Game, set and AI at the US Open

When over a million tennis fans attend the US Open at Billie Jean King Tennis Center in Flushing Meadows, Queens, the stakes to nail the fan experience are sky-high. Each day, around 75,000 fans pour into the grounds from either the picturesque boardwalk leading from the 7 train at the East Gate or they come by the South Gate, past the splattering fountains around the Unisphere–an iconic 120-foot-wide stainless-steel globe built for the 1964-1965 World Fair.

Another 14 million fans around the world will follow the action through the US Open app and website. “It’s incredibly important to get that first impression on day one right and keep doing it right,” said IBM Distinguished Engineer Stephen Hammer. “If that experience in-person or digitally isn't what the user is looking for, they might not come back.”

For that reason, the United States Tennis Association (USTA) and IBM are launching several new AI-driven features this year to give those fans round-the-clock information about every player on every court. The IBM and USTA partnership has “transformed how we operate and connect with fans,” said Brian Ryerson, Senior Director of Digital Strategy at the USTA in a press release. The team is “delivering real-time insights that enrich the fan experience, while significantly improving the reliability, scalability, and efficiency of our digital platforms and behind-the-scenes operations.”

Specifically, fans can converse with Match Chat, an interactive AI assistant, by either selecting an option from a menu of pre-written  prompts (e.g., “What happened in the match so far?”) or by typing their own questions, such as “Who converted more break points this match?” The US Open app also helps fans on-site with an “Ask The Open” assistant that will answer questions such as “Where can I find a Honey Deuce?,” the Open’s signature raspberry lemonade cocktail with melon pieces shaped like tennis balls.

Fans can also get live probabilities for any given player through another feature on the app called Live Likelihood to Win, which updates projections throughout a match based on AI-powered analysis of player statistics, expert opinion and match momentum.

Finally, the Key Points built with watsonx creates concise and accurate 3-bullet summaries of hundreds of articles on the US Open app and website.  Fans can click a button to get a “TL;DR” (Too Long; Didn’t Read) recap of an articles. “We can really increase the breadth and depth of what a user can experience within tennis,” said Aaron Baughman, IBM Fellow, in an interview with IBM Think. While the technology is innovative, it’s also not technology for technology’s sake, he said. Instead, he said, its function is to support the USTA’s two overarching goals: attracting more users and creating more engagement using the US Open app and website.

These enhanced AI features come at a time when audience demand for more interactive digital content is accelerating. According to a new study of more than 20,000 sports fans, 85% of tennis fans worldwide value AI-powered features, with real-time insights and personalized highlights central to how they engage with sports.

This year’s AI solutions, powered by AI agents and large language models (LLMs), are just the latest chapter in the US Open’s long history of embracing new technologies. The US Open has adopted several new technologies first, including electric line calling, which can determine whether a ball is in or out, in 2006 on certain courts and across all courts by 2018. 

Combined with their willingness to embrace new technologies, US Open fans are also known for being particularly boisterous and energetic. “The energy of the US Open is like no other tournament you have played,” said Madison Keys in an IBM video. (Keys is an American professional tennis player who won the 2025 Australian Open Grand Slam earlier this year.)

As case in point, at this year’s US Open Mixed Doubles finals on August 20^th, even a chilly, rainy evening could not dampen the spirits of roaring fans in Arthur Ashe. The stadium felt at times like a nightclub as strobe lights flashed and three different DJs played musical sets between the sets taking place on the court below. As Keys explained, “There are moments where you can’t hear the ball, the chair umpire with the microphones—you can’t hear yourself.”

In this electrifying environment, Keys said, data can be critical to getting an edge over the competition. Seeing an opponent’s stats before a match, for example, can help a player adjust and evolve their play, she said.  The collection of this data, and much of the technology powering the magic of the US Open, happens far out of sight of players and fans alike.

The key to transforming this data into insight are AI solutions, powered by AI agents and large language models (LLMs) such as IBM Granite, which process more than 7 million data points captured throughout the tournament. Every point played on each of the event’s 17 courts produces 156 distinct data points, including speeds and ball positions. This data flows into a vital broadcast complex underneath Arthur Ashe Stadium known as the IBM Data Operations room, where dozens of key personnel and statisticians ensure the accuracy and comprehensiveness of the incoming data. For close calls, fans can look to large video screens in the stadium to see an animated ball move through the air and the exact place it struck the court, whether in or out of the baseline or sidelines.

In addition to the player and court data, the tech powering the US Open collects a massive volume and variety of both structured and unstructured data that includes the umpire’s input (such as who scored a point, or who took a medical break and for how long), info from courtside statisticians, radar data, and thousands of media articles. The USTA and IBM also maintain a database of nearly 20 years’ worth of historical US Open data.

