AI and Data Trends in Healthcare Operations (2026): Nine Shifts Operators Should Track

TL;DR: Healthcare AI stopped being a clinical story and became an operational one. Eighty percent of health systems are exploring, piloting, or implementing generative AI in the revenue cycle, and the four most common provider use cases are all documentation or billing adjacent. The results are uneven: 63% of provider organizations use AI in revenue cycle operations and 15% report positive return, while only 3% of health care executives report significant ROI from AI overall. Meanwhile denials rose to an 11.6% initial rate, $48B in net revenue leaked in 2025, and providers spent $25.7B fighting claims that get overturned 70% of the time. This is a review of the nine trends that matter for operators, and the reason so much AI spend has not reached the margin yet.

The AI conversation in healthcare has moved from the exam room to the business office. That shift is visible in what organizations actually buy.

Bain and KLAS surveyed 228 US provider and payer executives in 2025 and found that 70% of providers and 80% of payers now have an AI strategy in place or in development. The four most common provider use cases were ambient notetaking, clinical documentation improvement, coding, and prior authorization. Not diagnosis. Not treatment planning. Documentation and getting paid.

That is a rational response to the economics. Kaufman Hall put adjusted year-to-date hospital operating margin at roughly 1.3% at the close of 2025, and margins through May 2026 were down about 5% year over year. When margins are that thin, the AI that gets funded is the AI attached to a hard dollar.

What follows is nine trends behind that, what the primary research actually says, and why the operational return has been slower than the adoption curve.

Key takeaways

• The revenue cycle is where healthcare AI went to work. HFMA's April 2025 survey of 519 CFOs and revenue cycle leaders found 80% exploring, piloting, or implementing generative AI for RCM, with about 40% already piloting or implementing.

• Adoption is well ahead of proven return. A separate HFMA poll of 101 provider organizations found 63% using AI or automation in the revenue cycle and only 15% reporting positive ROI. Deloitte found just 3% of health care executives reporting significant returns from AI.

• Denials keep getting worse and more expensive to fight. Kodiak Solutions put the 2025 average initial denial rate at 11.6% and net revenue lost to final denials and uncollected balances at $48B, up 25% year over year. Premier found providers spent $25.7B on claims adjudication, with roughly 70% of denials ultimately overturned.

• Regulators moved on algorithmic denials. California's SB 1120 took effect January 2025 and similar laws followed in Texas, Maryland, and others. CMS-0057-F decision deadlines began January 2026, with public prior authorization metrics from March 2026 and FHIR APIs required by January 2027.

• Ambient documentation is healthcare's first at-scale AI win. Abridge reached a $5.3B valuation in June 2025. Kaiser Permanente's Permanente Medical Group reported roughly 16,000 documentation hours saved in year one.

• Governance is the emerging bottleneck. Black Book found 29% of hospitals have implemented and enforced AI policies, 70% have had a pilot fail, and 22% could produce a complete AI audit trail within 30 days.

• Consolidation created the data problem. 82% of US physicians are now employed by hospitals or corporate entities, which means most groups run inherited, mismatched systems.

Trend 1: The revenue cycle is where healthcare AI actually went to work

HFMA's Pulse Survey of 519 CFOs and revenue cycle leaders, fielded in April 2025, found 80% exploring, piloting, or implementing generative AI tools for revenue cycle management, up from 58% merely considering it in 2023. About 40% are already piloting or implementing.

Asked where AI would have the biggest impact, 73% named prior authorization and 67% named denials and underpayment management. Those are the two workflows that consume the most administrative labor and directly determine whether earned revenue is collected.

The Bain and KLAS work confirms the pattern from the budget side, describing AI as moving from pilots to production with a focus on hard-dollar return. For operators this is the useful framing: the healthcare AI market has quietly become a revenue operations market, and the buying criteria have shifted accordingly.

Trend 2: The denials economy got worse in 2025

The numbers here are the most concrete in healthcare operations, and they are moving the wrong way.

