AI Officiating in American Sports: What Robot Umpires Teach Any Business About Automation


A catcher taps his helmet. Twelve cameras around the ballpark have already tracked the pitch. Fourteen seconds later a graphic on the outfield videoboard shows the ball clipping the edge of a rectangle, the call flips from ball to strike, and play resumes without an argument.

That sequence became routine in Major League Baseball in 2026. It is worth studying even if you have never watched a full nine innings, because two of the biggest leagues in the United States spent six years deciding how much of a human job to hand to a camera. The answer they landed on is more interesting than the hardware.

Baseball chose the narrow version of the problem

MLB’s Joint Competition Committee approved the Automated Ball-Strike Challenge System in September 2025, and it came into Spring Training, the regular season and the postseason from 2026. Twelve Hawk-Eye cameras track pitch location and send the result over a private 5G network. Each club starts with two challenges and keeps any it wins. Only the pitcher, catcher or batter can ask for one, signalled by a tap on the cap or helmet.

Note what the league did not do. It had a working system capable of calling every pitch, tested in the Atlantic League from 2019 and across the minors after that. Players preferred the challenge format, and that preference decided the design. The umpire still calls the game. The machine only settles disputes, and only when a player on the field spends one of two limited tokens to raise one.

The numbers behind that choice are modest and specific. Big league umpires already call roughly 94 percent of pitches correctly according to UmpScorecards. During Spring Training 2025, teams won about 52 percent of their challenges, and each review took an average of 13.8 seconds.

The NFL made the same call with a different problem

Football took a parallel route. From the 2025 season the NFL replaced the chain gang with Sony Hawk-Eye virtual measurement, using six 8K cameras to track the ball’s position across all 30 stadiums and the international venues. Sony says a measurement takes about 30 seconds, saving up to 40 seconds against the chains.

The detail most people skipped: the chain crew stayed on the field as a backup. A league with billions in annual revenue, replacing a piece of equipment made of rope and sticks, kept the rope and sticks on standby. That is not nostalgia. It is a fallback plan for a system that, as Sony has acknowledged, is not perfectly accurate and can fail.

The pattern worth copying

Both leagues did the same four things. They picked one measurable question instead of automating a whole role. They set a time budget the system had to fit, because a 90 second review would break the rhythm of a game. They built an appeal path controlled by a human. And they kept the old method alive while the new one proves itself.

Most failed automation projects break the first rule. A company decides to automate “customer support” or “invoice approval” rather than one decision inside it, such as whether a submitted invoice matches a purchase order within tolerance. The scope stays vague, the model gets blamed, and the project quietly dies. The teams that get this right, including the AI and business automation specialists at iTech Business Solutions, usually spend more effort on the process map than on the model itself, because the map is where you find the one question worth answering.

What is still unresolved

The ball-strike challenge system does not remove pitch framing. Catchers still shape borderline pitches to look like strikes, because only two calls a game can be contested. An automated system with a limited budget of appeals changes behaviour around the edges rather than eliminating it, which is exactly what happens when you add an automated check to a human workflow without changing the incentives underneath.

There is also a commercial layer forming on top. ABS reviews appear as branded animations on videoboards and in broadcasts, which turns an officiating tool into inventory a sponsor can buy. Anyone following how these rule changes land week to week will get more from ongoing coverage of American leagues at The Sports America than from a vendor press release, because the arguments about pace of play and fairness are happening in real time.

The takeaway for your own systems

If you are planning an AI project this year, borrow the structure rather than the technology. Write down the single question the system will answer. Decide the maximum time an answer may take

before it becomes useless. Give a named human the power to overrule it. Keep the manual process running until the new one has a full season behind it.

Baseball spent six years and three levels of professional ball getting to a system that corrects about two calls a game. That restraint is the lesson. The leagues that moved slowly on automation ended up with tools their own participants trust, and trust is the thing that determines whether anyone actually uses what you build.