Healthcare doesn't have an AI problem. It has a human-attention problem.
“Make reasonable changes so that every patient is served
in the right way.” — Taylor
Johnson, American Medical Association
The hospital had a talent show.
That sounds almost absurd when you think about what was
happening on the seventh floor of University of Michigan Health C.S. Mott
Children’s Hospital in Ann Arbor.
There were children receiving cancer treatment.
There were chemotherapy appointments.
There were exhausting days, difficult diagnoses and long
stretches of uncertainty.
And every week, there was a talent show.
Patients performed card tricks.
They showed drawings.
They danced.
They played guitar.
Staff members joined in.
Families applauded.
People sometimes stood outside the room just to watch.
The event began in February 2025 with patients and staff and
eventually grew into a weekly gathering of at least 20 people.
One of those patients was Donovan “DJ” Sheedy.
DJ was diagnosed with osteosarcoma at 15.
During treatment, he connected with neurologic music
therapist Matthew Bessette, M.A., MT-BC, and began developing his guitar
skills.
Eventually, he started performing classic rock songs at the
talent show.
His mother, Cassandra Sheedy, watched him play.
And she noticed something extraordinary.
Not a change in his laboratory values.
Not a change in his imaging.
Not a change in his treatment plan.
A change in who he appeared to be.
“When he played, he became himself again.”
That sentence stopped me.
Because she didn't say:
“He became healthier.”
She didn't say:
“He forgot he was sick.”
She said:
He became himself again.
That distinction may tell us more about the future of
healthcare than another thousand-page report about artificial intelligence.
Because healthcare has become exceptionally good at
describing the patient.
We have diagnoses.
Codes.
Problem lists.
Risk scores.
Claims.
Prior authorizations.
Eligibility records.
Clinical documentation.
Revenue-cycle dashboards.
Artificial intelligence.
Machine learning.
Automation.
Predictive analytics.
Interoperability.
Agents.
APIs.
And yet, somewhere inside all that infrastructure, we can
still lose the person.
That is the paradox.
Healthcare has more information than ever—and somehow
less attention.
And I think that is where the conversation about AI, medical
billing and administrative burden needs to go next.
Not:
How much more work can AI do?
But:
How much unnecessary work can AI eliminate?
The uncomfortable question nobody wants to ask
Healthcare loves efficiency.
At least, healthcare loves talking about efficiency.
We optimize workflows.
We automate workflows.
We digitize workflows.
We build dashboards to monitor workflows.
Then we hire people to manage the exceptions created by the
workflows.
Then we build software to manage those people.
Then we build AI to automate the software.
At some point, we should probably stop and ask:
What if the workflow itself is the problem?
That is not a rhetorical question.
It is an economic one.
Consider the modern medical practice.
A patient schedules an appointment.
Someone verifies insurance.
Someone checks eligibility.
Someone determines whether a referral is required.
Someone determines whether prior authorization is required.
Someone submits documentation.
Someone checks a payer portal.
Someone waits.
Someone calls.
Someone faxes.
Someone follows up.
Someone discovers that a piece of information was missing.
Someone asks someone else for the information.
Someone updates another system.
Someone resubmits something.
Then the patient receives care.
Then the claim is created.
Then the claim encounters another set of rules.
Then someone discovers another discrepancy.
Then someone works the denial.
Then someone appeals.
Then someone follows up again.
And somewhere in this magnificent administrative ballet,
somebody eventually asks:
Why did this happen?
That question usually arrives far too late.
I don't think healthcare has a medical billing problem
Here is the contrarian part.
I don't think healthcare has a medical billing problem.
At least, not primarily.
I think healthcare has an information problem that
eventually becomes a billing problem.
Billing is simply where many upstream problems become
financially visible.
If the patient's insurance information is wrong, billing
discovers it.
If eligibility wasn't properly established, billing
discovers it.
If an authorization requirement wasn't identified, billing
discovers it.
If documentation doesn't support what happened, billing
discovers it.
If two systems contain conflicting information, billing
discovers it.
