AI Policy

Introduction

Holli includes an AI assistant, also called Holli, built into the practice management platform. This policy explains what the AI does, what it deliberately does not do, what data it uses, and what we expect from practitioners who use it. It should be read together with our Privacy Policy.

What Holli's AI Does

The AI assistant supports practitioners with practice workflows:

• Chat and document questions: answers questions grounded in the documents a practitioner has uploaded to their knowledge base, plus platform-maintained reference materials, with numbered citations back to the sources.
• Session preparation: drafts session prep outlines from the practitioner's knowledge base and the client context they have entered, including intake responses.
• Summaries and recaps: drafts internal session summaries, client-facing recaps, and journey recaps across a client's session history.
• Document drafts: drafts referral letters, follow-up messages, and other practice documents, formatted with the practitioner's branding.
• Live transcription: converts session audio to text in real time (with consent, see below) so the practitioner can focus on the client.
• In-app help: answers questions about how to use Holli itself.

Humans Stay in Charge

Every AI output in Holli is a draft for the practitioner. The practitioner reviews, edits, and approves AI-generated content before it is used or shared with a client. The product is designed around this principle: generated documents go through a preview and sign-off step, and the interface reminds practitioners to apply their own professional judgement.

AI suggestions use careful language (“consider”, “may”, “you might explore”) rather than definitive clinical statements, and client-facing summaries only include techniques and observations that the practitioner actually recorded in their notes.

What the AI Is Not

• Not a medical device or diagnostic tool. Holli does not diagnose conditions, prescribe treatment, or provide medical advice. Even features that flag things for attention (such as soft safety screening on session notes) are suggestion-only prompts for the practitioner to consider, not clinical assessments.
• Not a substitute for practitioner judgement. The practitioner remains fully responsible for their professional decisions and for the care of their clients.
• Not a guaranteed source of truth. AI models can make mistakes, and document retrieval can miss relevant material. Outputs must be verified before use.
• Not client-facing. Clients cannot chat with the AI. It works for the practitioner only.

What Data the AI Uses

When a practitioner uses an AI feature, the relevant content is sent to our AI providers to generate the response. Depending on the feature, this can include client names, session notes, intake responses, and excerpts from uploaded documents. Specifically:

• OpenAI (United States) processes text for chat, retrieval, summaries, and document generation.
• Deepgram (United States) processes session audio for live transcription.

We use these providers through their business APIs. Your data is not used to train their models. Data storage and cross-border processing are covered in more detail in our Privacy Policy.

Recording & Transcription Consent

Live transcription involves recording a client's voice, so consent matters. Before AI processes any audio-derived content from a session, the practitioner must confirm in the app that the client has consented to recording. This is enforced on the server: if a session used live audio transcription and recording consent was not recorded, Holli blocks AI processing of that session's transcript content (such as summaries generated from it).

Practitioners are responsible for actually obtaining that consent from their clients, in line with their professional obligations and the laws that apply to them. Practitioners can delete transcript data from a session at any time.

Citations & Transparency

When Holli answers questions from documents, it uses retrieval-augmented generation: the AI answers from retrieved excerpts of the practitioner's uploaded documents and platform reference materials, and shows numbered citations so the practitioner can check the source. For document-grounded answers, the AI is instructed to answer only from the retrieved material, and to say when it could not find the information rather than invent an answer.

Limitations to Keep in Mind

• AI-generated content can contain errors, omissions, or awkward phrasing.
• Document retrieval may be incomplete: relevant material can be missed, especially in large or poorly scanned documents.
• Transcription accuracy depends on audio quality, accents, and terminology, and may contain mistakes.
• Anti-hallucination instructions reduce, but cannot eliminate, the risk of incorrect content.

Always verify AI outputs before relying on them or sharing them with a client. Generated PDFs include a disclaimer that the content is a supportive summary only and does not constitute medical advice.

Responsible Use by Practitioners

Practitioners using Holli's AI features are expected to:

• Review, edit, and approve all AI-generated content before using it or sharing it with clients or third parties
• Obtain informed consent from clients for data collection, and specifically for session recording and transcription
• Not present AI outputs as diagnosis, treatment, or guaranteed outcomes
• Keep client records accurate, and not rely on Holli as the sole record of care
• Comply with the privacy, health-record, and professional laws that apply to their practice

Changes to This Policy

Holli is in active development and its AI capabilities will evolve. We will update this policy when the AI's capabilities, providers, or data handling change in a material way, and we will notify users of significant changes by email or through the application.

Contact Us

If you have questions about how Holli's AI works or how your data is used, please contact us:

Related Policies

• Privacy Policy (how we handle your data)
• Accessibility Statement (our accessibility commitments)