Maralee Lellio fought for more time with Andrew, Ayla and Rosa. Her story raises an uncomfortable question for physicians and healthcare leaders: Are we using AI to give human attention back to healthcare—or simply making unnecessary work faster?
“Patient safety, physician expertise and the humanity of clinical practice must guide how these tools are developed and used.” — American Medical Association, American Academy of Family Physicians, American Academy of Pediatrics, American College of Obstetricians and Gynecologists, American College of Physicians, and American College of Surgeons, September 30, 2026
Healthcare doesn't need more automation. It needs fewer
problems to automate.
Maralee Lellio Wanted to Live.
Not optimize a workflow.
Not improve a dashboard.
Not reduce clicks.
Not increase productivity.
Live.
That distinction matters more than it first appears.
In 2018, Maralee Lellio was 28 years old and living
in northeastern Ohio.
She was a teacher.
She was a mother.
She was married to Andrew Lellio.
She had a young daughter, Ayla Lellio.
And she imagined the future the way most young parents do.
She would keep teaching.
Ayla would grow up.
Maybe they would have another child.
There would be birthdays, school mornings, family dinners,
arguments over bedtime and all the wonderfully ordinary chaos that makes a
life.
Then Maralee found a lump.
She did what patients are told to do.
She sought medical attention.
She was reassured that it was likely benign.
It wasn't.
The lump grew.
It became painful.
In January 2019, at age 29, Maralee was diagnosed with Stage
2B triple-negative breast cancer.
She underwent chemotherapy and a bilateral mastectomy.
For a while, things seemed to be moving in the right
direction.
Then came the headaches.
Then dizziness.
Then worsening symptoms.
A brain scan initially did not reveal cancer.
Eventually, an MRI showed a large brain tumor.
Her breast cancer had progressed to Stage IV disease.
The prognosis was devastating.
At one point, Maralee accepted that she might die within a
few years.
She thought about Ayla.
She thought about the second child she and Andrew had
dreamed about.
She thought about all the ordinary things she might never
get to experience.
Then Andrew said something that changed her trajectory.
He told her he believed she could survive.
He reminded her that some people live far beyond the
prognosis they are given.
Maralee later described the moment as a light switch
turning on.
She decided:
“I'm going to live.”
Not because anyone promised her a cure.
Not because an algorithm predicted a favorable outcome.
Not because her medical record produced a green checkmark.
She decided that the future was still worth pursuing.
Her care eventually moved to Cleveland Clinic.
There, Maralee received radiation and later a PARP
inhibitor, a targeted therapy particularly relevant to her BRCA1-positive
cancer.
Her treatment worked.
She eventually had no evidence of active disease.
Then came the part of the story that sounds almost
impossible.
Maralee wanted another child.
Her oncologist, Halle Moore, MD, supported her
decision after carefully weighing the risks.
In July 2024, Maralee and Andrew welcomed their second
daughter, Rosa Lellio.
Ayla became a big sister.
Maralee became a mother of two.
And she returned to teaching.
Cleveland Clinic describes Maralee today as the mother of
two daughters, Ayla and Rosa, and quotes her describing life with them
as “messy and perfect.” [1]
That phrase stopped me.
Messy and perfect.
Because perhaps we have spent too much time trying to
eliminate the wrong kind of mess.
There is the mess of life.
And there is the mess of healthcare.
They are not the same thing.
A child waking up at 2 a.m. is messy.
A toddler spilling milk is messy.
A patient asking one more question is messy.
A physician spending an extra five minutes listening is
messy.
A family changing plans because life happened is messy.
That is human.
That is life.
Healthcare should make room for that kind of mess.
Instead, we have built an enormous amount of administrative
mess that nobody asked for.
Faxing.
Re-keying.
Rechecking.
Prior authorization.
Denial appeals.
Duplicate documentation.
Missing information.
Portal messages.
Payer portals.
Coding corrections.
Eligibility problems.
Authorization problems.
Claim edits.
Payment reconciliation.
Manual follow-up.
And then we proudly announce:
“Don't worry. We have AI.”
Maybe.
But before we celebrate the robot, perhaps we should ask why
the robot has so much work to do.
The Healthcare Industry Has a Strange Definition of
Innovation
Here is my contrarian proposition:
Healthcare does not have an automation problem nearly as
much as it has a problem deciding what deserves human attention.
We keep creating complicated processes.
The processes create exceptions.
People manage the exceptions.
The exceptions create more work.
Organizations hire more people.
Technology companies build software to process the work.
Then someone adds AI.
And suddenly we call the whole thing transformation.
Sometimes it is.
Sometimes it is simply a faster way to do something we
should have stopped doing.
That distinction is becoming increasingly important.
Because healthcare AI is entering almost every layer of the
system.
AI documentation.
AI coding.
AI scheduling.
AI prior authorization.
AI claims.
AI denials.
AI payment posting.
AI patient communication.
AI revenue-cycle management.
AI clinical decision support.
AI everything.
The question is no longer whether AI will enter healthcare.
It already has.
The better question is:
Where should intelligence be applied?
And even more importantly:
What work should disappear before we automate it?
The Patient Is Not the Workflow
Maralee's story has almost nothing to do with medical
billing.
And that is exactly why I want to start there.
I am not suggesting that medical billing caused her cancer.
It didn't.
I am not suggesting that an AI billing platform could have
changed her cancer outcome.
It couldn't.
That would turn a deeply human story into a marketing
gimmick.
The real lesson is much broader.
Human attention is finite.
