Healthcare has a lot of work that AI can help with, and most of it is unglamorous: notes, forms, phone calls, prior authorizations and claims. It also has stricter rules about data than most industries. This guide covers practical use cases and the design basics that keep patient data safe.
This article is general information, not legal advice. Talk to your compliance team or a healthcare attorney about your specific situation.
Where AI is helping today
Clinical documentation. Ambient tools listen to a visit, with the patient's consent, and draft a note for the clinician to review, edit and sign. The clinician saves typing time and stays responsible for the record.
Scheduling and patient messages. An assistant can answer common questions such as hours, directions and preparation instructions, handle rescheduling, and route anything urgent to staff. A clear handoff to a person is essential.
Prior authorization and claims. AI can draft prior authorization requests from the chart, check claims for missing codes or information before they go out, and summarize the reasons behind denials. Trained staff still review and submit.
Summaries and search. AI can summarize a long patient history before a visit, or let staff search internal policies and payer rules in plain language. That second case is a good fit for RAG, where answers come from your own documents. (New to the term? Read RAG vs fine-tuning.)
Operations. Forecasting staffing, predicting no-shows and planning supplies. Many of these use cases don't need patient-identifiable data at all.
The common pattern: AI drafts, and a person decides. Starting with administrative work is lower risk than starting with clinical decisions.
What HIPAA means for an AI project
HIPAA applies to covered entities, which include most healthcare providers, health plans and clearinghouses, and to their business associates: vendors that handle protected health information (PHI) on their behalf. A few points matter most for AI work.
Business associate agreements. Any vendor that stores or processes PHI for you, including AI and cloud providers, needs a signed business associate agreement before PHI is shared. Not every AI product offers one, so check before the pilot, not after. Free public chatbots are generally not appropriate for patient data.
Minimum necessary. Share only the PHI a task actually needs. A scheduling assistant doesn't need a patient's diagnosis history.
De-identification. If a use case can work without identifiers, remove them. HIPAA describes two methods: Safe Harbor, which removes 18 specific types of identifiers, and Expert Determination, where a qualified expert assesses the re-identification risk.
Access controls and audit logs. People should see only what their role allows, and every access should be logged. This applies to AI too. An assistant that searches records has to respect the same permissions as the systems it reads from.
Security safeguards. Encrypt data in transit and at rest, host it securely, review vendor risk, and have an incident response plan.
Data use terms. Confirm in writing that your vendor won't use your data to train models for anyone else.
Two other things to keep in mind: state privacy laws can add requirements on top of HIPAA, and software that supports diagnosis or treatment decisions can fall under FDA oversight in some cases.
Design basics to plan from day one
- Human review for anything that goes into the medical record or to a payer.
- Clear disclosure when a patient is talking to an assistant rather than a person.
- An escalation path for urgent symptoms.
- Logs of inputs and outputs, stored and protected as PHI.
- Testing for errors and uneven performance across patient groups before launch, and monitoring after.
- A named owner for the system on both the clinical side and the IT side.
Where to start
Pick one administrative workflow with a clear cost, such as drafting prior authorizations or answering common patient questions. Map the data it touches, confirm your agreements with every vendor involved, and decide how you'll measure success. Pilot with one team, review the results with the people doing the work, and expand from there.
How Placix can help
Placix helps healthcare teams move from idea to a working pilot without cutting corners on patient data. We can help you choose the first workflow through an AI strategy and readiness review, then design and build it with the basics in this article from day one: minimum necessary data, role-based access, audit logs, human review and clear handoffs. For workflows that take actions, we build AI agents with human approval.
- Need people rather than a project? We recruit healthcare IT, data and AI talent, on contract or full-time.
- See the roles we place in Healthcare and Life Sciences.


