Bringing Entity Language Models To Life… with Mike Trout – Part 1

Continuing our Sports Analytics Journey

In my previous blog, Entity Language Models: Democratizing Entity Propensity Models, I introduced the concept of Entity Language Models (ELMs), an evolution of traditional Entity Propensity Models (EPMs) that blend predictive and causal analytics with the natural language fluency and contextual adaptability of Generative AI. That blog made the case for the importance of ELMs in highlighting how they unlock explainability, operational impact, and continuous learning.

This blog brings that idea to life with a real-world baseball example around one of baseball’s greatest players, Mike Trout of the Los Angeles Angels.

We’ll explore how a Mike Trout-specific ELM could support key decisions across workload, recovery, in-game strategy, and even trade value (sorry, Mike).

This post will highlight the steps in the Thinking Like a Data Scientist (TLADS) methodology that are impacted by the evolution to an operational ELM. You’ll see how EPMs quantify an entity’s propensities (like Mike Trout) and transform those propensities into real-time decisions, natural language outputs, and learning loops that adapt and get smarter over time (Figure 1).

Figure 1: Thinking Like a Data Scientist Methodology

Let’s review the impacted TLADS steps.

TLADS Step 3: Model Business Entities – Mike Trout as a Strategic Asset

In TLADS, Step 3 models the business entities—humans (patient, nurse, student, teacher, operator, technician, athlete) or devices (chiller, compressor, turbine, motor)—that drive customer/stakeholder, operational, and societal value. To bring the concept of a business entity to life, we will jump back into my baseball world and try to model the predictive performance and behavioral tendencies of Mike Trout, a player whose performance can significantly influence the outcome of games and seasons.

Trout isn’t just a player—he’s an amalgamation of behavioral and performance tendencies, including pitch response, performance under pressure, zone awareness, decision-making in base situations, discipline in chase rates, consistency in late innings, swing-path efficiency, and adaptability to different pitcher types. Modeling these tendencies forms the foundation for developing actionable intelligence to optimize his care, development, and utilization.

TLADS Step 5: Brainstorm Analytic Scores – Mike Trout’s Entity Propensity Model

In TLADS Step 5, we generate analytic or propensity scores that seek to quantify the entity’s performance and behavioral propensities. For Trout, we built an EPM composed of scores such as High-Leverage Performance, Pitch-Type Effectiveness, Contact Quality, Momentum-Adjusted Performance, and Hot/Cold Zone Effectiveness.

Each score includes (Figure 2):

- Short description

- Analytic strength rating from 0 to 100 (where 100 is the strongest correlation)

- Required data sources (Statcast, Baseball Reference, scouting reports, etc.)

- Different game-time applications

Figure 2: Mike Trout Entity Propensity Model (EPM)

Step 7: Map Scores to Decisions – From Propensity to Precision

This is where the transition from an EPM to an ELM starts. In TLADS Step 7, scores are translated into clear, contextual decisions and actions. That is the function of an ELM.

Where an EPM might say, “Trout’s xSLG vs RHP sliders is .640,” an ELM says: “Trout thrives against tonight’s starter’s slider. Recommend keeping him in the 3-hole to maximize RBI potential in innings 5–8.”

Here is an example of how we could operationalize Mike Trout’s ELM through an AI Assistant Coaching Agent (Figure 3).

Figure 3: Mike Trout Entity Language Model (ELM)

Step 8: Create a Learning-Based User Experience – The ELM Feedback Loop

Step 8 in TLADS is where user experience becomes intelligence. With ELMs, every interaction—every time a coach accepts or overrides a recommendation—feeds back into the model. That feedback informs future outputs. For example:

“Last time you batted Trout vs this pitcher, he had a key pinch-hit double. Adjusting his Momentum Score upward.”

This is where ELMs move from “smart models” to adaptive teammates. The more you use them, the better they understand your world.

Conclusion: Bringing ELMs to Life – Part 1

I love using sports to highlight the power of AI and new data science approaches—such as nanoeconomics, entity propensity models, and entity language models—to improve individual and team (organizational) performance. It’s in my blood, so that won’t change.

In Part 2, I will discuss how you can use my Thinking Like a Data Scientist methodology to develop, deploy, and optimize your EPM and ELM assets. I will also elaborate on the crucial EPM/ELM feedback loop that enhances AI’s ability to create economic assets that appreciate, rather than depreciate, in value through their ability to continuously learn and adapt.

Yep, why bother talking about it if you can’t deploy something and derive value from it?