Andrew Lellio gave Maralee attention.
He watched her.
He drove her to Cleveland Clinic.
He worked from home so he could monitor her.
He stayed beside her.
He helped her when she was too weak to get up.
He made sure she ate.
He was there.
The Cleveland Clinic team gave her attention.
Dr. Halle Moore gave her attention.
Maralee gave attention to her own future.
That attention mattered.
Now think about a typical physician practice.
A doctor opens the EHR.
There are messages.
Tasks.
Alerts.
Refills.
Forms.
Authorizations.
Denials.
Unsigned documents.
Coding questions.
Results.
Patient messages.
Administrative requests.
The patient needs attention.
But the system needs attention too.
And the system never stops asking.
That is the hidden cost of healthcare complexity.
The system competes with the patient for human attention.
And we have become remarkably good at letting the system
win.
The Most Valuable Resource in Healthcare Isn't AI
It isn't data.
It isn't software.
It isn't even money.
It is attention.
Physicians have limited attention.
Nurses have limited attention.
Medical assistants have limited attention.
Billers have limited attention.
Practice managers have limited attention.
Patients have limited attention.
Caregivers have limited attention.
Everyone has a finite cognitive budget.
Yet our healthcare infrastructure behaves as though human
attention were unlimited.
It isn't.
And eventually the bill comes due.
The American Medical Association's recent physician research
continues to show the enormous burden of administrative work, particularly
prior authorization.
Physicians report delays in care, negative effects on
outcomes, patient abandonment of treatment, serious adverse events and
substantial amounts of physician and staff time devoted to authorization
processes.
That is not simply a workflow inconvenience.
It is an attention problem.
And attention problems eventually become patient-care
problems.
Thirteen Hours Is Not a Workflow
Consider the often-cited prior-authorization burden.
Physicians and staff can spend approximately 13 hours per
week dealing with prior authorization.
Thirteen hours.
That is more than one full workday.
Every week.
Imagine telling a physician:
“We're going to take one day out of your week, every week,
and give it to paperwork.”
We would call that unacceptable if we described it honestly.
Instead we call it:
Administrative burden.
What a wonderfully polite phrase.
“Administrative burden” sounds like carrying a heavy
backpack.
It is closer to asking a physician to run a call center
between patients.
And then wondering why they are exhausted.
We Celebrate the Wrong Heroes
Healthcare has a strange cultural relationship with heroism.
We celebrate the biller who gets the impossible claim paid.
The nurse who stays late.
The physician who answers messages at 10 p.m.
The practice manager who knows exactly which payer portal to
use.
The coder who finds the missing documentation.
The employee who remembers the workaround nobody wrote down.
These people are heroes.
But there is a dangerous side effect.
Organizations can become dependent on heroes.
A process breaks.
Someone fixes it.
The organization survives.
Everyone goes home.
Until it breaks again.
Then the same person fixes it.
Again.
And again.
Eventually the organization says:
“Our team is really good at handling exceptions.”
Maybe.
Or maybe your system is really good at producing exceptions.
There is a difference.
The Hero Trap
The hero trap looks like this:
- Information
enters the system.
- Something
is missing.
- A
human notices.
- A
human investigates.
- A
human calls someone.
- A
human opens another system.
- A
human finds the information.
- A
human corrects the record.
- A
human resubmits the transaction.
- The
organization celebrates the recovery.
- Nobody
redesigns the original process.
Then it happens again tomorrow.
That isn't resilience.
That's institutionalized rework.
And here's where AI can become dangerous.
Not because AI is dangerous by itself.
Because AI can become an extremely efficient employee inside
a fundamentally inefficient architecture.
Imagine hiring the smartest assistant in the world and
giving that assistant 50,000 badly completed forms.
The assistant can process them faster.
Congratulations.
You now have the world's fastest bad-form processor.
But why are there 50,000 bad forms?
That's the question.
The AI Arms Race Has a Blind Spot
The healthcare industry is racing toward AI.
Good.
We should.
AI can be transformative.
But there is a blind spot in the conversation.
We frequently assume that intelligence automatically creates
simplicity.
It doesn't.
Intelligence can process complexity. It does not
necessarily eliminate complexity.
That distinction should be printed on every healthcare AI
pitch deck.
If documentation is inconsistent, AI can interpret it.
If claims are denied, AI can predict them.
If coding is manual, AI can accelerate coding.
If prior authorization is cumbersome, AI can automate parts
of it.
All useful.
But none of those necessarily changes the underlying
architecture.
The system still depends on the mess.
It simply becomes better at dealing with it.
That is not the same as removing the mess.
What If We Stopped Asking AI to Clean Up After Us?
Here's the question I would like every healthcare technology
company to answer:
What if the most valuable AI doesn't clean up the mess
faster?
What if it prevents the mess?
That changes everything.
Instead of:
Predict the denial.
Ask:
Why was the claim likely to be denied?
Instead of:
Automate the appeal.
Ask:
Could the original error have been prevented?
Instead of:
Make the biller faster.
Ask:
Why does the biller have to touch this transaction at
all?
Instead of:
Build another dashboard.
Ask:
Why does someone need a dashboard to discover something
the system should already know?
That is a much more uncomfortable conversation.
It is also where the real opportunity may be.
Healthcare Has Been Optimizing Around Noise
Healthcare organizations have become incredibly
sophisticated at compensating for variability.
Different physicians document differently.
Different staff members enter information differently.
Different systems represent information differently.
Different payers require different things.
Different workflows interpret the same information
differently.
