AI for Healthcare

Why AI matters in healthcare

Clinicians spend 2 hours on documentation for every 1 hour of patient care. The data on ambient AI scribes is clear: 30-50% reduction in documentation time, higher patient satisfaction, less burnout. Beyond the chart, AI moves revenue cycle, prior auth, and patient engagement.

The constraint: HIPAA, PHI, BAA agreements, FDA on clinical-decision tools. None of this is optional.

The core healthcare AI stack

Clinical documentation

• Abridge — our pick for ambient documentation.
• Microsoft Nuance DAX — enterprise scale.
• Notable — AI workforce for admin tasks.

Patient engagement

• Kapso — HIPAA-friendly WhatsApp messaging.
• Klara, Luma Health — patient communication platforms with AI.

Revenue cycle

• Notable, Olive AI (R1), Waystar — claims, denials, prior auth automation.

Medical imaging (specialist territory)

• Aidoc, Viz.ai, Heuro — radiology AI.

Deep dive 1: Ambient clinical documentation

The single biggest AI ROI in healthcare. Scribe listens to the visit, drafts the note in the EHR.

The workflow:

1. Clinician launches Abridge or DAX on phone or laptop.
2. Patient consents (verbal is fine in most states, document it).
3. Visit happens; clinician doesn't take notes.
4. AI drafts SOAP note in 2 minutes after visit ends.
5. Clinician reviews, edits, signs into EHR.

The constraint: pilot small. Have one clinician use it for 2 weeks. Review every note for accuracy. Roll out only when 95%+ notes need only minor edits.

Deep dive 2: Patient engagement and intake

AI handles the boring parts of patient comms. Appointment reminders, intake forms, post-visit check-ins, refill requests.

Pattern: Kapso or Klara handles WhatsApp/SMS. Bot triages incoming questions. Clinical questions escalate to a nurse. Admin questions handled inline. Cuts staff phone load by 40-60%.

Deep dive 3: Revenue cycle

Insurance denials are a tax. AI fights back.

The workflow:

1. Claim denied → AI parses the denial reason.
2. AI matches to your library of successful appeals for that reason.
3. AI drafts the appeal with the right CPT/ICD codes and prior auth documentation.
4. Coder reviews and submits.

Notable, Olive, or custom Claude prompts all work. The big systems use enterprise tools; SMB practices can build this on Claude + Sheets.

Deep dive 4: Prior authorization

The #1 source of clinician burnout administrative-wise. AI fills the form, attaches the chart notes, submits, follows up.

Governance

• HIPAA + BAA. Every AI vendor that touches PHI must sign a Business Associate Agreement. No exceptions.
• PHI in public LLMs. Never. Period. Use vendor APIs with zero-data-retention.
• Clinical accuracy. AI-drafted note must be reviewed and signed by the clinician. Never auto-submit.
• Bias and equity. Audit AI tools for performance across patient demographics.
• FDA. Clinical-decision-support tools (diagnostic AI) are FDA-regulated. Documentation tools generally aren't, but check.
• State laws. Some states have specific telehealth + AI disclosure rules.

30-60-90 day plan

Days 1-30: select an ambient documentation pilot (Abridge or DAX). One clinician, two weeks. BAA in place before day one.

Days 31-60: expand documentation to a department. Add patient engagement AI on the front desk (WhatsApp/SMS).

Days 61-90: add revenue cycle AI on denials. Measure: documentation time reduction, denial overturn rate, patient satisfaction.

Maturity model

• Level 1: One clinician pilots an AI scribe.
• Level 2: Department-wide ambient documentation. Patient engagement AI deployed.
• Level 3: Revenue cycle, prior auth, and patient triage AI in production.
• Level 4: AI-augmented clinical workflows across documentation, decision support, and ops.

Recommended prompts

• Knowledge base article from tickets — adapt for patient FAQs.
• Internal change comms — for clinician AI rollout.

Where to go next

• Browse Healthcare tools → (filter to Healthcare)
• For practice-level admin automation, see the AI Chief of Staff playbook.
• For revenue cycle and AP, see the Accounting playbook.