About

Our Mission

Walk-Off Labs builds products that help baseball teams turn data into actionable insights that are completely explainable. We're a small group of engineers and academics with deep expertise in statistics, machine learning, and game theory.

The Problem We Solve

Every team in baseball has access to sophisticated player models and probability estimates. The tools exist. The data exists. But most teams aren't using them optimally.

Traditional approaches use the data to optimize game scenarios in isolation. They tell you the probability of an outcome, but they don't consider how to sequence decisions across an entire game, or even season, to maximize expected wins. Baseball managers are left to make these more challenging calculations on their own.

Our Approach

We use game theory to leverage the extensive-form structure of baseball and calculate subgame perfect Nash equilibria (SPE) strategies that are optimal with respect to the entire game, and even multiple games. In other words, instead of asking "what decision optimizes the outcome of the next plate appearance?", we ask "of all possible sequences of decisions that we and our opponent(s) could make over a series of games, which path maximizes our expected wins?"

Our platform, Perfect Game, takes in a player model (either our own Bayesian machine learning model or a model provided by the user) as well as user specified constraints for how each player can be used (positional eligibility, pitcher usage limits, etc.). It outputs a comprehensive strategy that specifies the win value of every admissible decision that could be made throughout the hundreds of billions of scenarios that could occur in a given game. Every decision is entirely explainable by the user inputs and game theory.