If nobody owns an unresolved task, billing discovers it.
If the organization discovers an error only after the claim
is submitted, billing gets blamed for the error.
Billing becomes the crime scene.
But the crime may have happened several rooms upstream.
And then we wonder why the billing department is so busy.
Of course it is busy.
We're sending it the bodies.
The rework tax
Every healthcare organization pays a hidden tax.
I call it the rework tax.
It is the human effort required to repair information that
should have been correct earlier.
A staff member checks something twice.
A physician answers another message.
A biller searches another portal.
An employee calls an insurer.
A coder investigates a discrepancy.
Someone requests missing documentation.
Someone re-enters information.
Someone explains the same thing to another department.
None of those tasks individually looks catastrophic.
That's why they survive.
One extra phone call doesn't destroy a practice.
One missing field doesn't bankrupt a health system.
One denial doesn't cause a crisis.
But multiply them by:
1,000 encounters.
10,000 encounters.
100,000 encounters.
Now you have infrastructure.
The problem isn't that humans are inefficient.
The problem is that humans are being used as middleware.
Think about that.
The most expensive, adaptable, intelligent component in the
healthcare system—another human being—is routinely deployed to reconcile two
computer systems that don't agree.
We have somehow reached the point where a nurse, physician,
medical assistant or biller can become the human equivalent of an API.
That should bother us.
The seventh-floor lesson
This is why the story of DJ matters.
The talent show did not cure cancer.
Nobody at C.S. Mott would confuse a guitar with
chemotherapy.
That isn't the point.
The point is that healthcare is not only about treating
disease.
It is about protecting the human being experiencing the
disease.
DJ didn't want to spend every day in his room doing
chemotherapy.
The talent show gave him something to do.
It gave him a reason to leave the room.
It gave him something to anticipate.
It gave him an identity that was not “cancer patient.”
Landen Silveus, another patient, learned card tricks while
hospitalized and eventually performed them at the talent show.
He described the experience simply:
“It got me out of my room.”
And:
“The talent show gave me something to look forward to.”
That is not a trivial outcome.
That is healthcare.
Gregory Yanik, M.D., a pediatric hematology-oncology
physician at Mott, sometimes joins the performances himself.
He described the event as:
“This is the highlight of my whole week.”
Think about the irony.
A physician working in pediatric oncology has an enormous
amount of clinically important work to do.
Yet one of the highlights of his week is standing in a room
while patients perform music and magic.
Why?
Because the room reminds everyone what medicine is actually
for.
People.
Not claims.
Not queues.
Not dashboards.
Not workflows.
Not documentation.
People.
Now replace the guitar with a claim
Here is where the story becomes uncomfortable.
Imagine replacing DJ's guitar with a claim.
The analogy sounds strange.
It shouldn't.
DJ's guitar created a moment where the system stepped back
and the person stepped forward.
Our administrative systems often do the opposite.
The person enters the system.
Then the system creates tasks.
Then the tasks create queues.
Then the queues create exceptions.
Then the exceptions create more work.
Eventually, the human being disappears behind the machinery.
We call this administrative burden.
But perhaps that phrase is too gentle.
Maybe we should call it what it actually is:
attention extraction.
Healthcare continuously extracts human attention to
compensate for fragmented information.
Physicians pay.
Staff pay.
Patients pay.
Families pay.
Practices pay.
Health systems pay.
And the currency is not always dollars.
Sometimes it is five minutes.
Sometimes it is an afternoon.
Sometimes it is a delayed treatment.
Sometimes it is a frustrated employee.
Sometimes it is a physician who goes home mentally exhausted
because the clinical work was only half the job.
The numbers are not subtle
The American Medical Association's latest
prior-authorization survey makes the scale difficult to ignore.
Physicians report completing about 40 prior
authorizations per week.
Those requests consume approximately 13 hours of
physician and staff time each week.
94% report that prior authorization contributes to
burnout.
40% employ staff dedicated exclusively to prior
authorization work.
And only 33% believe the latest insurer commitments
will make a meaningful difference.