Kodiak Solutions, drawing on data from more than 2,300 hospitals and 350,000 physicians on its platform, reported that the average initial denial rate rose to 11.6% in 2025 from 11.4% in 2024, and the median final denial rate rose from 2.5% to 2.7%. Net revenue lost to final denials plus uncollected patient balances reached $48B in 2025, up 25% from $38.6B in 2024. Clinical denials, meaning prior authorization and medical necessity, drove nearly all of the increase.

Experian Health's State of Claims 2025 survey found 41% of providers now face denial rates of 10% or higher, a figure that has risen every year since 2022.

The cost of fighting is its own line item. Premier's national survey of 280 hospitals across 23 states found providers spent $25.7B on claims adjudication in 2023, up 23%, averaging $57.23 per claim, with initial denial rates averaging around 15% and reaching as high as 49% at some organizations. Denied claims averaged three rounds of review at 45 to 60 days per round. And roughly 70% of denials are ultimately overturned and paid, which led Premier to estimate about $18B of that spend as potentially unnecessary.

Read that last pair of numbers together. Most denials are wrong, and providers spend billions proving it, one claim at a time, months after the service.

The operational implication is not that denials should be appealed faster. It is that the pattern is predictable. If 70% of denials get overturned, then the information needed to prevent most of them existed before submission. Catching it requires connecting clinical documentation, the payer's specific policy, the contract terms, and the historical outcome of similar claims, which are four systems that do not talk to each other in most organizations. We covered the mechanics of this in denial and underpayment detection, and our revenue cycle recovery demo shows what a prioritized appeal worklist looks like on fictional data.

Trend 3: Only 14% of providers use AI against denials

Here is the gap that defines the category. Experian Health found 67% of providers believe AI can improve the claims process, and 14% actually use AI to reduce denials. Among the minority that do, 69% report that it reduced denials or improved resubmission success.

So the tools work for most who deploy them, and almost nobody has deployed them. That is not a technology adoption curve problem. It is a prerequisite problem.

An AI that predicts whether a claim will be denied needs to see the clinical documentation, the coding, the payer policy in force on the date of service, the specific contract terms, the patient's eligibility and authorization status, and the organization's own denial history with that payer for that procedure. In most provider organizations those live in the EHR, the practice management system, a contract folder, a payer portal, and a clearinghouse, with no common key between them.

The result is that the AI available to most organizations sees a fraction of the picture and produces recommendations that the billing staff correctly ignores. The 14% is a measure of how many organizations have solved the data problem, not how many have heard of the technology.

Trend 4: The algorithmic denial backlash arrived from three directions at once

Payers deployed AI in utilization review before providers deployed it in billing, and 2025 and 2026 brought the response.

The Senate Permanent Subcommittee on Investigations reported in October 2024 that UnitedHealthcare's post-acute care denial rate rose from 8.7% in 2019 to 22.7% in 2022, with skilled nursing denials up roughly ninefold, coinciding with deployment of the nH Predict algorithm. Humana and CVS were also criticized. Class action litigation over nH Predict remained active into 2026, and senators sent new records demands to all three insurers in July 2026.

States moved next. California's SB 1120, the Physicians Make Decisions Act, took effect January 1, 2025, prohibiting health plans from basing medical necessity denials solely on AI or algorithms and requiring licensed clinician review. Maryland followed effective October 2025, Texas with SB 1188 effective September 2025, and Arizona, Connecticut, and Nebraska among others. Per KFF, roughly 25 states have issued AI-in-insurance guidance based on the NAIC's 2023 model bulletin.

CMS moved on timing and transparency. Under CMS-0057-F, as of January 1, 2026, impacted payers must decide expedited prior authorizations within 72 hours and standard requests within seven calendar days, and must give specific denial reasons. Public reporting of prior authorization metrics, including approval and denial rates and appeal outcomes, began March 31, 2026. The four required FHIR APIs, covering prior authorization, patient access, provider access, and payer to payer, are required by January 1, 2027.

