Back to QTips
Operategy

The Five AI Tools Already Running in Your Health System

Ambient scribes, EHR agents, general assistants, revenue cycle software and operations tools are already inside health systems. Here is what each does, what the organizations running them report, and the moves you make first.

6 min read

You run the service line, you know which clinics run late and why, and then a vendor puts a product name on a slide nobody in the room has heard before. AI reached your facility faster than the training did, and that failure belongs to the industry and never to you. The five kinds of tool below are already running inside health systems, and each one carries a short list of questions that makes a vendor prove it works.

The five kinds of tool you will meet

1. Ambient documentation

An ambient scribe is software that listens to the visit through a phone or a room microphone and drafts the note, and nothing reaches the chart until the clinician reviews and signs it.

Microsoft reported more than 100,000 clinicians using Dragon Copilot as of March 2026, and Stanford Health Care passed 1 million notes with it by May 2026. Samaritan Health Services reports a 38% drop. MaineHealth reports a 23% drop in note time per patient, with 70.3% of encounters documented through Abridge.

The leader's use. The audio leaves the room and goes somewhere, so ask who processes it, what the contract says about that handoff, and which physicians read their notes in full before signing.

2. The AI already inside your EHR

At the HIMSS health technology conference in March 2026, Epic said more than 85% of its customers use Epic AI, the agents inside your electronic health record. Each named agent does one narrow job: Art drafts notes and hospital course summaries, Penny works coding and denial appeals, and Emmie schedules and reschedules patients.

Prior authorization is the payer's approval a drug or procedure needs first, and Summit Health reports medication submission time down 42%, with 92% of AI generated responses accepted without edits. Ochsner Health reports more than 14,900 appointments rescheduled through Emmie, saving almost 750 hours of staff time. Rush University Medical Center reports a 58% drop in billing related service messages.

The leader's use. Some of these agents are already switched on inside your instance, so find out which ones, then ask for the baseline, the number your team logged before the tool arrived.

3. The general assistant your organization licenses

ChatGPT, Claude, Microsoft Copilot and Gemini all have enterprise versions for healthcare. OpenAI announced ChatGPT for Healthcare in January 2026 with AdventHealth, Cedars-Sinai, HCA Healthcare and UCSF Health among the first customers, and Anthropic launched Claude for Healthcare that month. Banner Health rolled out BannerWise, a private assistant built on Claude, to more than 55,000 employees across its 33 hospitals.

The American Medical Association surveyed nearly 1,700 physicians in March 2026 and found 81% use AI professionally, up from 40% in 2023. The top jobs: summarizing research and standards of care at 39%, drafting discharge instructions and care plans at 30%, chart summaries at 28%, and patient portal replies at 19%. The adoption already happened, and what you add is the shared standard pointing that effort at one goal.

The leader's use. You leave a meeting with a transcript nobody will read, so paste it into the assistant and ask for the decisions, the owners and the dates. Then paste the 40 page vendor proposal and ask for the five questions a CFO would raise.

No patient identifiers go into a general assistant until your organization has signed a business associate agreement for that tool, the contract that lets an outside vendor handle protected health information. PHI is anything in a record that identifies a patient. When in doubt, you take the names out first.

4. Revenue cycle and prior authorization

Revenue cycle is the billing path a claim travels from visit to payment. A denial is a payer's refusal to pay. These tools sort denials by payer and reason, draft appeals against the payer's own rules, and check eligibility before the claim goes out. Universal Health Services runs more than 50 AI applications across revenue cycle, reporting expense reductions above $1 million and revenue gains in the tens of millions. Intermountain Health has about 300 AI projects underway across the system.

The leader's use. The sorting and the first draft belong to the software. A wrong denial decision belongs to your organization and never to the vendor, so the last look before an appeal goes out has to be a person's.

5. Operations and capacity

Block time, bed capacity, staffing and the contact center all have AI products, and block time is the operating room hours a service owns on the schedule. CommonSpirit Health replaced its phone trees with a virtual agent and expanded to 30 virtual command centers.

A UI Health cardiology clinic in Chicago asked ChatGPT to build nurse schedules in two shift patterns. The tool produced workable schedules fast, could not account for absences, and needed a nurse manager to verify coverage and fairness. That study ran in the Journal of Nursing Administration in April 2026, and the split it describes is the one you will manage everywhere: the tool proposes, the manager decides.

The leader's use. Block release is the operating room time a service gives back, so model it before the meeting and bring three options to the scheduling fight with the numbers already run.

The four moves that work in every group

You are the warrior and AI is the sword, which means you supply the operational context and the accountability while the tool supplies the labor and the speed. These four moves hold across all five, and mastering them is the standard.

  1. Name the workflow before the tool. A tool pointed at prior auth or the Tuesday block meeting can be measured, and one pointed at innovation cannot.
  2. Draw the PHI line on day one. One page for every manager: which tools a signed agreement covers, which it does not, and what gets de identified.
  3. Take the baseline first. Every vendor number above arrived without one, and the numbers you report will not.
  4. Keep the human review step in writing. Clinicians review and sign, managers verify the schedule, a person decides the denial, and the rollout plan says so.

Where most organizations actually are

Black Book Research surveyed 182 hospital leaders in October and November 2025. Of those, 29% had implemented and enforced policies covering AI model inventory and sign offs, while 48% were still drafting. An audit trail is the record of which tool did what and who approved it, and only 22% felt confident they could produce a complete one within 30 days. Your organization is somewhere in that spread, and the audit trail is what proves where. AI governance took a median 4.2% of 2026 IT and quality budgets.

What we could not verify

The adoption and savings figures above come from the vendors and their customers, none of whom publish the baseline or the confounders, meaning the other changes underway at the same time. Ask for the before number. Practitioner threads on public forums were not reachable for this edition, and the sweep of what nurse managers, revenue cycle staff and practice administrators say lands in the next one.

Build this at your organization

Your next demo is already on the calendar. Email us the word [ TOOLS ] at info@thequadco.ai and the one page tool map, with the vendor question list to take into your next demo, is in your inbox by tomorrow. Want it drawn for your own organization, with your contracts and your PHI line? Reply and say so.

Sources

*Published September 2026. This Q-Tip is an operations aid and not legal, coding or compliance advice.*

Take it with you

Download PDF

[Q]Tips

One QTip a week, free, from September 15, 2026.

AI intelligence for healthcare leaders. No cost, no hype, one article a week.