That is not a small workflow inconvenience.
That is an operating model.
And it raises an important question:
What exactly are we optimizing?
If an organization spends 13 hours a week managing a process
that everybody agrees is burdensome, the answer cannot simply be:
“Let's make the employees faster.”
Maybe the better question is:
Why does this work exist?
We have confused complexity with sophistication
Healthcare has a strange relationship with complexity.
If something is complicated, we assume it must be
sophisticated.
Not necessarily.
Sometimes complexity is just accumulated history.
One rule gets added.
Then another.
One payer creates another requirement.
Another system gets purchased.
Another portal appears.
Another exception is created.
Another department gets involved.
Another spreadsheet is born.
And eventually someone says:
“That's just how healthcare works.”
No.
That's how our accumulated decisions work.
Those are different things.
A sophisticated system should reduce unnecessary complexity.
It shouldn't merely become better at managing it.
This is one of the biggest mistakes I see in healthcare
technology.
We build beautiful interfaces around ugly processes.
We automate repetitive work without questioning whether the
work should exist.
We give employees faster ways to do things that perhaps
shouldn't be done at all.
Automating a bad workflow can make a bad workflow faster.
Congratulations.
You've created a high-speed traffic jam.
And then we put AI on top
Now comes artificial intelligence.
AI is going to transform healthcare.
I believe that.
But I think much of the current conversation is aimed too
low.
We talk about AI writing notes.
AI summarizing charts.
AI coding encounters.
AI reviewing claims.
AI predicting denials.
AI drafting appeals.
AI answering messages.
AI analyzing documents.
All useful.
But notice the pattern.
Much of it happens after the complexity already exists.
We are using AI to clean up the mess.
What if AI could help prevent the mess?
That is a fundamentally different idea.
Instead of asking:
“What should AI do with this claim?”
Ask:
“What should have been true before this claim ever
existed?”
That's a much more interesting question.
From reactive revenue cycle to deterministic revenue
cycle
A reactive revenue cycle asks:
What went wrong?
A deterministic revenue cycle asks:
What must be true before this transaction can
successfully proceed?
That distinction changes everything.
Imagine a claim.
A reactive system waits for failure.
A deterministic system attempts to establish readiness.
Before submission, it asks:
Is the patient correctly identified?
Is the insurance information current?
Is eligibility confirmed?
Is an authorization required?
If authorization is required, has it been obtained?
Is the relevant documentation present?
Does the documentation support the service?
Is the coding consistent with the underlying clinical
information?
Are payer-specific requirements satisfied?
Are there conflicting pieces of information?
Does someone own the unresolved exception?
If something is missing, why are we sending it downstream?
That last question may be the most important.
Don't send uncertainty downstream and then hire somebody
to chase it.
Resolve it upstream whenever possible.
The middleman isn't always the villain
This is another distinction worth making.
The problem isn't that every person in billing, RCM or
administration is unnecessary.
Quite the opposite.
Good people doing difficult work are often the reason the
healthcare system functions at all.
The problem is asking those people to repeatedly compensate
for structural weaknesses.
A great biller can rescue a broken process.
A great biller should not have to rescue the same broken
process 400 times.
There is an enormous difference between:
human judgment
and
human repair.
Healthcare needs more of the first.
It needs less of the second.
The future should not be:
AI replaces billers.
That is lazy thinking.
The future should be:
AI eliminates predictable administrative repair so humans
can focus on the exceptions that actually require judgment.
The biller becomes less of a data detective and more of an
exception specialist.
That is a better job.
A more valuable job.
And potentially a much more satisfying one.
The physician's attention is a healthcare asset
We measure physician productivity constantly.
Visits per day.
RVUs.
Collections.
Panel size.
Documentation time.
Access.
Throughput.
Length of stay.
But there is another asset we rarely measure:
attention.
Attention is finite.
A physician has a limited amount of cognitive bandwidth each
day.
So does a nurse.
So does a medical assistant.
So does a practice manager.
So does a patient.
Every unnecessary administrative interruption consumes some
of it.
