IBM Predicts Tennis Matches In Real Time. The Models Read The Style Guide. Of Course They Did.
IBM deployed AI agents trained in USTA editorial style and tennis terminology to power the US Open app for 14 million fans. Their Live Likelihood to Win feature continuously recalculates each player's victory probability using live scores, statistics, and expert analysis as matches unfold. The system processes billions of data points, making IBM the invisible infrastructure behind the entire fan experience.
This demonstrates domain-locked language modeling, where AI is fine-tuned on an organization's specific editorial voice and vocabulary rather than left as a generic assistant. The principle is narrow specialization: a model trained on tennis parlance and USTA style will outperform a generalist model in that context. Sports are the ideal stress test for AI because they generate massive structured data in real time, forcing systems to perform under live conditions rather than cozy lab benchmarks.
IBM, in partnership with the USTA, is running the operation. They are reaching 14 million fans worldwide through the US Open app.
- Download the US Open app from your phone's app store and open any live match page. You will see the Live Likelihood to Win percentage updating in real time as points are scored.
- Watch how the probability shifts after key moments like break points or aces. This is live Bayesian updating in action, and you are watching the model react to new evidence.
- Compare the app's predictions against your own intuition during a match. When you disagree, ask yourself what data the model might be weighing that you are not. That gap is the entire point of predictive AI.