Fin is already the best-performing AI Agent for customer service. But we believe it can go further. Our goal is to build a version of Fin that learns from every successful resolution, so it continuously improves and gets better at doing what it does best: helping your customers.

To get there, we’re taking a big step forward. The Fin AI Group will begin training custom models in-house. These models are designed to help Fin learn from the moments when it resolves issues clearly and completely.

When Fin answers an informational question, it pulls up and ranks content from your knowledge base. That step—choosing the right content—is just as important as the model that generates the reply.

The Fin AI Group will start by training custom models to improve how Fin retrieves and ranks that content. These models will learn from successful conversations. Every resolved interaction will become a chance for Fin to improve at finding the right answers, in the right order—for your business, and for every business using Fin.

We’re building these models with strong safeguards in place. This includes technical and organizational measures to ensure best practice for customer data handling and model development and deployment.

We’re leveraging a secure anonymization process, using trusted tools like Microsoft Presidio, to remove identifying information before creating training data for these models. Model training is performed within our existing infrastructure, and the fine-tuned models will only be used to improve Fin performance for our customers. We will build and deploy these models in compliance with the responsible AI governance principles we follow as part of our ISO 42001 certification.