The key to managing this data is IBM watsonx.data, a hybrid, open data lakehouse that allows IBM to access and manage data from multiple sources and prepare it for AI models. All collected data is unique to the USTA, which makes it key to creating a differentiated experience on the US Open app.

Few fans are aware of the behind-the-scenes technology, and they don’t need to be. “The beauty of it is that users don’t need to know the math or tech; they can go look and quickly understand what’s happening,” said Baughman. “Match Chat gives them a way to dive deep into the data we have using natural language and assistants, which people are becoming increasingly comfortable using and asking questions to for day-to-day life or work tasks,” added Tyler Sidell, IBM’s Technical Program Director,  Sports and Entertainment Partnerships, in an interview with IBM Think.

When users ask it a specific question, like “How many aces did Madison Keys have in set one?” many different AI agents work together beneath the surface of the Match Chat interface to retrieve the information needed.

As Baughman explained, the agents behind Match Chat work together to elevate the data that matters most. First, an initialization agent organizes a request from a user. Next, a tools agent pulls information from a dozen data feeds that are relevant to the question asked, and a facts agent synthesizes that data into coherent, understandable text. Next, a judge agent reviews the factualness, relevance and stylistic components of the text to ensure the quality and accuracy of the. Finally, a corrective agent will write instructions on how to fine-tune or correct a response, which then regenerates the text and returns it to the judge agent.

This process repeats until there’s nothing more to correct, and the data flows to the answering agent, which creates an output that the system can display to a user. And all of this happens in a matter of milliseconds.

Going further under the hood, to make the AI-powered Match Chat efficient, the IBM infrastructure team used automation so that the OpenShift clusters—the platforms for managing and deploying containerized applications in the hybrid cloud environment—could accommodate the increased capacity needed when peak numbers of fans were using the feature during the tournament. Instead, the automation allowed the team to build new worker nodes that it could bring online as needed during the tournament and turn off after the main event.

“We can literally click a button on our automation to say, ‘Hey, grow to this amount of CPUs and RAM or add new worker nodes right away,’” said Andrew Young, an Associate Partner of Infrastructure at IBM, in an interview with IBM Think. Using an infrastructure tool from HashiCorp called Terraform, which integrates with OpenShift clusters, the team automated the provisioning of resources across the US Open hybrid cloud infrastructure to maximize performance and minimize cloud costs. Automation is used to grow and shrink other parts of the infrastructure as well, such as the databases housing the moment-to-moment information.

Just as top tennis players learn key lessons from every tournament they play, the IBM team debuted Match Chat at Wimbledon earlier this year and identified improvements to the system that helped them prepare for the US Open. During the 2025 Wimbledon tournament, Match Chat recorded 5 million interactions.

“Match Chat worked better than expected at Wimbledon,” said Young. “We gradually turned it on—first on Centre Court, then No. 1 Court, and so on. We had more capacity than we thought we needed. So, for the US Open, we have right-sized it to maximize its efficiency.”

A second key back-end component being used at the 2025 US Open is Apptio, which analyzes the US Open’s cloud consumption and helps identify opportunities to save. It’s currently managing the cost of 5,197 resources spread across 12 different cloud accounts.

“With the number of resources, the time it would take for a human to go in and spot check all those resources is not remotely viable. It would take too long,” said Susanne Bull, Solution Architect at IBM, in an interview with IBM Think. “Apptio gives us a tool to spot check and say, ‘Oh, hey that looks wrong. We should go look at that right now, to  save us time and money.”

“With platforms like this, it’s very easy to just leave the lights on, and Apptio helps us make sure that we turn the lights off when there’s nobody in the room,” added Hammer. 

Rounding out the critical tools that operate in the background to help reduce costs and downtime for the 2025 US Open is Instana, a service to monitor applications on the US Open digital platform. Instana updates every second and can identify, trace and resolve issues before they impact fans’ US Open experience. “With customized dashboards, we can see whether one database is running hotter than another database, for example, or whether we need to increase the capacity of the OpenShift ingress controller,” said Young. Instana also helps IBM monitor third-party integrations—there are over 30 different companies that integrate with the platform to perform various services for the USTA related to creating or distributing content.

As the team that develops and designs the US Open digital platform, IBM Consulting works with these vendors to create a seamless, integrated experience. The combination of these back-end tools is built to work together so that even if traffic swells or security incidents happen (millions of incidents are mitigated per year), there won’t be any interruptions for the fan.