Where AI tools fit in healthcare, by kind of organization

Six kinds of healthcare organizations and 44 problems that cost them money or put them at risk, each with a sourced number. It shows which of the problems an AI tool can help with without reading patient data and how each tool's work is checked for accuracy, and it ends with an explorer of every problem and the services that address it.

By Tyler KentUpdated

Part one

Six kinds of organizations

A tool that fits a 12-person dental office is wrong for a six-hospital system. The two differ in size, in how they are paid and in the rules they answer to, so we split healthcare into six kinds of organizations: from the largest health systems to small practices and to post-acute and senior care organizations, plus the companies and associations that serve them.

  1. Health systems

    Systems with six or more hospitals, and teaching hospitals.

    3,567 of the 5,121 hospitals the American Hospital Association counts as community hospitals belong to a health system, counting systems of any size. Source: AHA.

    Its 8 problems and the tools for them
  2. Community hospitals

    Independent and small-system hospitals, rural hospitals and critical access hospitals.

    1,797 of those 5,121 hospitals are rural, and some of them also belong to a system. Source: AHA.

    Its 7 problems and the tools for them
  3. Health centers

    HRSA-funded health centers and look-alikes, free and charitable clinics, local health departments.

    1,356 health center organizations served more than 32.7 million patients in 2025. Source: KFF.

    Its 7 problems and the tools for them
  4. Practices

    Physician groups, dental offices, behavioral health and therapy practices, other small clinical businesses.

    440,667 practice firms have employees, and 92.3% of them have fewer than 20 employees. Source: US Census Bureau.

    Its 7 problems and the tools for them
  5. Post-acute and senior care

    Skilled nursing, home health, hospice, senior living and PACE.

    About 14,400 freestanding skilled nursing facilities and 6,706 hospices served Medicare patients in 2024, and more than 12,000 home health agencies were certified to take part in Medicare. Source: MedPAC (also Chapter 8, Chapter 10).

    Its 8 problems and the tools for them
  6. Health-tech and associations

    Digital health companies, revenue cycle and staffing firms, associations and medical societies.

    Nearly 300 revenue cycle firms and billing departments belong to HBMA, one billing association. Source: HBMA.

    Its 7 problems and the tools for them

We left out health insurers, pharmacies, labs, and drug and device makers. This analysis covers the organizations that deliver care and the firms that work for them.

Part two

The 44 problems we found

For each of the six kinds of organizations, we used our own judgment to pick seven or eight problems that cost a lot of money or carry serious risk, then sorted all 44 into six areas, from money and payers to daily operations. Here is how we found them.

  1. Three seats

    For each kind of organization, we looked for the problems that weigh on three people: the chief executive or administrator, the finance leader and a manager who runs the front line.

  2. Sources you can check

    We took the problems those leaders report in trade group surveys, federal rules and data files, research and industry reports, and journals.

  3. Seven or eight for each kind

    We kept seven or eight problems for each kind, picked by our own judgment of the money or risk at stake.

  4. One number, checked

    For each problem we found one number, such as a cost, a rate or a deadline, and checked it at its source in October 2026. Every number links to that source in the explorer at the end of this page.

Fig. 1

The 44 problems, by kind of organization and by area

  • Each square is one problem
  • A problem in another area

Step 1 of 4

Seven or eight problems for each kind of organization

Every problem on the list costs money or carries risk, and each has a number we checked at its named source, such as a federal rule or data file, a trade group survey, a research or industry report, or a journal. Across the six kinds of organizations, that makes 44 problems.

Step 2 of 4

13 of the 44 problems are about money and payers

That is more than any other area. The numbers behind them include hospital drug expenses up 13.6% in 2025, 41.2% of rural hospitals operating in the red, and nearly half of community health centers having less than 90 days of cash on hand.

Step 3 of 4

Ten problems are about rules, surveys and audits

Many come with a deadline or a penalty. A nonprofit hospital that misses its community health needs assessment owes a $50,000 excise tax for each year it fails, and 44.3% of nursing homes were fined at least once in three years.

One deadline is already set for websites: from May 11, 2027, HHS-funded organizations with 15 or more employees must make their websites and apps meet WCAG 2.1 AA, the technical standard that lets people with disabilities use them.

Step 4 of 4

The other 21 problems fall in four smaller areas

Patients, referrals and customers (8), staffing (5), paperwork and back office (5) and daily operations (3). Among them, medical assistants had the highest turnover of the health center jobs in a 2025 NACHC staffing survey, at 24%, and physicians complete an average of 40 prior authorizations a week.

Every kind of organization has at least one problem about money and payers and one about rules, surveys and audits on its list.

Kent Applied analysis, October 2026. Every problem's number and source are in the explorer.

Show Fig. 1 as a table
The 44 problems by area and by kind of organization.
AreaHealth systemsCommunity hospitalsHealth centersPracticesPost-acute and senior careHealth-tech and associationsAll
Money and payers23222213
Rules, surveys and audits32112110
Patients, referrals and customers0111238
Paperwork and back office0012115
Staffing1111105
Daily operations2010003
All areas87778744

The money problems show up first in costs. For hospitals, every major category of expense grew in 2025, and drugs grew fastest.