That makes administrative friction more than a financial
problem.
It becomes a clinical resource allocation problem.
If a physician spends 20 minutes resolving an avoidable
administrative issue, those 20 minutes came from somewhere.
Maybe from the next patient.
Maybe from chart review.
Maybe from thinking.
Maybe from family.
Maybe from sleep.
Maybe from the physician's ability to recover.
The cost isn't always visible on the balance sheet.
But the cost exists.
The hidden invoice nobody sends
Healthcare sends patients invoices.
It sends payers claims.
It sends practices reports.
But there is another invoice nobody sends.
It is the invoice for attention consumed by unnecessary
work.
The medical assistant who spends an hour searching three
portals.
The physician who responds to a message that should have
been routed automatically.
The office manager who discovers that an authorization
expired two days ago.
The biller who finds a discrepancy that could have been
caught at registration.
The patient who makes another phone call because nobody
knows who owns the problem.
Nobody receives an invoice for this.
But somebody pays.
Usually everyone.
The question physician-owners should ask
If I were sitting with a physician-owner today, I would not
begin with:
“What billing software do you use?”
I'd ask:
“Where does your practice repeatedly have to remember
something that the system should remember for you?”
That question is revealing.
Where does someone maintain a spreadsheet because the
software doesn't maintain state?
Where does somebody manually check a payer portal every
morning?
Where does someone keep a personal list of unresolved
authorizations?
Where does the practice rely on one employee who “knows how
everything works”?
Where does a task disappear because nobody owns it?
Where does the same error happen repeatedly?
Where does the practice discover a problem only after money
is already at risk?
Those are not merely staffing problems.
They are information architecture problems.
Try a billing autopsy
Here's a practical exercise.
Don't start by buying another AI tool.
Take your last 20 denied or delayed claims.
Perform an autopsy.
For each one, ask:
1. What failed?
Not “what did the biller have to do?”
What actually failed?
2. Where did the failure originate?
At scheduling?
Registration?
Eligibility?
Referral?
Authorization?
Clinical documentation?
Coding?
Claim creation?
Payer processing?
3. When could we have known?
This is the critical question.
Could the problem have been identified:
Before the appointment?
Before the procedure?
Before documentation was closed?
Before coding?
Before claim submission?
Before denial?
4. Who knew?
Did somebody already possess the information?
If so, why didn't the right system know?
5. Who owned the problem?
If the answer is “everyone,” the answer is probably nobody.
6. What would have prevented it?
Not repaired it.
Prevented it.
7. Can the prevention be encoded?
If yes, that's where automation becomes interesting.
The Five-Whys test
Take one recurring billing problem.
Ask “why?” five times.
For example:
Why was the claim denied?
Because required authorization information was missing.
Why?
Because the authorization was not linked to the encounter.
Why?
Because authorization was tracked separately.
Why?
Because the scheduling workflow did not maintain
authorization state.
Why?
Because the practice treats authorization as a separate
administrative task rather than part of the patient's operational record.
Now you've found something interesting.
The problem wasn't “billing.”
The problem was workflow state.
That is where technology can create leverage.
What OnnX is actually trying to change
This is the philosophy behind OnnX.
Not:
“Let's build a better outsourced billing service.”
Not:
“Let's put an AI chatbot on top of RCM.”
Not:
“Let's replace every human in the revenue cycle.”
The idea is much more fundamental.
Make revenue a more predictable consequence of better
information.
OnnX is designed around the belief that much of the
downstream revenue-cycle burden originates upstream.
So the system should move upstream.
Capture information earlier.
Structure it.
Validate it.
Detect contradictions.
Identify missing information.
Assign ownership.
Maintain workflow state.
Escalate uncertainty.
Close the loop.
Then learn from the outcome.
The objective isn't to make the administrative maze more
entertaining.
It is to remove sections of the maze.
The OnnX architecture
Think of the revenue cycle not as a straight line but as a
connected information system.