Different people work around the limitations of software
differently.
Then we build more software to manage the variability.
Eventually the infrastructure becomes enormous.
And everybody asks:
“Why is healthcare so complicated?”
Maybe because we built an entire economy around compensating
for complexity.
This is where I believe the industry needs a reset.
Stop optimizing around noise. Remove the noise.
This Is the OnnX Thesis
This is the philosophy behind OnnX.
Healthcare billing is often treated as a workflow problem.
I believe much of the problem begins earlier.
It is a data-structure problem.
The revenue cycle receives the consequences of decisions
made upstream.
Clinical documentation.
Patient information.
Coverage.
Eligibility.
Authorization.
Coding context.
Encounter details.
Operational information.
When those elements are incomplete, inconsistent or poorly
structured, downstream teams inherit the problem.
The biller becomes the detective.
The coder becomes the interpreter.
The administrator becomes the coordinator.
The practice manager becomes the escalation department.
And the physician often becomes the person asked to fix
documentation after the fact.
That's backwards.
The person closest to the patient should not be forced to
become the cleanup crew for the revenue cycle.
Clean Claims Shouldn't Be Heroic
Think about the phrase:
Clean claim.
We celebrate it.
We measure it.
We optimize it.
But why is a clean claim considered an achievement?
Shouldn't it simply be the natural result of good
information?
If the right information was captured...
If it was structured correctly...
If it was available when needed...
If authorization requirements were known...
If eligibility was accurate...
If the workflow did not require redundant entry...
Then the claim should have a much easier journey.
The real opportunity isn't merely:
Better denial management.
It is:
Fewer preventable denials.
Those are two very different strategies.
The Downstream Addiction
Healthcare has become addicted to downstream solutions.
The claim fails.
Fix it.
The patient can't get authorized.
Fix it.
The documentation is incomplete.
Fix it.
The payer denies it.
Appeal it.
The payment is wrong.
Reconcile it.
The data doesn't match.
Correct it.
This creates an endless loop:
Problem → Human intervention → Temporary fix → Repeat.
We call it workflow.
I call it organizational déjà vu.
Same problem.
Different Tuesday.
The Most Important Question in Revenue Cycle Management
Forget for a moment about collections.
Forget about A/R.
Forget about denial rates.
Ask:
Where did the problem first become inevitable?
That question is radically different.
A denial may happen today.
But the conditions that produced the denial may have existed
days or weeks earlier.
The authorization may have been incomplete.
The documentation may have been ambiguous.
The information may have been entered incorrectly.
The system may have lacked the necessary context.
The wrong person may have had to interpret the information.
By the time the denial reaches the billing department, the
original cause may be ancient history.
Yet the biller gets the blame.
That's like blaming the fire department for the fire.
The New KPI: Preventable Work
Healthcare has plenty of metrics.
We need another category.
Preventable work.
How much of the organization's labor exists because
something earlier went wrong?
Measure:
Rework.
How many transactions require correction?
Manual touches.
How many people must intervene?
Duplicate entry.
How many times is the same information typed again?
Clarification.
How often must employees contact each other because
information is incomplete?
Exception handling.
How often does the transaction leave the normal process?
Preventable denial.
How many denials could reasonably have been prevented
earlier?
Administrative time.
How many staff hours are consumed by correcting predictable
problems?
Now we are measuring something meaningful.
The 30-Day Experiment
A clinic does not need to buy another platform tomorrow.
Try this first.
For 30 days, ask your staff to record every recurring
administrative problem.
Not the dramatic ones.
The boring ones.
Especially the boring ones.
Because the boring problems are often the expensive ones.
Track five things.
1. Rework
How many times did someone correct information?
2. Re-entry
How many times did someone type the same information into
another system?
3. Clarification
How often did one person need to contact another because
information was incomplete?
4. Exceptions
How many transactions required a workaround?
5. Preventable downstream events
How many denials, delays or corrections could have been
avoided upstream?
At the end of the month, don't immediately buy software.
Look for the pattern.
The pattern is your product specification.
Follow the Information, Not the Org Chart
This is one of the simplest exercises a clinic can perform.
Pick one transaction.
A new patient.
A procedure.
A referral.
A prior authorization.
A claim.
Now follow the information from beginning to end.
Don't follow the organizational chart.
Follow the data.
Where does it originate?
Who touches it?
Who changes it?
Who interprets it?
Who copies it?
Who approves it?
Who enters it again?
Who waits for it?
Who fixes it?
And where does it first become ambiguous?
That's where the real problem lives.
Handoffs Are Where Information Goes to Get Complicated
Consider a typical healthcare transaction.
Physician.
Medical assistant.
Front desk.
Authorization team.
Payer.
Billing.
Clearinghouse.
Payer again.
Payment.
Appeal.
That's a lot of hands.
Every handoff is an opportunity for information to degrade.
A field can disappear.
A date can change.
A diagnosis can be interpreted differently.
A requirement can be misunderstood.
A document can sit in someone's inbox.
And then somebody says:
“Why didn't billing catch this?”
Why did billing have to catch it?
That's the more interesting question.
AI Should Not Make Humans Faster at Doing the Wrong Thing
This is perhaps my biggest concern with the current AI
enthusiasm.
We are measuring automation.
But we should measure human work eliminated.
Those are not the same thing.
Suppose an AI system reduces claim-review time from ten
minutes to two.
Great.
But why is somebody reviewing the claim?
Suppose AI cuts authorization processing from 30 minutes to
five.
