implemented in a proof of concept that was designed to prove technical feasibility. This teed up development of a pilot that enabled GenAI to respond to email queries that were reviewed and refined for accuracy by USGA subject matter experts before sending the reply to the user. This work provided an opportunity to further train the GenAI tool for future implementations by the USGA.
This “learning” is taking place under conditions that don’t require an immediate response, unlike a chatbot that needs to react within a few seconds, enabling continued enhancement of the database and increasing the tool’s accuracy. Even at this early stage, the USGA sees the potential efficacy of the tool and is excited to unlock future data-driven opportunities.
The customized customer service query-and-response model Deloitte built for the USGA could be applicable for any number of organizations that rely on specialized industry knowledge and regulatory components. Their queries, feedback, and reference materials might once have been filed away in a cabinet. That no longer needs to be the case, and Deloitte is helping the USGA and other clients leverage their expertise and democratize access to it.
An organization’s expert-level knowledge may be in a variety of formats, from text documents and emails to social media posts or videos. GenAI can help turn this unstructured data into accessible, relevant information an organization and its constituents may need—whether that’s to fix a home appliance, better understand innovative medical technologies, or—just maybe—lower their golf handicap.