Patient
↓
Insurance
↓
Eligibility
↓
Referral
↓
Authorization
↓
Encounter
↓
Documentation
↓
Coding
↓
Claim
↓
Payer
↓
Payment
↓
Exception / Resolution
The conventional approach treats many of these as separate
systems and separate tasks.
The more interesting approach is to treat them as connected
states.
If something changes upstream, the downstream workflow
should know.
If something becomes uncertain, somebody should own it.
If something is missing, the system should identify it.
If something is resolved, the workflow should close.
If the same exception happens repeatedly, the system should
learn from it.
That is not simply billing automation.
It is administrative intelligence.
AI should be boring
Here is another contrarian position:
The best healthcare AI may be boring.
Very boring.
It doesn't need to impress a physician with a clever
paragraph.
It doesn't need to sound human.
It doesn't need to write poetry.
It doesn't need to tell jokes.
It needs to notice that something is wrong before someone
else has to notice it.
It needs to know:
“This information is missing.”
“These two systems disagree.”
“This authorization expires tomorrow.”
“This task has no owner.”
“This claim is not ready.”
“This exception has happened 47 times.”
“This workflow has been waiting too long.”
That kind of AI isn't glamorous.
It is useful.
And usefulness is underrated.
The goal isn't to make AI look intelligent.
The goal is to make the system behave intelligently.
The AI agent should know when not to act
There is another important principle.
Healthcare AI should not be rewarded for doing the most.
It should be rewarded for knowing what it can safely do.
An administrative agent might:
- Retrieve
information.
- Compare
information.
- Identify
missing information.
- Classify
an exception.
- Prioritize
work.
- Draft
a response.
- Route
a task.
- Remind
an owner.
- Monitor
a deadline.
- Escalate
unresolved issues.
- Execute
bounded actions under explicit rules.
But when the system encounters ambiguity, it should stop.
That isn't failure.
That's intelligence.
The ability to say “I don't know” may be one of the most
important features of healthcare AI.
Don't remove the human. Remove the pointless human touch.
This is the distinction I wish more AI companies would make.
A human should remain in the loop when judgment matters.
But humans shouldn't have to remain in the loop simply
because two systems can't exchange information.
There is a difference.
Human judgment:
“Does this clinical situation warrant escalation?”
Human repair:
“Let me log into another portal and see whether the
authorization number matches the spreadsheet.”
The first requires a human.
The second requires better infrastructure.
What the industry gets wrong
There are several traps healthcare organizations should
avoid.
1. Automating before understanding
If you don't understand why the work exists, automation may
simply hide the problem.
Map the process first.
Then automate.
2. Measuring activity instead of outcomes
“Number of tasks automated” is not a meaningful healthcare
outcome by itself.
Better questions:
How many exceptions were prevented?
How many human touches disappeared?
How quickly are exceptions resolved?
How much revenue became more predictable?
How many administrative hours were returned?
3. Building another dashboard
Healthcare already has dashboards.
So many dashboards.
Dashboards have become the digital equivalent of putting
another clock on the wall.
Visibility is useful.
But visibility without action simply gives you a better view
of the problem.
4. Removing humans from ambiguity
Don't automate judgment just because you can.
Automate predictable work.
Escalate uncertainty.
5. Treating physicians as end users instead of
co-designers
If a technology changes physician workflow, physicians
should help design it.
Otherwise we risk creating software that looks brilliant in
a product demonstration and becomes another administrative burden in practice.
The metric nobody talks about
Healthcare technology companies love ROI.
Good.
But ROI shouldn't only mean dollars recovered.
We should measure something else:
Human attention recovered.
Imagine a practice could quantify:
- 11
hours of staff time returned each week.
- 6
fewer manual payer interactions per day.
- 25
fewer repetitive messages.
- 40
fewer unresolved tasks.
- 15
fewer avoidable claim corrections.
Those are not merely efficiency statistics.
They represent recovered human capacity.
That capacity can go somewhere.
A staff member can spend more time helping a patient.
A physician can think.
A manager can improve the practice.
A nurse can communicate with a family.
Someone can simply breathe.
Return on attention
We talk about return on investment.