For operators there are two consequences. The first is that payer denial behavior is about to become public data at the plan level, which changes contract negotiations. The second is that the prior authorization API mandate creates, for the first time, a programmatic channel for a workflow that currently consumes an average of 13 hours of physician and staff time per week per practice, per the AMA's 2025 survey of 1,000 physicians. That survey also found physicians complete an average of 40 prior authorizations weekly, 40% of practices employ staff dedicated exclusively to prior auth, and 94% say it contributes to burnout.

Organizations that have their authorization requirements, clinical criteria, and patient data connected will be able to use that API on day one. Organizations that do not will watch a genuine efficiency opportunity pass. Our patient access demo shows where referral-to-visit leakage accumulates, on fictional data.

Trend 5: Ambient documentation is the first at-scale win, and the EHR is absorbing it

The clearest AI success in healthcare is also the one with the cleanest data conditions, which is not a coincidence.

Abridge raised a $250M Series D at a $2.75B valuation in February 2025 when it was in more than 100 health systems, then a $300M Series E led by Andreessen Horowitz at $5.3B in June 2025 at more than 150 health systems. Named deployments include Johns Hopkins with 6,700 clinicians, Mayo Clinic, Duke, and UNC Health, with UPMC committing to scale past 12,000 clinicians.

The measured results come from named studies rather than vendor decks. Kaiser Permanente's Permanente Medical Group reported roughly 16,000 documentation hours saved in year one across more than 2.5 million uses. A Sutter Health rollout published in JAMA Network Open found mean note time per appointment fell from 6.2 to 5.3 minutes and burnout fell from 42.1% to 35.1%. Mass General Brigham reported clinicians saving about an hour of note entry, with 60% saying it made them more likely to extend their careers.

The category is now consolidating into the EHR itself. Microsoft launched Dragon Copilot in March 2025, merging Dragon Medical One with DAX ambient AI, which at launch was assisting more than 3 million ambient patient conversations monthly across 600 organizations. Epic unveiled its clinician assistant Art and patient assistant Emmie in August 2025, along with CoMET foundation models trained on 118 million patient records and 151 billion medical events from its Cosmos dataset, with Ambient AI Charting released around February 2026.

Worth naming the reason this worked when other applications stalled. Ambient documentation operates on a single self-contained input, the conversation in the room, and produces a single output into one system. It requires no reconciliation across systems, no entity resolution, and no agreement about what a metric means. Every operational AI application that has struggled fails at precisely the step ambient scribing gets to skip.

Trend 6: The 2026 story is ROI discipline

McKinsey's survey of roughly 150 US health care leaders, fielded in the fourth quarter of 2025, found 50% of organizations have implemented generative AI, up from 25% in late 2023 and 47% in 2024, with more than 80% of leaders reporting at least one use case deployed to end users. Integration with existing systems and internal capability gaps were the top barriers.

Deloitte's 2026 US Health Care Executive Outlook, based on 120 health plan and health system C-suite executives, is where the discipline problem shows. Sixty-four percent say AI could reduce costs through standardized and automated workflows and 56% expect significant operational efficiency value. Then: 51% either have not measured AI returns or say it is too soon, 31% report moderate returns, and 3% report significant ROI. Deloitte's healthcare CFO survey published in August 2026 adds that only 18% of CFOs at organizations scaling AI consistently measure its financial impact.

The HFMA poll of 101 provider organizations lands in the same place from the revenue cycle side: 63% using AI or automation, 15% reporting positive ROI.

There are two honest readings. One is that many deployments are genuinely not producing value. The other is that the value is real and unmeasured, because the baseline was never captured. Both are true in practice, and both are fixable by the same discipline: pick the metric before the pilot, capture the baseline, and measure the same way after. In a 1.3% margin environment, an unmeasured initiative is an initiative that gets cut.

