Explainable AI for Disease Prediction in Pets

Predictive AI only generates value when people trust it enough to act upon it. In regulated, data-heavy industries such as insurance that trust isn't assumed. It has to be built, methodically, long before any model ships.

In this webinar, Dr. Christian Debes (Head of Data Analytics & AI, Spryfox) walks through the real engineering and collaboration process behind Fetch Health Forecast, a live AI product that accurately predicts the likelihood of 45 disease categories based on data collected on nearly 800,000 dogs over 17 years of insurance claims.

Joining him are Dr. Audrey Ruple, Professor of Veterinary Medical Informatics at Virginia Tech, and Dr. Aliya McCullough, Chief Veterinary Officer at Fetch Pet Insurance, the domain experts who helped shape the model from the inside.

Together, they'll show why AI explainability (XAI) is a discipline you build from the first line of model design, not a feature you add at the end.

The session will cover:

• Why explainability starts with the domain expert in the room: How the collaboration between data scientists and veterinary specialists shaped every stage of the Fetch Health Forecast model.
• The case for building deliberately simple models first: How baseline "stupid models" catch data leakage, validate signals, and create a shared language between data scientists and domain experts before complexity is introduced.
• Translating model outputs into real-world decisions: How feature importance and partial dependence plots were used to surface insights a veterinarian, an insurer, or a pet owner could actually act on.

Attendees will walk away with:

1. A repeatable framework for explainability-first (XAI) model development: A concrete, stage-by-stage process that can be applied to any data-heavy prediction problem.
   
   
2. Practical criteria for evaluating XAI tools against your audience: How to assess if an explanation method will actually build trust with the stakeholders who need to act on model outputs.
   
   
3. A defensible answer to "why does your model predict this?": The habits and documentation practices that ensure your team can easily justify individual predictions to regulators, customers, or internal reviewers.