Great.
But why did the authorization require so much manual
interpretation?
Suppose AI cuts denial appeals from 20 minutes to three.
Excellent.
But why did the claim become a denial?
The first question celebrates efficiency.
The second question challenges architecture.
Healthcare needs both.
But we have been overinvesting in the first.
Simplify First. Automate Second.
This should be a law of healthcare technology.
Simplify first. Automate second.
Otherwise, you risk automating a bad process.
Then everybody becomes dependent on the bad process.
Then the organization spends millions maintaining the bad
process.
Then nobody can change it because:
“That's how the system works.”
This is how technical debt becomes operational culture.
What Should AI Actually Do?
AI can:
- detect
patterns,
- identify
anomalies,
- extract
information,
- classify
documentation,
- predict
risk,
- identify
missing information,
- surface
inconsistencies,
- assist
with decisions,
- automate
repetitive work,
- and
help humans focus on higher-value tasks.
But I would give AI a more demanding mission.
Not:
Make humans faster.
Instead:
Make unnecessary human work disappear.
That's a higher standard.
It forces us to distinguish between:
work that requires judgment
and
work that exists because the system is poorly designed.
A physician's judgment is valuable.
Typing the same patient information into three systems is
not.
A biller's expertise is valuable.
Hunting through five portals for information that should
already be available is not.
A nurse's clinical assessment is valuable.
Calling three departments to find a missing authorization is
not.
Human work should be expensive.
We should spend it accordingly.
What the Physician Organizations Just Told the AI
Industry
This week's joint statement from six major physician
organizations is important for precisely this reason.
The AMA, AAFP, AAP, ACOG, ACP and ACS pushed back against
the idea that AI should be treated as inherently superior to physicians.
They emphasized context.
Experience.
Professional judgment.
Patient relationships.
And humanity.
Their message was not anti-AI.
Quite the opposite.
They explicitly acknowledged AI's tremendous potential.
But they made the hierarchy clear:
Technology should enhance medicine.
It should not diminish the physician.
And it should not undermine the patient's trust.
That is an important distinction for healthcare founders.
The goal isn't to make the human less important.
The goal is to make unnecessary human work less important.
Maralee's Story Makes That Principle Real
Think about the people in Maralee's story.
Maralee Lellio.
Andrew Lellio.
Ayla Lellio.
Rosa Lellio.
Halle Moore, MD.
These aren't workflow objects.
They aren't records.
They aren't encounters.
They aren't claims.
They are people.
Maralee wanted to live.
Andrew wanted his wife.
Ayla needed her mother.
Rosa entered a world where her mother was still there.
Dr. Moore treated a person whose life extended far beyond
the clinical encounter.
That is what healthcare is supposed to protect.
The administrative infrastructure matters.
But infrastructure is not the mission.
The patient is the mission.
A Better Definition of Healthcare Efficiency
We have traditionally defined efficiency as:
More output with fewer resources.
That definition is inadequate for healthcare.
A physician seeing 30 patients instead of 20 isn't
automatically more efficient if every patient gets less attention.
A billing department processing 10,000 claims instead of
7,000 isn't automatically more efficient if denials increase.
A call center answering more calls isn't automatically more
efficient if patients have to call three times.
A portal that receives more messages isn't necessarily a
success if patients use it because they can't reach anyone.
Healthcare needs a different definition.
More human value with less unnecessary work.
That is the metric I want.
The Human Attention Return
Imagine every healthcare technology company had to report
one new metric:
Human Attention Returned
How many hours did the product return to:
Physicians?
Nurses?
Medical assistants?
Billers?
Practice managers?
Patients?
Caregivers?
Now the conversation changes.
A billing system that saves 1,000 hours matters.
But what happens to those hours?
If they simply create 1,000 more administrative tasks,
nothing meaningful changed.
If those hours become patient-facing time?
Now we have something.
If physicians go home earlier?
Something changed.
If staff stop working Saturday mornings?
Something changed.
If patients get callbacks sooner?
Something changed.
If clinicians have more time to think?
Something changed.
The Real ROI of Healthcare Technology
We talk about ROI as dollars.
We should.
But healthcare technology has another return.
Return on attention.
What does the organization get back?
Time.
Focus.
Judgment.
Presence.
Patience.
Human connection.
These are harder to put into a spreadsheet.
They may be more important than what fits in the
spreadsheet.
Three Questions for Every AI Vendor
Before buying another AI platform, ask:
What work disappears?
Not what the software automates.
What work actually disappears?
What work does your product create?
Every system creates some new work.
Be honest about it.
What happens when the AI is wrong?
This is perhaps the most important question.
In healthcare, failure is not merely a technical issue.
It can become a clinical, financial, legal or ethical issue.
Five More Questions I Would Ask
Where does your system intervene?
Upstream?
Midstream?
Downstream?
How does it improve data quality?
Or does it simply interpret poor data later?
How many human touches remain?
Don't tell me the system is “90% automated.”
Show me the remaining 10%.
That's where the interesting problems usually live.
Can we audit what happened?
If the answer is no, keep asking.
Does the technology prevent exceptions?
Or does it process exceptions faster?
Those are very different products.
The OnnX Perspective
The philosophy behind OnnX is intentionally
different.
Medical billing has become an enormous downstream ecosystem.
Billing companies.
Clearinghouses.
Payer portals.
Coding systems.
Denial systems.
Authorization systems.
Analytics.
Work queues.
Dashboards.
Appeals.
And now AI agents.