Trend 7: Interoperability finally has volume

TEFCA looked ceremonial for its first year and then turned into infrastructure. Per an ASTP/ONC update in February 2026, there are 11 designated QHINs, more than 23,000 organizations live on the framework, and monthly exchange grew from roughly 10 million records in January 2025 to about 500 million by February 2026, with more than 1.5 billion documents exchanged since the December 2023 go-live.

Automation is producing money at the transaction layer too. The 2025 CAQH Index, published February 2026, estimates US healthcare avoided $258B in administrative costs in 2024 through electronic transactions, a 17% increase in cost avoidance, with roughly $21B in remaining savings available from full automation. Electronic prior authorization adoption rose from 31% to 40%, and a manual prior authorization costs providers $10.97 against $5.79 fully electronic.

The distinction that matters for operators is between exchange and understanding. TEFCA moves documents. It does not tell your finance team whether the procedure documented in that record was profitable, or whether the payer's behavior on this claim class has changed, or how this site compares to your other eleven. Getting the data is now considerably easier than making sense of it, which is a reversal from five years ago and a reallocation signal for where effort should go.

Trend 8: Consolidation built the data problem

The Physicians Advocacy Institute and Avalere Health report covering 2018 through early 2026 found that 82% of US physicians are now employed by hospitals or corporate entities including insurers, private equity firms, and pharmacy chains, with just 18% remaining in physician-owned practices. Hospitals and corporate entities own 63.9% of practices, against 29.8% in 2018.

That is the structural reason operational data in healthcare is so hard. A multi-site group is usually not one organization that grew. It is eight organizations that were acquired, each arriving with its own practice management system, its own EHR instance or configuration, its own charge master conventions, its own payer contracts, and its own way of naming a procedure.

The consequence shows up in every operating review. Nobody can answer whether site four is more profitable than site seven, because the two sites count a visit differently and allocate cost differently. Nobody can compare procedure economics across surgeons, because supply cost sits in one system and case time in another. Nobody can tell whether the new contract is performing, because expected reimbursement is not computed anywhere.

Our multi-site consolidation demo shows fourteen clinics measured on one yardstick, and our procedure economics demo shows profit by surgeon and insurer down to supply choice, both on fictional data.

Trend 9: Governance became the new bottleneck

Black Book Research surveyed 182 hospital leaders between October and November 2025 and found governance well behind deployment. Twenty-nine percent have implemented and enforced AI policies covering model inventory, lineage, and sign-offs, with 48% still drafting. Seventy percent have had at least one AI pilot fail due to weak endpoints, workflow misalignment, or data gaps. Only 22% could produce a complete AI audit trail for regulators or payers within 30 days. The median hospital allocates 4.2% of its 2026 IT quality and safety budget to AI governance, ranging from 6.8% at large systems to 2.3% at small hospitals.

The regulatory floor is rising underneath that. OCR's January 2025 HIPAA Security Rule proposal, the first overhaul in about two decades, would mandate encryption, multifactor authentication, and annual penetration testing, and would explicitly pull ePHI used in AI training data and prediction models into required security risk analyses. It remains a proposed rule as of mid-2026. The ACA Section 1557 final rule already bars discrimination through AI-based patient care decision support tools.

The security context is not abstract. HIPAA Journal's 2025 report counted 772 breaches of 500 or more records reported to HHS OCR, the most ever recorded, exposing protected health information of roughly 139.7 million individuals.

The 22% audit-trail figure is the one to sit with. It says that most organizations running AI cannot currently demonstrate what data a model saw or how it reached an output. That is a compliance exposure, and it is also why so many pilots fail: a recommendation nobody can trace is a recommendation clinical and billing staff will not act on.

What separates the 3% from everyone else

Put the trends together and the pattern is consistent. Ambient documentation succeeded because it works on one self-contained input. Denial prevention, prior authorization automation, procedure economics, multi-site comparison, and contract performance all stalled because they require reasoning across systems that were never designed to be read together.