I think healthcare needs another concept:
Return on attention.
If technology saves $50,000 but creates another 500 alerts,
it may not have created much value.
If technology reduces administrative noise and gives a
physician 30 uninterrupted minutes every afternoon, that may be enormously
valuable.
Attention is not an abstract resource.
Attention is how medicine happens.
A physician cannot think deeply while constantly switching
contexts.
A nurse cannot provide meaningful human connection while
simultaneously managing five administrative queues.
A practice manager cannot redesign a broken process while
spending every afternoon putting out fires.
So perhaps the real question is:
How much human attention does this technology give back?
That may become one of the most important measures of
healthcare technology.
A children's talent show and a billing platform have
something in common
This may sound like the strangest conclusion in healthcare
technology.
But DJ's guitar and a medical billing system actually share
a lesson.
The talent show created space for something human to happen.
It did not add another clinical protocol.
It did not create another dashboard.
It did not generate another administrative queue.
It created an opportunity.
A teenager could be a guitarist instead of a cancer patient.
A parent could be a proud parent instead of a worried
caregiver.
A physician could sing instead of simply rounding.
A hallway could sound like a concert instead of a hospital.
That is what good healthcare infrastructure should do.
It should create space for the human being to reappear.
The same principle applies to administrative technology.
If AI simply creates more alerts, more dashboards, more
tasks and more things to monitor, we have missed the point.
If AI removes unnecessary work, prevents avoidable errors
and gives people back time to do the work only humans can do, then we have
created something valuable.
The future of the medical biller
I don't believe the future is “AI replaces the biller.”
I think that framing is too simplistic.
The future could look more like this:
AI handles predictable information work.
Humans handle ambiguity.
AI monitors workflow state.
Humans handle judgment.
AI identifies patterns.
Humans decide when exceptions matter.
AI retrieves evidence.
Humans interpret difficult cases.
AI watches deadlines.
Humans intervene when the stakes are high.
AI learns from resolved exceptions.
Humans improve the underlying process.
The biller becomes less of a person chasing information and
more of a person managing meaningful exceptions.
That is not the elimination of the human role.
It is the elevation of it.
The ethical line matters
There is an obvious temptation in healthcare AI.
If the objective is revenue, the system can become optimized
for revenue at the expense of truth.
That cannot happen.
AI should never manufacture documentation.
It should never invent clinical facts.
It should never manipulate coding simply to increase
reimbursement.
It should never turn uncertainty into false certainty.
The objective should be:
Represent reality accurately.
If the clinical documentation supports the claim, help make
that information usable.
If the information is incomplete, identify the gap.
If the information conflicts, surface the conflict.
If the system does not know, say so.
That is where governance matters.
Healthcare AI needs:
- Auditability.
- Access
controls.
- Appropriate
human oversight.
- Data
minimization.
- Security.
- Clear
accountability.
- Appropriate
HIPAA safeguards.
- Business
associate agreements where applicable.
- Defined
boundaries for autonomous action.
- Monitoring
for errors and drift.
The smartest system is not the one that acts most
autonomously.
It is the one that knows where autonomy should stop.
What physician-owners can do Monday morning
You don't need to launch an AI transformation program next
week.
Start smaller.
Pick one recurring administrative headache.
Then ask five questions.
1. How often does it happen?
2. How many people touch it?
3. Where does the information originate?
4. At what point does the organization discover the
problem?
5. What could have prevented it?
Then calculate the human cost.
Not just dollars.
Count the touches.
Count the messages.
Count the phone calls.
Count the portal logins.
Count the handoffs.
Count the minutes.
Count the exceptions.
You'll probably discover something uncomfortable.
The biggest problem may not be the amount of work.
It may be the number of times the same information has to be
rediscovered.
Healthcare's next productivity revolution may be
subtraction
We've spent decades adding.
More systems.
More software.
More integrations.
More reports.
More rules.
More alerts.
More dashboards.
More staff.
More vendors.
More workflows.
Now AI arrives and gives us the ability to add even faster.
