The Cost of AI in the ED: Who Actually Pays?

The vendor closed the demo with a slide titled "ROI in Six Months." The numbers on the slide were clean. The cost was $74 per provider per month for the enterprise tier, scaled across two hundred and eighty clinicians, with a published time savings of thirteen minutes per provider per day. The ROI calculation converted those minutes into recoverable RVUs at the institution's average per-RVU reimbursement rate. The bottom line landed at $2.1 million in projected first-year value against $234,000 in subscription cost. The CFO smiled. The CMIO nodded. The ED medical director — me — asked a question. Who pays when the time savings do not convert to RVUs? Who pays for the integration engineer? Who pays for the eighteen months of after-hours support cases? Who pays for the override-review process? Who pays for the resident curriculum to use the tool well? The room got quiet. The vendor changed the subject. The procurement went to committee for another six months.

That conversation is happening in every emergency department in the country in 2026, and the honest answer is that nobody at the table has a clean accounting of who actually pays for AI in the ED. The contract price is the smallest piece of the total cost. The reimbursement pathway is narrower than the vendor implies. And the unfunded operational burden — the integration, the training, the documentation overhead, the governance committee work, the failure-mode review — lands almost entirely on the physician and nursing workforce that the tool was supposed to relieve. Emergency medicine leaders need to understand the full cost stack before they sign, and the 2026 reimbursement landscape before they assume the tool will pay for itself.

What AI Actually Costs in 2026

The published 2026 pricing on clinical AI in the ED falls into three tiers, and the difference between them matters for procurement.

Per-provider subscription tools, dominated by ambient documentation scribes, run roughly $99 to $299 per provider per month at list price, with the American Academy of Family Physicians citing $150-$200 per month as the typical figure for AI-powered scribes. AWS HealthScribe runs $0.10 per minute of audio, which works out to about $1,500 for one hundred fifteen-minute visits. Common per-tier pricing models offer a basic tier around $199 per month for five hundred minutes and a pro tier around $449 per month for two thousand minutes. Enterprise licenses for health systems run $20,000 to $100,000 or more annually, with one-time setup and onboarding fees of $1,000 to $5,000 layered on top. Specific vendors like Freed publish tiers from $39 per month at the starter level through $119 per month at the premier level, with custom enterprise pricing for groups.

The second tier is the FDA-cleared clinical AI device. Viz.ai's large-vessel-occlusion stroke detection system receives roughly $1,000 per use under the CMS New Technology Add-on Payment program. IDx-DR for diabetic retinopathy screening receives around $40 per use. Caption Guidance for cardiac ultrasound receives an NTAP add-on. These are the AI tools that have successfully navigated the CMS NTAP pathway and have a defined Medicare payment for the additional cost of the AI augmentation above the standard DRG.

The third tier is the CPT-code-supported AI service. CPT 2026 expanded the assistive and augmentative AI section with new Category III codes — 0972T for multi-spectral burn-wound imaging paired with an algorithmic detection device (officially in the 2026 code book), and 0992T and 0993T for noninvasive cardiac risk assessment derived from augmentative software analysis of perivascular fat with and without concurrent CT, both effective January 1, 2026. The legacy Category III codes 0623T through 0626T for coronary atherosclerotic plaque quantification will be deleted effective January 1, 2026 and replaced by the permanent CPT 75577, which CMS proposed for Medicare payment in the 2026 Physician Fee Schedule.

That sounds like coverage. It is not. It is a small set of specific clinical applications, and the AI tools an emergency department is actually evaluating in 2026 — ambient documentation scribes, AI triage augmentation, AI clinical decision support, AI imaging interpretation outside the narrow NTAP-approved indications — are almost entirely outside the reimbursement framework. The hospital is the buyer of last resort.

The Hidden Cost Stack

The line-item subscription price is the easy number. The real cost of clinical AI in the ED is roughly four times the contract figure once you account for the operational stack that the vendor does not include in their ROI model.

Integration with the electronic health record is the single largest hidden cost. The ambient AI scribe that ingests audio and produces a draft note has to integrate with the institution's EHR, which means an Epic or Cerner integration engineer at minimum, an HL7 or FHIR API contract, a documentation security review, and a multi-month go-live process. The published cost-of-implementation analyses out of the 2026 healthcare-AI consulting literature put this number at $50,000 to several hundred thousand dollars depending on the institution's existing infrastructure, and it does not show up on the vendor's per-provider pricing page.

