AI Use Policy

Overview

This AI Use Policy describes how TrovaInsights uses artificial intelligence in our platform, the design choices we have made to keep answers grounded in your data, and the limitations you should keep in mind when using the AI features.

How answers are produced

When you ask a question in the TrovaInsights conversational interface, we do not send a free-form query directly to a language model and act on whatever it returns. Your question is interpreted by a language model that selects from a set of structured tool functions we have defined. Those tool functions run queries against a governed semantic view layer in your data warehouse. The results are returned to the language model, which composes a written answer based on them.

The semantic view layer defines how business metrics are calculated. Metric definitions live there once, and both our AI features and our dashboards read from the same definitions. This is intended to keep answers grounded in your data and consistent across the platform.

Trust markers

Every AI-generated answer in TrovaInsights carries a trust marker that tells you how the answer was produced. There are three levels.

A green Verified marker means a pre-built tool function answered a question it was designed for. The metric, the shape of the analysis, and the framing of the result were all defined in advance by our team.

An amber Informed marker means the AI composed a custom analysis using our flexible query tool. The numbers are grounded in the semantic view, but the analytical framing was composed in the moment.

A gray General marker means no data query was executed. The answer is conversational context only.

Trust markers are a signal about how an answer was produced. They are not a guarantee that any particular answer is correct.

Model training

We do not use your business data to train, fine-tune, or otherwise improve any generally available machine learning model, ever. Our agreements with the AI providers we use as subprocessors prohibit them from doing so with your business data.

Human oversight

Our AI features are designed to assist, not to replace human judgment. You are responsible for reviewing AI-generated answers before relying on them for material business, financial, legal, tax, or similar decisions. AI output is not professional advice.

Limitations

Language models are probabilistic. Even with the semantic view grounding described above, AI-generated answers may contain errors, omissions, or misinterpretations. Answers may be sensitive to how questions are phrased.

The trust markers are intended to help you calibrate your confidence, but they do not substitute for review. When an answer will drive a material decision, we recommend confirming the underlying numbers in the corresponding dashboard or through your usual review process.

Contact

Questions about how our AI features work can be directed to app_admin@trovainsights.com.