Sports Analytics
Player performance, fan lifetime value, and roster decisions all need probability distributions, not point estimates.
Our chief scientist Chris Fonnesbeck spent 7 years in professional baseball research with the Phillies, Yankees, and Brewers. Sports analytics has the same structural problems Bayesian methods solve everywhere: small sample sizes, hierarchical structure, abundant prior knowledge, and the need for interpretable outputs.
Hierarchical models provide a means for integrating information at multiple scales and adjusting for biases associated with small sample sizes.
Work in This Space
Los Angeles Dodgers
Time Series Analytics
Bayesian time series modeling assistance for one of MLB’s most analytically sophisticated organizations.
Real Madrid C.F.
Fan CLV Modeling
Fan lifetime value modeling using PyMC-Marketing BG/NBD and MBG/NBD models. Engagement contributed covariate support improvements back to pymc-marketing.
Trusted By
- Los Angeles Dodgers
- Real Madrid C.F.
What We Solve
Fan Lifetime Value
Probabilistic BG/NBD and Pareto/NBD models for fan cohort segments. Hierarchical CLV with covariate support for season ticket holders, casual buyers, and streaming-only fans.
Player Performance Modeling
Hierarchical Bayesian models pool information across players with similar profiles. Credible intervals communicate genuine uncertainty to GMs and coaching staff.
Spatial & Positional Analytics
Bayesian spatial models using Gaussian Processes over 2D playing surfaces. Goaltending evaluation, field zone analysis, and positional performance metrics.
Contract Valuation Under Uncertainty
Posterior predictive distributions for future performance scenarios. Monte Carlo contract value simulations for multi-year deals with full uncertainty quantification.
Why Bayesian
Why Bayesian for Sports Analytics?
- Small sample sizes — Early-season player performance and rare events need models that borrow strength across similar players and situations through partial pooling.
- Hierarchical structure — Players within teams within leagues, each borrowing statistical strength from peers. The natural structure of sports data maps directly to Bayesian hierarchy.
- Prior knowledge is abundant — Career statistics, physical measurements, and domain heuristics are natural priors that Bayesian models incorporate formally rather than discarding.
“Hierarchical models provide a means for integrating information at multiple scales and adjusting for biases associated with small sample sizes.”