The industry has become very good at moving information
through an increasingly complicated maze.
But what if the answer isn't another better maze?
What if the answer is less maze?
OnnX is being built around the premise that much of the
revenue-cycle problem begins upstream.
If the information is structured correctly at the point of
capture, fewer downstream systems have to guess.
Fewer people have to interpret.
Fewer transactions require correction.
Fewer exceptions need escalation.
The revenue cycle becomes more deterministic.
Less reactive.
Less dependent on heroics.
That is the goal.
Don't just automate the mess. Prevent the mess.
This Isn't About Eliminating Billers
This distinction matters.
Prevention-oriented technology should not be confused with
simply eliminating jobs.
Healthcare needs knowledgeable people.
Billing expertise matters.
Coding expertise matters.
Practice-management expertise matters.
The question is whether highly skilled people should spend
their day doing highly skilled work.
Or whether they should spend it correcting predictable
system failures.
Those are not equivalent.
A great biller should be solving difficult revenue-cycle
problems.
Not searching three portals because an address was entered
incorrectly.
A great practice manager should be improving the practice.
Not becoming the organization's human error-correction
engine.
The Jobs That AI Should Replace
Let's be provocative.
AI should replace some work.
Absolutely.
But let's be specific.
AI should replace:
Copying.
Re-keying.
Sorting.
Searching.
Matching.
Flagging.
Routing.
Repetitive verification.
Predictable reconciliation.
Administrative detective work.
The objective isn't:
Replace the human.
The objective is:
Replace the unnecessary task.
That's a much better AI strategy.
The Future May Belong to Boring AI
Here's another prediction.
Some of the most valuable healthcare AI in the next decade
will be incredibly boring.
No humanoid robot.
No dramatic demo.
No giant screen.
No science-fiction interface.
Maybe:
The claim never needed correction.
Maybe:
The authorization was automatically complete.
Maybe:
The physician never had to re-document the same
information.
Maybe:
The biller never saw the transaction because nothing went
wrong.
Maybe:
The patient never knew the software existed.
That's boring.
And beautiful.
Because the best infrastructure is often invisible.
The Difference Between Automation and Prevention
Automation asks:
Can a machine do this task?
Prevention asks:
Why does this task exist?
Automation asks:
Can AI process this exception?
Prevention asks:
Why did the exception happen?
Automation asks:
Can we make the denial workflow faster?
Prevention asks:
Can we make the denial less likely?
Automation asks:
Can we reduce clicks?
Prevention asks:
Why are there so many clicks?
That is the difference.
The 30-Day Challenge for Physicians and Clinic Owners
Try this.
For the next 30 days, don't ask your staff:
“How productive were you?”
Ask:
“What unnecessary work did the system create for you
today?”
Write down every answer.
At the end of the month, rank the problems.
Then choose one.
Don't automate it.
Prevent it.
If you can't prevent it, simplify it.
If you can't simplify it, automate it.
If you can't automate it, at least make the responsibility
clear.
That sequence matters.
What Should Clinics Measure?
A modern practice should consider tracking:
First-pass claim rate
How many claims succeed without intervention?
Manual touch rate
How many human interventions occur per transaction?
Rework rate
How many transactions require correction?
Duplicate-entry rate
How often is information entered more than once?
Exception rate
How often does a transaction leave the standard workflow?
Preventable denial rate
How many denials were reasonably foreseeable?
Authorization cycle time
How long does the process take?
Administrative hours per provider
How much physician and staff time is being consumed?
Patient-facing hours
How much time actually reaches the patient?
And one more:
Human attention returned.
That is the metric I would put on the executive dashboard.
Legal Implications
The more healthcare relies on AI, the more important
accountability becomes.
Healthcare organizations should consider:
- HIPAA
and privacy obligations.
- Data
security.
- Business
associate requirements.
- Access
controls.
- Auditability.
- Documentation.
- Human
oversight.
- Payer-contract
requirements.
- Fraud
and abuse concerns.
- False
Claims Act exposure.
- State-specific
requirements.
- Clinical
liability.
- Vendor
accountability.
- Model
governance.
The legal question isn't simply:
“Does the AI work?”
It is also:
“Can we explain what happened?”
Who made the decision?
What information was available?
What did the system recommend?
Who approved it?
What happened when the system was wrong?
Healthcare AI requires accountability architecture.
Not just model architecture.
Ethical Implications
The ethical question is even simpler.
What happens to the human when we optimize the system?
Does the physician gain time?
Does the patient gain access?
Does the nurse gain attention?
Does the biller gain meaningful work?
Or does the organization simply increase throughput?
Because efficiency can be morally neutral.
Efficiency can also be used to scale the wrong thing.
A faster bad process is still a bad process.
The Privacy Paradox
Healthcare is entering an era in which more systems want
more data.
AI wants data.
Analytics wants data.
Payers want data.
Platforms want data.
Vendors want data.
But more data does not automatically mean better healthcare.
Sometimes it means more complexity.
The future should not simply be:
More data.
It should be:
Better structured information.
Information that is available at the right moment.
In the right context.
To the right person.
For the right decision.
That is harder.
And much more valuable.
The Patient Doesn't Care Which Department Owns the
Problem
This is another truth healthcare sometimes forgets.
A patient doesn't care whether the problem belongs to:
Billing.
Scheduling.
Authorization.
Coding.
Eligibility.
IT.
The payer.
The clearinghouse.
The physician.
The hospital.
The patient experiences the problem.
Imagine a patient saying:
“I completely understand that the authorization issue
belongs to a different department.”
