The Next Frontier
Driveline Baseball founder Kyle Boddy said it best in a recent tweet: “Before, it was fairly easy to see those working on the next big thing. But large baseball organizations cannot / do not fund these initiatives, and curious individuals are enlisted into these organizations due to their talent and put to work on operational, margin-maximizing efforts. All very reasonable and incentive-driven concepts. But we’re going to run out of the easy stuff pretty soon, and very few people are working on what’s next.”
All over the news, we see business leaders from technology behemoths like Alphabet and Microsoft describe artificial intelligence (AI) as a “once-in-a-generation kind of opportunity” whose closest analogy is when “the PC became standard issue in the early 90s.” For an industry I generally view as on the cutting edge, I have heard no talk of how MLB teams can utilize AI to their benefit. If tech leaders believe AI’s effect on how we work to be greater than the effect the computer had, and baseball’s valuation paradigm completely shifted when computers started being used, how will AI change the scalable processes that every MLB front office concerns itself with?
At the core of transaction theory is how to value assets using statistical forecasting and how to devise a macro strategy to leverage edges a team has in valuation. I have no information on how teams go about both of these, but I’d imagine it revolves around getting great baseball minds in the room together and codifying what they believe about baseball and why. There’s clearly an edge in being above average at blending your organization’s human intuition with its forecasting’s intuition. But what does the future of this blending look like?
Bridgewater Associates, the world’s largest hedge fund by assets under management, recently released their white paper for the AIA Forecaster, “a Large Language Model (LLM)-based system for judgmental forecasting using unstructured data.” I believe that the AIA Forecaster provides a blueprint for MLB teams to most effectively use their intellectual property in a scalable fashion in an effort to optimize transacting. An LLM-based system that uses judgmental forecasting, rather than just statistical forecasting, would be a meaningful shift and provide a different dimension to decision-making that other teams don’t have access to. In addition, if the quality of a team’s intellectual property is greater than opposing teams, it seems possible that the edge within that information could be exacerbated by creating a system that can aptly take it all into account.
The key distinction here is that of judgmental forecasting, which they later go on to define as “the process of aggregating unstructured data (e.g., news articles, scientific reports, etc.) and using this data in conjunction with past experience to logically predict some future outcome.” Judgmental forecasting contrasts itself with statistical forecasting, a different methodology that “utilizes some combination of tabular data and simulation to develop a mathematical model of what might occur in the future.”
MLB organizations are familiar with statistical forecasting. As the white paper aptly summarizes, “Wherever there is uncertainty about the future, there is forecasting.” The name of the game in baseball analytics and front office decision-making is handling uncertainty. Models, scouting, and collective decision-making are all vehicles to handle it. The AIA Forecaster is a comprehensive system designed to do the same.
Statistical forecasting doesn’t lack sophistication by any means. In baseball terms, statistical forecasting generates rational, mathematically grounded distributions of player outcomes and values them accordingly. What’s difficult about the forecasting process as a whole is taking those distributions and baking in non-tabular data like injury reports, market analysis, and scouting reports to aid in the valuation process. Bridgewater has found the same to be true.
To illustrate how the AIA Forecaster works, they give a real estate example, but for the sake of this post I’ll use a baseball one. Suppose it’s two weeks prior to the trade deadline and a pro scouting department wants to predict how likely it is that a trade target of theirs actually gets dealt by the deadline. The pro scouting department already has statistical models to tackle this problem. They use factors such as the player’s projected value, his team’s playoff probability, and his team’s past willingness to make trades to put a probability behind whether or not this player will be dealt. Although a sound methodology in its own right, this forecasting has its shortcomings.
Let’s add a layer to the story and suppose that news hits via ESPN.com explaining that the GM of the player’s team is facing increased pressure to transact at the deadline. The news story also notes that the player’s performance has been affected by a nagging back injury. As humans, it’s difficult to incorporate this unstructured yet noteworthy information because of current modeling limitations. The most comprehensive forecast would blend the withstanding statistical modeling with an additional model that could incorporate news that has an effect on outcomes. The problem here isn’t just understanding whether he’ll be traded, but also how to leverage the probability of him being traded to make more informed decisions on what to offer.
The Twins fire sale at the deadline portrays a real-world example of why adding unstructured data to models can bear surplus value. Their fire sale came as a shock, but a judgmental forecasting approach that incorporates unstructured data may have foretold their trade deadline strategy. Not only were the Twins in the midst of their lowest attendance total in the last 16 seasons, but ownership also mandated a payroll reduction. An October 2024 article highlighted that the Pohlad family sought to sell the franchise and retained an investment banking firm to facilitate a sale. All of these factors certainly had a major effect on the team shockingly dealing Carlos Correa, their prized reliever Jhoan Duran, and three other leverage bullpen arms.
I don’t know the exact process of how teams plan for the trade deadline, but I do know that the deadline is exactly that—a deadline. Any deadline or time constraint means that having as much information as soon as possible can yield a better plan. If you had a system that could reliably forecast your opponents’ trading tendencies, the amount of possible worlds shrinks and a more robust, planned operation can be enacted. Because of this, trade deadline deals become contingent on one another, therefore emphasizing the importance of understanding the scope of possible trades in a scalable way that doesn’t include reaching out to every team on every player and trying to put yourself in that GM’s shoes.
The idea behind an AIA Forecaster–esque system seems somewhat straightforward, but to achieve it, a team that believes strongly in its merit would have to commit to a disciplined approach. In a recent podcast, Greg Jensen, Bridgewater’s co-CIO, explained what goes into creating proprietary unstructured data that feeds into this model and made no bones about the fact that its process is anything but easy. However, his ideology appeared to mirror that of many great baseball organizations.
Jensen attributes Bridgewater’s success to being “focused [on] two things—how do you deeply understand how the global financial system works and how do you build portfolios.” He continues:
“To drive that, the idea that we have to do that by compounding understanding. You have to have the discipline to write down what you believe and why you believe it, share that with others so that they can assess what’s wrong about that and right about that, build that out, and keep compounding understanding… To me, basically, the focus on those things, getting a culture of people who care deeply about how those two things work, and then taking everything we ever learned and having the discipline to write it down, translate it into algorithms, and keep moving forward that way—those are the magical pieces.”
What Jensen outlines here doesn’t seem to fall far from how baseball organizations think. The best teams strive to think empirically, just like Bridgewater does. To create scalable infrastructure, it’s imperative to have data that backs up exactly why you want to build out your systems. The sharpest baseball teams already have an upper hand in creating an AIA Forecaster–like system because they have the “magical pieces” that Jensen lays out—people who care deeply about these questions and have the discipline to think empirically when it comes to what they believe.
I can’t say for certain how AI will affect baseball front offices, but history suggests that the first team to apply it well will gain a meaningful edge on its competitors. In this case, that edge would not simply come from adoption, but from being first to begin creating the unstructured data that can fuel a system of this kind. As that data compounds over time, so too would the model’s understanding. The road ahead in answering this question won’t be easy, and the team that succeeds will likely be the one willing to break from conventional thinking and commit to building something before its value is obvious.
