How Gran Minerva uses AI

Which AI we use

Our AI models run on Google Cloud's Vertex AI, in the United States. We use Google Gemini models for most features, and Anthropic Claude models, served by Google, for some background analysis. Reading the tone of text uses Google's Natural Language API, a global service that may process that text outside the United States. Your data doesn't leave Google Cloud to reach any of these, and Google does not use it to train its AI models.

What the AI does

• Answers your questions. The Minerva assistant answers questions about your accounts, like “who is most at risk this month?”
• Sorts and reads your files. It works out what kind of data you uploaded and where it goes, and picks out risk signals in text such as support notes.
• Reads tone. It reads whether text like support tickets and emails sounds positive or negative.
• Writes suggestions. It helps draft playbook steps and summaries.
• Helps our support team. It sorts emails you send to support@. For common questions it may send you one of our pre-written help answers right away. Anything else gets a draft reply that a person reviews before it is sent.

What the AI sees

Before your data is analyzed, we replace names, emails and other personal details with codes. Most AI features work only on those codes. Here are the exceptions, so you know exactly where they are:

1. When you ask the assistant about a specific account, it may look up that account's real name to answer you. That happens inside Google Cloud, only for your question, and only for people in your company who can already see that account.
2. When you email us at support@ or info@, AI reads your email, including your name and contact details, to route it and help us reply.

What the AI doesn't do

• It doesn't make decisions about people. Risk and health scores are signals for your team. You decide what to do.
• It isn't always right. AI answers and scores can be wrong or incomplete, so please check anything important before you act on it.
• Your data isn't used to train Google's or Anthropic's AI models.

How our own risk model learns

Gran Minerva's risk scores come from our own model, not from a chatbot. To make it better for everyone, it learns from patterns in numbers across customers: things like health trends, login activity, ticket counts and whether an account renewed. It never uses names, emails or other contact details, and it never shows one company's data to another.

Questions

Email privacy@granminerva.com. Our subprocessors page lists every company that handles data for us.