That is the whole story of the gap between 80% adoption and 3% significant ROI. It is a context problem, not a model problem, and it is not unique to healthcare. The same gap appears in private equity, where 86% of dealmakers use generative AI and nearly 40% of GPs expect no material financial impact this year, and in commercial real estate, where 88% of firms are piloting and 5% have hit their program goals.

Four things have to be true before operational AI in healthcare produces a number a CFO will act on:

One definition per metric, owned centrally. A visit, an encounter, a case, a denial, expected reimbursement, and cost per case each need one organizational definition, precise enough that two analysts compute the same number, with each site's local source mapped to it in code rather than in a policy binder. This is finance work, it requires no technology, and everything downstream depends on it.

Entity resolution across systems. One patient, one provider, one payer, one procedure, regardless of how eight systems spell them. Without it, every cross-site comparison is approximate and every payer-level analysis is understated.

Contracts and policies made computable. Expected reimbursement cannot be calculated from a PDF in a shared drive. Until contract terms and payer policies are structured facts, underpayment detection is guesswork and denial prevention is reactive.

Lineage on every figure. Every number traceable to source records in seconds. This is what makes an appeal defensible, what makes a physician trust a profitability figure, and what answers the audit-trail problem that 78% of hospitals currently have.

None of this requires replacing the EHR. Organizations that try to fix reporting with a system migration usually spend two years and arrive with the same problem in a new interface.

How OutcomeCatalyst fits

OutcomeCatalyst builds the layer between the systems you already run and the AI you want to use. We connect your EHR, practice management, billing and clearinghouse data, scheduling, supply and inventory systems, and payroll, along with the payer contracts and policies that govern reimbursement, into one governed model your team and your agents can reason over. Nothing gets migrated and no system gets replaced.

For a health system, a multi-site group, or a practice platform that means:

• One definition per operational metric, so a visit, a case, and a denial mean the same thing at every site.

• Entity resolution across systems, so patients, providers, payers, and procedures resolve to one record regardless of source.

• Contracts made computable, so expected reimbursement is calculated rather than assumed, which is what turns underpayment detection from an audit into a control.

• Lineage on every number, which is what makes an appeal defensible and an audit trail producible.

• Live connections rather than exports, so the operating review reflects this week rather than last month.

That foundation is what most of the stalled use cases in this article were missing. Denial prediction, prior authorization readiness ahead of the 2027 API mandate, procedure-level profitability, and site-to-site comparison all become tractable once the underlying data resolves.

For the deployment side, which agents to assign first in a clinical operations setting and what must stay under human sign-off, see AI agents for healthcare operations and the broader architecture in AI agents, workflows, and knowledge graphs in healthcare. If you are weighing an internal build, we compared both paths in buy versus build an AI context layer, and covered sequencing in data unification in 2026.

See the full healthcare workflow set, or start a conversation about your organization.

Common questions about AI and data in healthcare operations

How many health systems are using AI in the revenue cycle?

HFMA's April 2025 Pulse Survey of 519 CFOs and revenue cycle leaders found 80% exploring, piloting, or implementing generative AI for revenue cycle management, with about 40% already piloting or implementing. A separate HFMA poll found 63% currently using AI or automation in revenue cycle operations.

Why do so few organizations report ROI from healthcare AI?

Two reasons. Deloitte found 51% either have not measured returns or say it is too soon, and only 18% of CFOs at AI-scaling organizations consistently measure financial impact, so much of the value is unmeasured rather than absent. The rest is genuine: applications that require reasoning across disconnected systems produce answers staff cannot verify and therefore do not act on.

How bad are claim denials in 2026?

Kodiak Solutions reported an average initial denial rate of 11.6% for 2025 and $48B in net revenue lost to final denials and uncollected patient balances, up 25% year over year. Premier found providers spent $25.7B on claims adjudication at an average of $57.23 per claim, with roughly 70% of denials ultimately overturned and paid.

Can AI actually reduce denials?

For organizations that have deployed it, mostly yes. Experian Health found 69% of the providers using AI against denials report reduced denials or improved resubmission success. The catch is that only 14% of providers use AI for this at all, because prediction requires connecting clinical documentation, coding, payer policy, contract terms, and denial history, which most organizations cannot do.