Nobody says that.
They say:
“Can somebody please fix this?”
The patient doesn't see our organizational chart.
Why should our technology?
The Most Important Handoff Is the One You Eliminate
Healthcare spends enormous energy improving handoffs.
Better handoff tools.
Better workflows.
Better communication.
Better alerts.
Better coordination.
Good.
But sometimes the best handoff is:
No handoff.
If the information can flow automatically, don't make a
person carry it.
If the system already knows, don't make someone ask.
If the data already exists, don't make someone re-enter it.
If a decision can be made safely upstream, don't create a
downstream work queue.
The best workflow may be the workflow you remove.
Why This Matters to Physicians
Physicians entered medicine to care for patients.
That doesn't mean they hate technology.
It means they want technology to respect the reason they
entered medicine.
The recent joint physician-organization statement on AI made
this point clearly.
AI should augment physicians.
It should not replace professional judgment.
It should not undermine trust.
It should support the humanity of clinical practice.
That means healthcare AI companies have a responsibility.
Not just to build smarter systems.
But to build systems that understand where human judgment
belongs.
And where it doesn't.
Why This Matters to Clinic Owners
Clinic owners have another problem.
Margins.
Staffing.
Payer pressure.
Administrative costs.
Recruitment.
Retention.
Denials.
A/R.
Compliance.
Technology costs.
They don't need another software vendor promising to
“revolutionize the workflow.”
They need fewer fires.
And fewer reasons to hire someone to put out those fires.
The best technology investment may therefore be the one that
quietly eliminates the recurring problem everyone has accepted as normal.
Why This Matters to Healthcare Founders
For founders, the temptation is obvious.
Find an inefficient workflow.
Add AI.
Sell efficiency.
But there is a better opportunity.
Find the source of the work.
Understand why it exists.
Find the first point where information becomes ambiguous.
Move intelligence upstream.
Prevent the downstream event.
That is harder.
It may also produce a much more defensible company.
Because you're not just replacing labor.
You're changing architecture.
Healthcare's Next Competitive Advantage
The next healthcare advantage may not be:
More automation.
It may be:
Less unnecessary work.
The organizations that figure this out will have advantages
in:
- physician
retention,
- staff
retention,
- operating
margins,
- patient
experience,
- revenue-cycle
performance,
- administrative
efficiency,
- and
potentially clinical capacity.
Not because they bought more AI.
Because they created less noise.
Three Things I Would Stop Doing Tomorrow
Stop Celebrating Rework
Rework is not productivity.
It is evidence.
Every correction tells you something about the system that
produced the error.
Stop Calling Every Problem a Workflow Problem
Sometimes the workflow is merely the messenger.
The actual problem is upstream information.
Stop Buying Automation Before Understanding the Work
Otherwise you may build a faster machine for producing the
same problems.
And then you will spend five years explaining why the
machine is “transformative.”
Three Things I Would Start Doing Tomorrow
Start Measuring Preventable Work
If you don't measure it, the organization will continue
producing it.
Start Designing Information Before Designing Workflow
Bad information creates expensive workflows.
Start Treating Human Attention as a Strategic Asset
Because it is.
A Question for Every Healthcare CEO
If you could give every physician in your organization five
additional hours per week, what would you want them to do?
More documentation?
Probably not.
More meetings?
Please, no.
More administrative tasks?
Obviously not.
More patient care?
More thinking?
More teaching?
More listening?
More time with families?
Now ask:
Why aren't we designing systems around that outcome?
A Question for Every Practice Manager
Ask your staff:
“What is the stupidest thing you have to do repeatedly?”
Don't laugh.
Don't defend it.
Don't explain why it exists.
Write it down.
Then ask:
“Why does the system require this?”
That conversation could be worth more than your next
software demo.
A Question for Every AI Founder
Before you tell me how many tasks your system can automate,
tell me:
How many tasks no longer need to exist because of your
system?
That is the question I want to hear more often.
Maralee's Story Comes Back Into Focus
Go back to Maralee.
She wasn't fighting for another workflow.
She was fighting for Ayla.
For Andrew.
For Rosa, who had not yet been born.
For teaching.
For birthdays.
For ordinary mornings.
For the little stresses she now describes as part of a life
she is grateful to have.
Cleveland Clinic's account makes something especially clear.
Maralee understood the seriousness of her diagnosis.
She didn't need false reassurance.
She needed facts.
She needed trustworthy physicians.
She needed autonomy.
She needed people willing to see her as more than a
diagnosis.
And she needed time.
That's what healthcare should protect.
Time.
Time to think.
Time to listen.
Time to decide.
Time to explain.
Time to heal.
Time to live.
The Patient Is Not the Workflow
We should probably put that sentence above every healthcare
technology roadmap.
The patient is not the workflow.
The patient is why the workflow exists.
The infrastructure is not the mission.
The infrastructure supports the mission.
The claim is not the mission.
The denial is not the mission.
The dashboard is not the mission.
The AI model is not the mission.
The workflow is not the mission.
The human being is the mission.
The Real Opportunity for AI
The future of healthcare AI should not be measured only by
intelligence.
It should be measured by restraint.
Can AI know when not to interrupt?
Can AI know when not to create another task?
Can AI recognize that the best outcome is sometimes no
workflow at all?
Can AI prevent a problem before it enters a queue?
Can AI make information cleaner before humans have to
interpret it?
Can AI eliminate repetitive work without eliminating
judgment?
Can AI give physicians time back?