Fig. 2

Hospital drug expenses grew 13.6% in 2025, nearly twice as fast as total expenses

Growth in US hospital expenses in 2025, by category, as AHA reports it from Strata Decision Technology benchmark data

Show the numbers as a table
Hospital drug expenses grew 13.6% in 2025, nearly twice as fast as total expenses
CategoryValue
Drug expenses13.6%
Supply spending9.9%
Total expenses7.5%
Workforce costs5.6%

Part three

How many healthcare businesses already use AI

About one in four healthcare businesses told the Census Bureau they had used AI in the two weeks from September 7 to 20, 2026. The Census Bureau's number covers its health care and social assistance sector, which also includes child care. That is close to the rate for all US businesses, and half the rate for information firms such as publishers, software makers and telecom companies.

Source: US Census Bureau, Business Trends and Outlook Survey, sector estimates and national estimates, published October 8, 2026.

Fig. 3

Health care and social assistance firms use AI at about the US rate, half the top sector's

Share of US employer businesses that said they had used AI somewhere in their business between September 7 and 20, 2026: the four sectors with the highest use, health care and social assistance, all businesses, and the two sectors with the lowest use

Show the numbers as a table
Health care and social assistance firms use AI at about the US rate, half the top sector's
CategoryValue
Information50.2%
Professional, scientific and technical services42.7%
Finance and insurance40.1%
Educational services39.9%
Health care and social assistance24.8%
All US businesses23.8%
Accommodation and food services10.1%
Agriculture, forestry, fishing and hunting9.5%

Buyers start with vendors they know

In an HFMA survey of 233 health systems in 2025, nearly 80% said an existing vendor would have a significant advantage over a new one in an AI pilot. Source: HFMA, Health system adoption of AI outpaces internal governance and strategy.

Many adults already use AI for health information

32% of US adults turn to AI tools for health information and advice, a KFF poll found in 2026. On Google searches that showed an AI summary, people clicked through to a regular result on 8% of visits, against 15% when there was no summary, so fewer of them reach an organization's own website. Sources: KFF; Pew Research Center.

Where the rules already limit what AI can do without a person

  • California. AI-written messages to patients about clinical information need a disclaimer and a way to reach a person, unless a licensed or certified provider reads and reviews them first. Source
  • Federal nondiscrimination rule (Section 1557). Machine translation of critical or technical health content must be reviewed by a qualified human translator. Source
  • New York City. An automated tool that screens job applicants needs a bias audit within the past year. Source
  • NIH. A grant application substantially developed by AI does not count as the applicants' original idea. Source

Each rule limits what AI can do without a person: a provider's review, a qualified translator, a bias audit, or the applicants' own ideas. The tools in the explorer below work the same way: a person approves what goes out.

Part four

Which problems need patient data

Under HIPAA, a tool that reads patient records needs a business associate agreement, the contract an outside company signs before it handles patient data for you. For most of the 44 problems, a useful AI tool never reads patient data, so no agreement is needed.

Fig. 4

The 44 problems, by whether an AI tool for them needs patient data

  • Needs no patient data
  • Needs patient data for part of the work
  • Needs patient data from the start
  • No AI tool advised

Step 1 of 5

30 of the 44 problems need no patient data

For 30 of the 44 problems, an AI tool can help while reading only contracts, policies, staff records, public data, websites and summary reports. A contract register, a website check or a weekly brief built from summary reports never sees anything that identifies a patient.

Step 2 of 5

Ten problems need patient data for part of the work

For these, the work can begin without patient data, and a later stage needs it. With a health system's claim denials, for example, a tool can start by tracking contract terms and appeal deadlines, and only reviewing the denied claims themselves needs patient data. With referrals to post-acute and senior care organizations, it can start by counting referrals by source, and only reading the referral packets needs patient data.

Step 3 of 5

Three problems need patient data from the start

Prior authorization and claim denials at practices, and the hospice cap in post-acute and senior care (Medicare's yearly limit on what it pays each hospice), need patient records, claims or day counts for each patient from the first step. Even a list of admission and discharge dates is patient data: under HIPAA's Safe Harbor method, the standard way to remove identifiers from data, every part of those dates except the year counts as an identifier.

Step 4 of 5

No AI tool is advised for hiring at practices

AI that ranks job applicants needs a yearly bias audit in New York City and falls under California's rules on automated decision systems. A practice's applicant tracking system already posts jobs and schedules interviews, so an AI tool would add too little to be worth those audit and legal requirements.

Step 5 of 5

Patient data matters most for practices and for post-acute and senior care

All three problems that need patient data from the start belong to practices or to post-acute and senior care. Those two kinds of organizations also have the fewest problems that need no patient data: four of seven for practices and four of eight for post-acute and senior care. Health systems and community hospitals each have six such problems, and health centers and health-tech firms each have five.

Kent Applied analysis, October 2026. A problem counts as needing no patient data when every tool advised for it reads only contracts, policies, staff records, public data, websites or summary reports, with no patient identifiers. Sources for the rules: 45 CFR 164.514; NYC Local Law 144; California Civil Rights Council.