What does CMS-0057-F require and when?

As of January 1, 2026, impacted payers must decide expedited prior authorizations within 72 hours and standard requests within seven calendar days with specific denial reasons. Public reporting of prior authorization metrics began March 31, 2026. The four required FHIR APIs, including prior authorization, are required by January 1, 2027.

Is ambient AI documentation worth it?

The evidence is stronger here than anywhere else in healthcare AI. Kaiser Permanente's Permanente Medical Group reported roughly 16,000 documentation hours saved in year one, and a Sutter Health study in JAMA Network Open found note time per appointment fell from 6.2 to 5.3 minutes with burnout down from 42.1% to 35.1%. It also succeeds partly because it needs no cross-system reconciliation.

What is the biggest governance risk with healthcare AI right now?

Traceability. Black Book found only 22% of hospitals could produce a complete AI audit trail for regulators or payers within 30 days, and only 29% have implemented and enforced AI policies covering model inventory and lineage. The pending HIPAA Security Rule update would pull ePHI used in AI training and prediction into required risk analyses.

Do we need to replace our EHR to use AI operationally?

No. Most operational questions require data the EHR never held, including contract terms, supply cost, and staffing, so a migration does not solve them. The workable path is to leave systems of record in place and add a governed layer above them that resolves entities and standardizes definitions.

What is the highest-leverage first step for a multi-site group?

Writing down one definition per operational metric and mapping each site's local source to it. A visit, a case, a denial, expected reimbursement, and cost per case. Twenty to thirty definitions, agreed by the people who own the numbers. It requires no software purchase, and every comparison and every AI application downstream depends on it.

Sources

• HFMA Pulse Survey of 519 CFOs and revenue cycle leaders, fielded April 2025 (80% exploring, piloting, or implementing generative AI for RCM, up from 58% considering it in 2023; about 40% piloting or implementing): hitconsultant.net

• HFMA poll of 101 provider organizations, fielded October to November 2024, published May 2025 (63% using AI or automation in revenue cycle operations, 15% reporting positive ROI; 73% expect prior authorization to see the biggest impact, 67% denials and underpayment management): hfma.org

• Experian Health, State of Claims 2025, 250 healthcare professionals surveyed June to July 2025 (41% face denial rates of 10% or higher; 67% believe AI can improve claims, 14% use AI to reduce denials, 69% of those report improvement): experian.com

• Kodiak Solutions, State of the Healthcare Revenue Cycle, published March 2026, data from 2,300+ hospitals and 350,000 physicians (initial denial rate 11.6% in 2025; median final denial rate 2.7%; $48B in net revenue lost to final denials and uncollected patient balances, up 25% from $38.6B): fiercehealthcare.com

• Premier Inc. national survey of 280 hospitals across 23 states, published 2025 ($25.7B spent on claims adjudication in 2023, up 23%; $57.23 average per claim; initial denials averaging about 15%; roughly 70% of denials overturned; about $18B potentially unnecessary spend): premierinc.com

• 2025 CAQH Index, published February 2026 ($258B in administrative costs avoided in 2024 through electronic transactions; about $21B in remaining savings; electronic prior authorization adoption up from 31% to 40%; manual prior auth $10.97 versus $5.79 electronic): globenewswire.com

• US Senate Permanent Subcommittee on Investigations, October 2024, via Healthcare Dive (UnitedHealthcare post-acute denial rate from 8.7% in 2019 to 22.7% in 2022; skilled nursing denials up roughly ninefold alongside nH Predict deployment): healthcaredive.com

• Fenwick and KFF on state regulation of AI in utilization review (California SB 1120 effective January 2025; Maryland, Texas SB 1188, Arizona, Connecticut, Nebraska following; roughly 25 states with AI-in-insurance guidance based on the NAIC 2023 model bulletin): kff.org