Training and adoption support is the second hidden cost. The 2026 multi-site JAMA study on ambient AI scribes was clear that the time-saving gains were concentrated in the clinicians who used the tool in more than half of their patient encounters, and only thirty-two percent of clinicians in that cohort hit that threshold. Getting the rest of the department to that adoption rate requires protected training time, post-go-live support cases, workflow redesign with operational and informatics staff, and ongoing champion-clinician engagement that the institution funds out of pocket. The 2026 healthcare-AI implementation literature reports total implementation costs typically multiple times the annual license fee in the first year.

Documentation overhead is the third hidden cost. Every AI-generated clinical note must be reviewed and signed by the physician, who is liable for any omissions or hallucinations the algorithm introduces. The override-and-edit time on AI scribes runs three to five minutes per note in the published implementation studies. At thirty patients per shift, that is ninety to one hundred fifty minutes of editing work the physician now does that was not in the workflow before, partially offsetting the documentation time savings the tool was sold on. The institution does not pay for that time. The physician absorbs it.

Governance and compliance is the fourth hidden cost. Every clinical AI tool the institution deploys requires AI governance committee work, FDA Predetermined Change Control Plan review when the vendor updates the model, departmental policy updates, malpractice insurance carrier notification, and ongoing post-market monitoring. The 2026 NEJM AI policy literature is converging on the position that this work needs dedicated institutional staffing — a chief AI officer, an AI governance manager, a clinical informatics analyst — and at most institutions today, that work is done in spare hours by the CMIO and the ED medical director on top of their existing roles, with no protected time and no incremental compensation.

Who Actually Pays

Stack the costs honestly and the question of who pays for AI in the ED has a clear answer for 2026.

The hospital pays the contract price. That is the visible cost on the operating budget, and the procurement committee evaluates it against the published ROI model the vendor provides.

Medicare pays a small, narrow set of NTAP add-on payments for a small, narrow set of FDA-cleared AI devices, capped at three years of new-technology status. The CPT 2026 expansion adds reimbursement for specific assistive AI services, primarily on the imaging and risk-stratification side, that do not capture the bulk of clinical AI deployment in the ED. The 2025-2026 NEJM AI policy work on fixing AI payment systems argues explicitly that current Medicare structures pay primarily for AI software that augments billable physician services and largely fails to reimburse AI tools that operate as standalone augmentation of clinical workflow — which is exactly the category most emergency department AI deployment falls into.

Commercial payers follow Medicare. They cover what Medicare covers. They argue about everything else. The 2026 reporting on insurer-hospital AI billing disputes documents a growing tension where hospitals deploy AI tools at scale and submit claims under existing codes, and insurers apply AI-based claim-review tools to deny those claims at scale. The economic friction is real, and it lands disproportionately on the institutions that can least afford it — rural and safety-net emergency departments with thin margins.

The physician pays the unfunded operational tax. Override review, documentation editing, governance committee time, training, residency curriculum integration, and the cognitive overhead of operating tools that sometimes hallucinate and sometimes generalize poorly to the patient population. None of that is on the vendor's pricing page. All of it lands on the clinician. The 2026 burnout literature makes the point bluntly: AI tools that are deployed without explicit institutional commitment to protect the time they save will absorb the savings into productivity expectations within twelve months. For the leadership framing on how the time-savings-versus-productivity-creep dynamic plays out in observation and care-pathway design, see my Medium essay Observation Units Are a Leadership Tool — Not a Billing Hack.

The patient pays at the margins. Higher hospital costs eventually reach the payer mix, the chargemaster, and the patient's out-of-pocket exposure. The 2026 healthcare economics literature is reasonably clear that clinical AI is currently a net cost driver at the institutional level rather than a net cost saver, and the institutions that have not yet reckoned with that fact will eventually pass the cost through.

What You Should Be Doing Now

Demand the four-times multiplier on every AI procurement decision. If the vendor quotes you $234,000 in annual subscription cost, model the real first-year cost at $900,000 to $1.1 million across integration, training, governance, and absorbed clinician overhead. If the ROI still works at that number, the deal is real. If it only works at the line-item subscription cost, the deal is not real and the institution will discover that twelve to eighteen months in.