Can AI give nurses time back?
Can AI give billers time back?
Can AI give patients time back?
If yes, now we're talking.
The Future May Be Invisible
The most impressive healthcare technology of the next decade
may not look impressive.
It may simply mean:
The physician doesn't have to re-enter the information.
The biller doesn't have to correct the claim.
The nurse doesn't have to chase the authorization.
The patient doesn't have to call again.
The practice manager doesn't have to open another
spreadsheet.
The claim doesn't become a denial.
The denial doesn't become an appeal.
The appeal doesn't become another human task.
Nothing dramatic happens.
And that is exactly the point.
The system simply works.
Don't Just Automate the Mess
This is the idea I keep coming back to.
Don't just automate the mess.
Prevent the mess.
Don't build the fastest denial machine.
Build a system that makes fewer denials necessary.
Don't make the biller work faster.
Make fewer transactions require the biller's intervention.
Don't make physicians document faster.
Make the information they create more useful downstream.
Don't give healthcare another dashboard.
Give it fewer problems that require dashboards.
Don't build AI that makes the system more complicated.
Build AI that makes the human experience simpler.
The OnnX Opportunity
That is where I believe OnnX fits.
Not as another outsourced RCM company.
Not as another layer of human intermediaries.
Not as another dashboard sitting on top of a fragmented
system.
The premise is much simpler:
Fix the information before it becomes the problem.
Use intelligence upstream.
Structure information earlier.
Reduce ambiguity.
Reduce manual handoffs.
Reduce rework.
Reduce preventable denials.
Make revenue-cycle processes more deterministic.
And ultimately:
Give human attention back to healthcare.
That is the opportunity.
The 10-Question Healthcare AI Test
Before your next AI purchase, ask:
1. What problem existed before your software?
2. Where does that problem actually begin?
3. Where does your technology intervene?
4. What human work disappears?
5. What new work does your product create?
6. How does your system improve information quality?
7. What happens when the AI is wrong?
8. Can the organization audit the decision?
9. Does the technology prevent exceptions or simply
process them faster?
10. How much human attention does the product return?
If a vendor can't answer those questions clearly, don't be
impressed by the demo.
Ask harder questions.
Three Final Challenges
For Physicians
Look at your calendar this week.
Find one hour spent doing something that had nothing to do
with caring for a patient.
Then ask:
Why did I have to do that?
Don't accept:
“Because that's how we do it.”
That's not an explanation.
It's a confession.
For Clinic Owners
Ask your team:
“What problem do we solve over and over again that should
have been prevented?”
Write down the answer.
Then find where the problem begins.
Not where someone fixes it.
Where it begins.
For Healthcare Technology Leaders
Before you tell me your AI can automate a workflow, tell me
something more interesting:
Can you make the workflow unnecessary?
That's the standard I believe healthcare technology should
increasingly demand.
Not more automation.
Less unnecessary work.
Not more dashboards.
Better information.
Not faster denial management.
Fewer preventable denials.
Not AI replacing humans.
AI returning human attention to healthcare.
So What Do You Think?
Is healthcare really suffering from an automation problem?
Or have we spent decades automating around problems that
should have been prevented upstream?
Where does your organization spend the most time fixing
something that should have been right the first time?
What is the one administrative task you would eliminate
tomorrow if you could?
I want to hear from physicians, clinic owners, billers,
coders, practice managers and healthcare technology leaders.
Tell me in the comments.
And if this perspective resonates, consider reposting it.
Someone in your network may be spending half their week fixing a problem nobody
has questioned in years.
A Free Starting Point
I have placed a free OnnX resource in Featured on
LinkedIn for physicians and clinic owners who want to examine their own
revenue-cycle and administrative friction.
No signup. No sales funnel.
The objective is simple:
Find where the work begins.
Find where information becomes noisy.
Find where humans are compensating for the system.
Then ask whether the problem can be prevented.
Frequently Asked Questions
Is Maralee Lellio cured?
The most precise description is that Maralee has had no
evidence of active disease following treatment. Her story involves
metastatic, Stage IV breast cancer, which is generally considered incurable.
She continues ongoing surveillance. [1]
That distinction matters.
Hope does not require inaccurate language.
Who is Andrew Lellio?
Andrew Lellio is Maralee's husband and the father of
Ayla and Rosa. He was a central source of emotional and practical support
throughout her illness and played an important role in changing her outlook
after her Stage IV diagnosis. [1][2]
Who is Halle Moore, MD?
Halle Moore, MD, is Director of Breast Medical
Oncology at Cleveland Clinic's Taussig Cancer Institute and Co-Director of the
Cleveland Clinic Comprehensive Breast Cancer Program. She became Maralee's
oncologist and helped guide her subsequent treatment and decisions about
pregnancy. [3]
Who are Ayla and Rosa Lellio?
Ayla Lellio is Maralee and Andrew's older daughter.
Rosa Lellio is their younger daughter, born in July
2024. [1]
Why use a cancer story in an article about medical
billing?
Because the article isn't actually about cancer or billing.
It is about human attention.
Maralee's story makes visible what healthcare can easily
forget: behind every workflow is a human life.
The administrative infrastructure should protect that life.
It should not consume the attention required to care for it.
Is AI the problem?
No.
Poorly designed systems are the problem.
AI can be extraordinarily useful.
The issue is where intelligence is applied.
If AI makes a bad process faster, the organization has not
necessarily become better.
It may simply have become more efficient at being
inefficient.
Does prevention mean eliminating billers?
No.