• CMS Interoperability and Prior Authorization Final Rule (CMS-0057-F) fact sheet (72-hour expedited and 7-day standard decisions from January 2026; public prior authorization metrics from March 31, 2026; four FHIR APIs required by January 1, 2027): cms.gov

• AMA 2025 Prior Authorization Physician Survey of 1,000 physicians, released May 2026 (40 prior auths per week, about 13 hours of physician and staff time; 40% employ dedicated prior auth staff; 94% say it contributes to burnout; 74% report increased denials over five years): ama-assn.org

• Abridge funding coverage via Fierce Healthcare, February and June 2025 ($250M Series D at $2.75B in 100+ health systems; $300M Series E led by Andreessen Horowitz at $5.3B in 150+ health systems): fiercehealthcare.com

• Kaiser Permanente Permanente Medical Group via AMA and Becker's (roughly 16,000 documentation hours saved in year one across more than 2.5 million uses) and Sutter Health via JAMA Network Open (note time per appointment 6.2 to 5.3 minutes; burnout 42.1% to 35.1%): ama-assn.org

• Microsoft Dragon Copilot launch, March 2025 (3M+ ambient patient conversations monthly across 600 healthcare organizations at launch) and Epic UGM 2025 coverage via CNBC (Art and Emmie assistants; CoMET models trained on 118M patient records and 151B medical events from Cosmos): news.microsoft.com

• McKinsey survey of approximately 150 US healthcare leaders, fielded Q4 2025 (50% of organizations have implemented generative AI, up from 25% in late 2023 and 47% in 2024; more than 80% of leaders report at least one use case deployed): mckinsey.com

• Deloitte 2026 US Health Care Executive Outlook, 120 health plan and health system C-suite executives, published December 2025 (64% say AI could reduce costs through automated workflows; 51% have not measured returns or say it is too soon; 31% moderate returns; 3% significant ROI) and Deloitte healthcare CFO survey, August 2026 (18% of CFOs at AI-scaling organizations consistently measure financial impact): deloitte.com

• Bain & Company and KLAS Research survey of 228 US provider and payer executives, published October 2025 (70% of providers and 80% of payers have an AI strategy in place or in development, both up from 60%; top provider use cases are ambient notetaking, clinical documentation improvement, coding, and prior authorization): bain.com

• ASTP/ONC TEFCA update, February 2026 (11 designated QHINs, 23,000+ organizations live, monthly exchange from about 10M records in January 2025 to about 500M by February 2026, 1.5B+ documents since December 2023): healthit.gov

• HIPAA Journal 2025 Healthcare Data Breach Report (772 breaches of 500 or more records reported to HHS OCR in 2025, the most ever; PHI of approximately 139.7 million individuals exposed): hipaajournal.com

• Physicians Advocacy Institute and Avalere Health physician employment report covering 2018 to early 2026 (82% of US physicians employed by hospitals or corporate entities; 18% in physician-owned practices; hospitals and corporate entities own 63.9% of practices versus 29.8% in 2018): physiciansadvocacyinstitute.org

• Kaufman Hall National Hospital Flash Report covering 1,300+ hospitals (adjusted year-to-date operating margin around 1.3% at the close of 2025; margins through May 2026 down about 5% year over year): kaufmanhall.com

• Black Book Research survey of 182 hospital leaders, fielded October to November 2025, via Advisory Board (29% have implemented and enforced AI policies covering model inventory and lineage, 48% drafting; 70% have had an AI pilot fail; 22% could produce a complete AI audit trail within 30 days; median 4.2% of IT quality and safety budget to AI governance): advisory.com

• Reed Smith on HHS AI guidance and the January 2025 HIPAA Security Rule proposal (encryption, multifactor authentication, annual penetration testing; ePHI used in AI training data and prediction models pulled into required security risk analyses; proposed rule as of mid-2026): reedsmith.com

OutcomeCatalyst connects the systems you already run into a governed intelligence layer your team and your agents can reason over. Demos on this site use fictional data. To see this on your own organization, start a conversation.