That should make us nervous.
Because technology can accelerate accumulation.
The next great healthcare technology companies may therefore
be the ones that practice subtraction.
Fewer handoffs.
Fewer duplicate entries.
Fewer manual checks.
Fewer unnecessary alerts.
Fewer disconnected queues.
Fewer exceptions.
Fewer reasons for someone to say:
“Let me check on that.”
That phrase may be one of the most expensive sentences in
healthcare.
The best AI may be the AI nobody notices
Imagine a future medical practice where:
The patient arrives.
Information is already structured.
Eligibility is already known.
Authorization requirements are already evaluated.
Missing information is identified before it becomes a
problem.
The encounter occurs.
Documentation supports what happened.
The claim is ready.
The payer receives coherent information.
Payment follows.
An exception occurs.
The system identifies it.
A human handles it because it actually requires judgment.
Nobody celebrates.
No one posts a screenshot of the AI.
No one says:
“Look what our revolutionary algorithm did.”
Because nothing dramatic happened.
And that is the point.
The system simply worked.
That is what mature technology looks like.
Invisible competence.
What DJ teaches us about healthcare technology
DJ eventually finished treatment.
He still returns to the seventh floor for the talent show.
Why?
Because the experience became bigger than the treatment.
He said the smiles of the children are why he keeps coming
back.
Think about that.
The patient who once needed the hospital became part of what
makes the hospital meaningful for other patients.
That is the kind of loop we should want in healthcare.
Not:
Problem → task → task → task → denial → appeal → payment.
But:
Person → information → action → resolution → learning →
better system.
The first loop consumes people.
The second loop learns from them.
The question I would put in front of every healthcare
technology company
Not:
How much work can your AI automate?
Ask:
How much unnecessary work disappears because your AI
exists?
Those are radically different questions.
The first rewards activity.
The second rewards subtraction.
The first can produce impressive demos.
The second produces better organizations.
The first may increase software utilization.
The second may increase human capacity.
And if the goal of healthcare is ultimately human health,
perhaps human capacity is the metric that matters most.
The OnnX thesis
This is the thinking behind OnnX.
Healthcare revenue should not depend on heroic
administrative effort.
It should increasingly become the predictable consequence of
high-quality clinical and operational information.
The system should identify uncertainty early.
It should structure information.
It should connect related states.
It should assign ownership.
It should maintain workflow state.
It should prevent predictable problems.
It should route exceptions to the right human.
And it should learn from what happens next.
That means moving the revenue cycle upstream.
Not waiting for the denial.
Not waiting for the rejection.
Not waiting for the unpaid claim.
Not waiting for the angry phone call.
Not waiting for someone to discover that a critical piece of
information disappeared three steps earlier.
Fix the information before it becomes expensive.
That's the opportunity.
The bigger healthcare question
DJ picked up a guitar because he needed something that
reminded him he was still himself.
Healthcare technology should serve the same purpose.
Not by pretending technology can replace humanity.
But by removing the unnecessary work that gets between
humans and humanity.
The physician should not spend the afternoon reconciling
information that the system could reconcile.
The nurse should not become a human routing engine.
The medical assistant should not be a walking reminder
system.
The biller should not have to reconstruct a patient's
administrative history from six disconnected sources.
The patient should not have to become the project manager of
their own care.
And the family should not have to navigate an administrative
maze simply to receive something the healthcare system already knows.
We can do better.
But doing better may require a different definition of
innovation.
Innovation isn't always adding capability.
Sometimes it is removing friction.
Sometimes it is removing a screen.
Sometimes it is removing a handoff.
Sometimes it is removing a form.
Sometimes it is removing a denial.
Sometimes it is removing the reason somebody has to call
somebody else.
And sometimes the most sophisticated thing technology can do
is simply get out of the way.
The challenge
So here's my challenge to physicians, clinic owners,
healthcare executives and healthcare technology builders:
Find the work in your organization that everybody accepts as
“just part of healthcare.”
Then question it.
Who created it?
Why does it exist?
Who touches it?