It means eliminating unnecessary work.
Experienced billers, coders and practice managers have
tremendous value.
Their expertise should be directed toward problems requiring
judgment—not predictable errors created by poor information flow.
What should a clinic measure first?
Start with:
Rework.
Manual touches.
Duplicate entry.
Exception rates.
Preventable denials.
Authorization delays.
Administrative hours.
Then ask:
How much human attention did we return?
Myth Busters
Myth: More automation means less work.
Not necessarily.
Automation can reduce one task while creating another.
Measure net human work.
Myth: The best billing platform has the most features.
No.
The best system may be the one that creates fewer
exceptions.
Myth: Denials belong to billing.
Some do.
Many originate earlier.
Documentation.
Eligibility.
Authorization.
Coding.
Coverage.
Data structure.
Workflow timing.
By the time the denial reaches billing, the original problem
may be weeks old.
Myth: AI eliminates judgment.
It shouldn't.
Good AI should make human judgment more valuable, not
less.
Myth: Patients care about our workflow.
They don't.
They care whether they receive care.
They care whether someone listens.
They care whether their medication arrives.
They care whether the doctor has time.
The workflow is our problem.
The outcome is theirs.
Practical Framework
Use this sequence when evaluating any recurring
administrative problem:
1. Identify it.
What keeps happening?
2. Trace it.
Where does the information originate?
3. Locate the first failure.
When does ambiguity begin?
4. Prevent it.
Can the problem be eliminated upstream?
5. Simplify it.
If it can't be eliminated, can the workflow be reduced?
6. Automate it.
Only now should AI enter the conversation.
7. Measure it.
Did human work actually disappear?
8. Learn from it.
What does the remaining exception teach you?
This is the opposite of automation-first thinking.
It is prevention-first architecture.
Future Outlook
The next decade will produce an extraordinary number of
healthcare AI products.
Some will be revolutionary.
Some will be useful.
Some will disappear.
Some will become expensive middleware.
And some will simply put a beautiful interface on an old
problem.
The winners may not necessarily be the companies with the
largest models.
They may be the companies that understand where
intelligence belongs.
The real opportunity is upstream.
Before the denial.
Before the appeal.
Before the authorization problem.
Before the duplicate entry.
Before the manual handoff.
Before the patient becomes another exception in a queue.
The healthcare system of the future should not simply
process problems more intelligently.
It should create fewer problems to process.
Final Thought
Maralee Lellio did not fight for her life so she could
become a better healthcare workflow.
She fought because she wanted to see Ayla grow up.
She wanted Andrew beside her.
She wanted to meet Rosa.
She wanted to keep teaching.
She wanted ordinary mornings.
Ordinary family dinners.
Ordinary arguments.
Ordinary stress.
Ordinary happiness.
The messy, imperfect life she once feared she would lose.
That's what healthcare exists to protect.
Not the claim.
Not the code.
Not the dashboard.
Not the workflow.
Not the AI model.
The human being.
So perhaps the most important question for healthcare
technology isn't:
“What can AI automate?”
Perhaps it is:
“What human attention can we give back?”
Because the next great healthcare innovation may not be the
system that does more.
It may be the system that finally understands what doesn't
need to be done at all.
Don't just automate the mess.
Prevent the mess.
And give human attention back to healthcare.
References
[1] Cleveland Clinic — “After Breast Cancer Spread to Her
Brain, Young Mother Finds Path to Second Child”
Cleveland Clinic — Maralee Lellio Patient Story
[2] PEOPLE — “At 29, She Treated Her Breast Cancer — then
Discovered it Spread to Her Brain. Now She's Back to Teaching, Spending Time
with Her Daughters”
By Wendy Grossman Kantor, October 7, 2026.
PEOPLE — Maralee Lellio Story
[3] Cleveland Clinic — Halle Moore, MD
Director, Breast Medical Oncology; Co-Director, Cleveland Clinic Comprehensive
Breast Cancer Program.
Cleveland Clinic — Halle Moore, MD
[4] American Medical Association — “Statement from
leading physician organizations on the role of augmented intelligence in
healthcare”
September 30, 2026. Joint statement from the AMA, AAFP, AAP, ACOG, ACP and ACS.
AMA — Joint AI Statement
About the Author
Dr. Daniel Cham is a physician and medical consultant
with expertise in medical technology consulting, healthcare management and
medical billing. He focuses on practical insights that help healthcare
professionals navigate complex challenges at the intersection of healthcare,
technology and medical practice.
Connect with Dr. Cham on LinkedIn to
learn more.
Continue the Conversation
Explore more perspectives on healthcare, medical technology,
medical billing and the future of healthcare operations through Dr. Cham's
website, podcast, YouTube channel and social platforms.
Listen to the
podcast on Spotify
Subscribe and
watch on YouTube
Knowledge creates better questions. Better questions
create better healthcare. Start there.
Disclaimer
This article is provided for general informational and
educational purposes only. It does not constitute medical, legal, financial,
compliance or professional advice. Individual patients, physicians and
healthcare organizations have different circumstances. Clinical decisions
should be made by qualified healthcare professionals, and legal or regulatory
questions should be reviewed with appropriately qualified counsel.
Hashtags
#Healthcare #HealthcareAI #MedicalBilling
#RevenueCycleManagement #HealthcareInnovation #MedicalTechnology
#PhysicianBurnout #AdministrativeBurden #PracticeManagement #PatientCare
#HealthcareOperations #DenialPrevention #PriorAuthorization #HealthTech #OnnX
