2026 SABR Analytics: Watch highlights from Doug Fearing’s keynote talk
At the SABR Analytics Conference on Saturday, February 28, 2026, in Phoenix, Arizona, Doug Fearing of Teamworks Intelligence delivered the keynote talk.
Fearing is the Chief Data Officer at Teamworks Intelligence. He was also the co-founder and CEO of Zelus Analytics. Following Zelus’s acquisition by Teamworks in 2024, he now leads the expansion of its advanced analytics capabilities across the Teamworks operating system.
From 2015 to 2018, he founded and led the Los Angeles Dodgers’ Baseball Research and Development team, growing from one full-time employee to 20 in software development, data analytics, and performance technologies. He also worked with the Tampa Bay Rays from 2010 to 2015 as a Senior Advisor with their R&D team. He holds a Ph.D. in Operations Research from the Massachusetts Institute of Technology, as well as a B.S. in Computer Science from Carnegie Mellon University.
Afterward, Fearing was honored by SABR CEO Scott Bush as the recipient of the 2026 SABR Analytics Conference Lifetime Achievement Award.
- Watch: Click here to watch a video replay of Doug Fearing’s keynote talk (YouTube)
Here are some highlights from Fearing’s talk:
On utilizing the Hierarchical Nearest-Neighbor Gaussian Process (HNNGP) Model
- “The model established a set of reference pitches and when a pitcher threw a new pitch, it predicted performance for that pitch based on a weighted similarity to the established reference set. But the real breakthrough was that the coefficients on those reference pitches could also be hierarchical, which meant we could estimate both pitcher skills and batter tendencies simultaneously. We had effectively built a model where the pitcher and batter effects were interacting mathematically much closer to how they were on the field. One of the elegant benefits of this HNNGP approach is appropriately smooth heat maps, things that you actually could present to coaches.”
On batted ball spray distributions
- “We weren’t just looking at zones or buckets on the field. We were building mix effects models to predict the horizontal spray distribution of ground balls to inform our infield shift decisions. The resulting model led to at least one interesting insight: a player’s variance in their spray distribution was proportional to their pull tendency. The players with the weaker pull tendencies were more unpredictable.”
On Latent Skill Framework using Joint Modeling
- “Instead of treating some statistics as inputs and others as outputs, the model effectively builds a joint distribution across all of a players observed performance data. Every season, every level, every handedness split simultaneously.”
On the analysis of college level players
- “That same level neutral structure that I talked about earlier in the latent skill projection framework that powered our pro projections, it allowed us to evaluate college players on the same scale as Double-A players. In addition to the benefit of being able to evaluate college players, that also meant for our pro players, we could still use their college data to inform our estimate of their skill.”
On asset value
- “Every major league team, through the minor league system, through the draft, is essentially managing a portfolio. There are lots of risks and rewards to prospects and how you value those financially. How you think about the organizational health is an important part of trades or also projecting future performance.”
Transcription assistance from Ryan Belina.
For more coverage of the 2026 SABR Analytics Conference, visit SABR.org/analytics.
Originally published: May 1, 2026. Last Updated: March 5, 2026.