Doctors Spend More Time on Paperwork Than on Patients: How AI Workers Give It Back

By Delos Workers team — 2026-11-11

Physicians lose 15.5 hours a week to non-clinical work. See how AI Workers handle prior authorizations, referrals and coding safely, plus a 90-day plan.

The scarcest resource in your hospital is clinician attention, and it is being spent on administration. Every form, authorization and referral that lands on a physician's desk is time taken from a patient. For hospital executives, this is not only a morale issue: it is a capacity, safety and cost issue.

A burden measured in hours and in people

The AMA reports that physicians spend roughly 15.5 hours a week on non-clinical work. In the US, administrative costs are estimated at nearly 8% of total health expenditure against 1 to 3% in other countries, and in hospitals they can exceed 40% of the cost of delivering care (Cohere Health, April 2025). The human cost is the part executives feel most: nearly half of physicians who leave the profession cite administrative burnout, and burned-out physicians are more than twice as likely to be involved in patient-safety incidents (same source). Every hour recovered from paperwork is therefore also a safety investment.

What early evidence shows

Ambient AI documentation is the most studied use case. A study indexed on PubMed (2025, ID 41198484) found that ambient AI cut documentation time sharply, for example median progress note time falling from 128 seconds to 27 seconds in one comparison. In another report, clinician self-reported burden fell from 52% to 39% after one month with an AI scribe (186 clinicians).

The AMA also finds that seven in ten respondents see opportunities for AI to automate clinical and administrative tasks that drive burnout. Adoption is already mainstream: 72% of employed physicians and 64% of those in private practice use health AI (AMA, May 2025). These are early results from specific settings; check the original studies before generalizing them to your own hospital.

Beyond the scribe: administrative Workers

Documentation is only the first layer. The bigger prize is the administrative chain around the patient.

• Prior authorizations and claims. A Worker assembles the clinical evidence, fills payer forms and tracks status, returning only the denials that need a human.
• Scheduling and referrals. A Worker manages reminders, rescheduling and referral follow-up through the channels patients already use.
• Coding and billing support. A Worker proposes codes from the note and flags gaps before a claim leaves, reducing rework.

The common thread: the Worker prepares, the professional decides and signs.

An illustrative scenario

A regional hospital group wants to lower the time nurses and physicians spend on forms. A Delos Worker, connected to scheduling and document systems with strictly scoped access, prepares referral packets and authorization files overnight. Staff review and sign in the morning. This is an illustrative scenario, not a customer result.

Non-negotiables for healthcare leaders

Patient data demands more than good intentions:

1. Data residency and encryption.
2. Role-based access, scoped per Worker.
3. Human sign-off on anything clinical.
4. A complete audit trail.
5. Compliance with GDPR and, for the EU, the AI Act, which classifies many health uses as high risk.

Choose a platform designed for these constraints from day one, not retrofitted.

Why the window is open now

Staffing shortages and burnout push hospital leaders to look for relief that does not require hiring. Clinicians themselves are asking for it: with a majority already using some form of health AI, the debate has shifted from permission to governance. Organizations that set clear rules now will scale safely; those that wait will face shadow AI use with no oversight. Regulation is also tightening, which makes documentation, human oversight and traceability a design requirement rather than an afterthought.

Common pitfalls to avoid

• Starting with the hardest clinical use case. Begin with administrative workflows where errors are visible and reversible.
• Ignoring the workflow. A tool that adds clicks to the EHR will be abandoned, however accurate it is. Involve nurses and secretaries from the first week.
• Weak consent and data handling. Define what data each Worker may read, where it is processed and how long it is retained.
• No feedback loop. Clinicians must be able to flag errors in one click, and those flags must reach the owner of the Worker.

A 90-day plan

• Days 1 to 15: choose one service (for example outpatient referrals) and measure the minutes spent per case and the denial or rework rate.
• Days 16 to 45: run a Worker in assisted mode with a small team and a named clinical lead.
• Days 46 to 90: review safety, time saved and staff feedback, then decide on extension to a second service. Share results with the medical board, not only with IT.

Next step

Delos AI Workers are built for regulated environments. A short discovery call can identify the one administrative process where your teams lose the most hours. Contact the Delos team to set it up.

Sources: AMA (May 2025; Sept 2026 survey summary); Cohere Health (Apr 2025); PubMed 41198484 (ambient AI documentation); Medical News in Brief (AI scribe study).