Map the reimbursement pathway before signing. For each AI tool, ask the vendor in writing which CPT code, NTAP application, or other reimbursement pathway applies, and require them to produce the documentation. If the answer is "the tool pays for itself through documentation time savings," that is a productivity argument, not a reimbursement argument, and the institution needs to put the productivity protection in writing before signing.

Protect the time. The institution needs a written commitment that the time savings from clinical AI tools will be returned to the clinician — through reduced patient load, protected administrative time, or reduced after-hours documentation — rather than absorbed into additional throughput. Without that commitment, the burnout-and-attrition cost of the deployment will exceed the financial cost within two years.

Build the governance funding into the procurement. The institution should not be deploying clinical AI without funded governance staffing — a chief AI officer or equivalent, a clinical informatics analyst, and a clear AI committee charter with protected time for the participating physicians. That cost is part of the AI deployment, not a separate line item.

Dr. Chet's Take

I have negotiated AI procurement contracts at academic medical centers, military medical commands, telemedicine networks, and air medical programs. The pattern is consistent across all four. The institutions that treat AI as a finance line item lose money. The institutions that treat AI as an operational deployment with a finance line item underneath get the published outcomes. The difference is in how the institution accounts for the unfunded operational stack — the training, the integration, the governance, the override review, the absorbed clinician time — and whether the institution funds it explicitly or implicitly transfers it to the clinical workforce.

The 2026 cost numbers are real. The reimbursement gap is real. The operational stack is real. The vendor's ROI slide is partial. The CFO's spreadsheet is partial. The honest accounting of who pays for AI in the ED requires the ED medical director at the table, with the operational authority to push back on the partial accounting and the institutional support to walk away from a procurement that does not hold up under the full math. That is not an adversarial role. It is the role the institution needs filled if the AI deployment is going to deliver on the published evidence rather than create new costs the institution will be carrying for the next decade.

AI Won't Wait. Neither Should You.

If your institution is evaluating clinical AI procurement without the full cost stack on the table — integration, training, governance, override review, absorbed clinician time, and reimbursement pathway — you are exposed. Consider enrolling in my course: AI in Emergency Medicine: Becoming AI Bulletproof. The course walks through the four-times-multiplier procurement model, the CPT 2026 and NTAP reimbursement landscape as it applies to ED-specific AI tools, the institutional contract language that protects clinician time from productivity creep, the AI governance funding template we use with regional consortiums, and the post-deployment financial review framework that catches partial ROI within six months rather than twenty-four.

Learn more: AI in Emergency Medicine: Becoming AI Bulletproof

If you're an emergency physician (or any clinician treating patients daily) trying to understand how AI will actually impact your clinical practice — not just the hype — I put together a free practical guide. You can download it here: AI in EM Survival Guide.

Dr. Chester "Chet" Shermer, MD, FACEP is a Professor of Emergency Medicine, TeleHealth, HEMS and Critical Care Transport, and State Surgeon for the Army National Guard. He is the founder of Global MedOps Command and the creator of AI in Emergency Medicine: Becoming AI Bulletproof.

Also on Medium: Read more from Dr. Shermer on Medium.

Sources

1. MedCentral, "CPT 2026 Updates Expand Lab, Category III, and AI Codes," September 2025, https://www.medcentral.com/coding-reimbursement/cpt-2026-updates-expand-lab-category-iii-and-ai-codes
2. Complizen, "CMS NTAP for AI Medical Devices: 2025 US Inpatient Reimbursement Guide," August 2025, https://www.complizen.ai/post/what-is-cms-ntap-ai-device-reimbursement-guide
3. Freed, "Cost of AI Medical Scribes: Pricing Guide and ROI Analysis [2026]," 2026, https://www.getfreed.ai/resources/cost-of-ai-scribes
4. NEJM AI (PMC), "Catalyzing Health AI by Fixing Payment Systems," November 2025, https://pmc.ncbi.nlm.nih.gov/articles/PMC12900248/
5. The Daily Record, "Insurers and hospitals turn to new AI for age-old billing disputes," March 2026, https://thedailyrecord.com/2026/03/12/insurers-hospitals-ai-billing-payments/