Corrected October 14, 2026. Three problems moved from needing no patient data to needing it for part of the work, because a later step for each reads patient records: automating the UDS report at health centers, checking charts before a Medicare claim in post-acute and senior care, and a results study on claims or visit records for health-tech firms.

Show Fig. 4 as a table
The 44 problems by whether an AI tool for them needs patient data.
Kind of organizationNeeds no patient dataNeeds patient data for part of the workNeeds patient data from the startNo AI tool advised
Health systems6200
Community hospitals6100
Health centers5200
Practices4021
Post-acute and senior care4310
Health-tech and associations5200
All six301031

What an AI tool does for the 30 problems that need no patient data

For most of these 30 problems, an AI tool does one of the six tasks below. In each, the tool reads only public information or files with no patient identifiers in them, such as names, record numbers or admission dates. AI tools can get things wrong and still sound sure of themselves, so each task also says how its accuracy is checked before anyone relies on it. In every one, a person makes the final call.

  1. Pull deadlines and terms out of documents

    Reads payer and vendor contracts, policies and funding notices, and lists each deadline and term with the page it came from.

    How its accuracy is checked. A person on your team checks every deadline and term against the page it came from before it goes into your contract register, and the tool never marks a term verified on its own. You can test it on a small set of your own contracts before you buy more.

    On the menu: Payer and Vendor Contract Register

  2. Check websites and public files against the rules

    Checks a hospital's price files, website accessibility and online listings against the rules that apply to them, on a schedule.

    How its accuracy is checked. It runs CMS's free validator, which checks the price file's format, and every week it compares a sample of posted prices with your chargemaster (the hospital's master price list) and payer rates, flagging any that differ. Automated tests catch only some accessibility failures, so a person still has to test the site, for example with a screen reader, before you say it meets the standard. Source: W3C, Selecting Web Accessibility Evaluation Tools.

    On the menu: Price Transparency and Accessibility Monitor

  3. Compare your documents with a checklist

    Compares your documents with the items HRSA reviews on a health center site visit, or the rules nursing home inspectors cite, and lists what's missing.

    How its accuracy is checked. The health center tool labels each match between a requirement and one of your documents “AI proposal” and quotes the line it relied on, and nothing counts as evidence until your compliance lead accepts it. The nursing home tool labels each checklist it drafts “AI draft” for your director of nursing to edit. Neither tool ever marks a requirement met.

    On the menu: Health Center Site Visit Readiness and Nursing Home Survey Readiness

  4. Turn exported numbers into a weekly brief

    Turns counts and totals exported from the systems you already run into one page for leaders each week.

    How its accuracy is checked. Ordinary code, not the AI, calculates every number from your reports, and the AI can't change them. The person you name to approve the brief, such as the owner, CFO or chief of staff, can check any number in one click, and the AI's own observations are labeled as AI.

    On the menu: Weekly Leadership Brief

  5. Watch public sources and flag what applies

    Watches grant notices, rural health funds and CMS nursing home inspection data, and flags what applies to you.

    How its accuracy is checked. Each notice gets a score for how well it fits your organization, and the score shows the eligibility rules and facts behind it. The AI's comments are labeled as AI and name the lines they relied on, and your staff decide what to act on.

    On the menu: Funding and Needs Assessment Desk and Nursing Home Survey Readiness

  6. Draft writing for a person to approve

    Drafts replies to online reviews, sections of grant applications and social media posts.

    How its accuracy is checked. Before drafting any part of a grant application, the tool checks whether that funder allows AI-assisted writing. A person reads every draft, and nothing is posted or sent until they approve it.

    On the menu: Website, Reviews and Social Desk and Funding and Needs Assessment Desk

Part five

Explore all 44 problems

Search them, or pick a kind of organization or an area. Tap any square for the problem's figure and source, who feels it, and the tools that address it, with prices from the services menu.

The 44 problems, by kind of organization and area

Each square is one problem, colored by the patient data an AI tool for it would need. Select a square to see its figure and source, who feels it, and the tools that address it.

  • Needs no patient data 30
  • Needs patient data for part of the work 10
  • Needs patient data from the start 3
  • No AI tool advised 1

Tap a row or column name to show only that kind or area.

Group the list by

All 44 problems, grouped by kind of organization.

Health systems 8

Community hospitals 7

Health centers 7

Practices 7

Post-acute and senior care 8

Health-tech and associations 7

Each figure was checked at its source in October 2026. Free to reuse with credit to Kent Applied.

We picked the problems from the sources listed with each one (such as federal rules and data files, congressional and research reports, trade group surveys, industry reports and journals) and checked each number at its source in October 2026. The groupings by area and by patient data are our own. The tools are Kent Applied's, at the starting prices on the services menu. The data is free to reuse with credit: download it as JSON.

AI tools

See one of your organization's problems here?

Tell us which one. We'll say which tool fits, or whether a smaller first step makes more sense.

Or see every service and its price. Prefer to skip the call? Get a written proposal.