How we use AI

Effective 3 October 2026

What it does

• Drafts emails, replies and reminders for you to review and send yourself.
• Reads bank and card statements and suggests a category for each line, asking you about the unclear ones. For health clients, a person reviews statements instead, for now.
• Answers questions about your own figures and schedule, by voice or typing. It calls a figure up, down or a record only when your numbers show it.
• Summarizes information and flags items that may need attention.

You decide what goes out

AI output is a draft or a suggestion. The client app never sends messages, pays anything or publishes anything itself. Drafts open in your own email or Messages, and you or someone you authorize decides whether to send them. AI can make mistakes, which is why every output can be reviewed and edited first.

No decisions about people

We don't use AI to make, or be a substantial factor in, consequential decisions about individuals, including employment, credit or lending, housing, insurance, healthcare, education, or legal services. Our systems don't score, rank or screen your customers, patients or applicants.

Your data and AI providers

• The client app's AI runs on Amazon Bedrock, which doesn't use your data to train models. We also use Anthropic and OpenAI in our own internal work, and we never send them protected health information.
• The AI in the client app runs on Amazon Bedrock inside our AWS account, and the app's voice runs on AWS too.
• For health clients, AI reads their records only on AWS under our Business Associate Agreement, and only once that client is set up there and their BAA with us is signed.
• For health practices, a person reviews statements, and only totals reach the app's Money view.
• We never use your information to train our own or anyone else's models.

Voice control, in development

We're building voice control of a computer and use it on our own machines. It isn't part of client services. It runs on a separate login that can't open client accounts, refuses to look at or read windows that mention a client or patient records, waits for a yes before risky commands, and logs every action. It may use AI providers outside AWS, which is why client data is kept away from it.

Telling your customers

Messages drafted with AI are approved and sent by you, under your name. Some laws require a disclosure on AI-generated patient messages. If a law like that applies to you, we'll help you set up the disclosure. You stay responsible for your own communications, including consent and opt-outs for the emails and texts you send.

Limits

Financial dashboards are for management use and are source material for your CPA. They are not accounting, tax, audit or legal advice.

Questions

Email nick@impressionsystems.co and we'll tell you which models touch your data and why.