How often does it fail?
Where does the information originate?
Why isn't the system handling it?
What would happen if we eliminated it?
And perhaps the most important question:
What would your people do with the time you gave back?
Because that is the question technology companies too often
forget.
The goal isn't to create more technology.
The goal is to create more capacity for care.
Frequently Asked Questions
Isn't medical billing inherently complicated?
Some of it is.
But complexity and unpredictability are not the same thing.
A complex process can still be structured, observable and
predictable.
The opportunity is to reduce avoidable variability while
preserving appropriate human judgment.
Will AI eliminate medical billing jobs?
That should not be the goal.
The more productive objective is to eliminate repetitive
administrative repair and move human workers toward exception management,
judgment, relationship management and process improvement.
Why focus upstream instead of simply improving denial
management?
Denial management is important.
But a denial is often evidence that uncertainty survived too
long.
Fixing the denial is downstream recovery.
Preventing the underlying information defect is upstream
control.
You need both.
Does OnnX replace an EHR?
No.
The concept is an intelligence and orchestration layer
across the administrative lifecycle rather than simply another system of
record.
Can AI safely make healthcare decisions?
Some bounded administrative actions can potentially be
automated.
Clinical judgment and ambiguous high-stakes decisions
require appropriate human oversight.
The correct question is not whether AI can act.
It is where AI should be allowed to act without human
intervention.
What should a small physician-owned practice automate
first?
Start with a workflow that is:
- Frequent.
- Repetitive.
- Measurable.
- Rule-driven.
- Currently
dependent on manual coordination.
- Expensive
when it fails.
Then measure the result before expanding.
What is the most important metric?
Don't stop at dollars.
Measure:
human touches per transaction, exception rate, time to
resolution, preventable denials, administrative hours returned and human
attention recovered.
Three myths worth killing
Myth #1: “More automation means a better system.”
Not necessarily.
More automation can simply produce more automated
complexity.
Myth #2: “The billing department owns the billing
problem.”
Often the billing department owns the symptom.
The cause may originate much earlier.
Myth #3: “AI needs to replace humans to create massive
value.”
Wrong.
AI can create enormous value by making humans unnecessary
for work that never should have required human attention in the first place.
Final thought
A hospital talent show probably won't show up on a
revenue-cycle dashboard.
Neither will the moment when a teenager picks up a guitar
and, for a few minutes, becomes himself again.
But maybe that is exactly the problem.
We have become extremely good at measuring what is easy to
count.
Claims.
Denials.
RVUs.
Minutes.
Dollars.
Tasks.
Clicks.
Queues.
But healthcare is ultimately about things that are harder to
count.
Trust.
Attention.
Hope.
Dignity.
Connection.
Time.
Human presence.
The seventh floor at C.S. Mott offers a simple reminder.
A child can be sick without being defined by sickness.
A physician can be a physician without being buried in
administration.
A staff member can be a professional without becoming a
human API.
A patient can receive care without becoming an
administrative project.
And technology can help make all of that possible.
But only if we stop asking:
“How can we automate more?”
And start asking:
“What should no longer require a human being to do it?”
That is the question I think healthcare should be asking
next.
Because the future of healthcare AI may not belong to the
technology that does the most.
It may belong to the technology that knows what doesn't
need to be done at all.
And if we get that right, the payoff won't simply be a
cleaner revenue cycle.
It will be something much more valuable.
More human attention for healthcare.
More time to think.
More time to listen.
More time to care.
More time to be present.
More time to be human.
Just like DJ.
Get Involved
If you're a physician, clinic owner, healthcare operator or
healthcare technology builder thinking about how AI can reduce administrative
friction rather than simply automate it, I would like to hear your perspective.
The most interesting question isn't whether AI will change
healthcare.
It will.
The interesting question is:
What work should disappear because of it?
About the Author
Dr. Daniel Cham is a physician, medical consultant
and healthcare technology entrepreneur focused on healthcare operations,
medical technology, medical billing and practice management.
He is the founder of OnnX, an AI-powered medical
billing SaaS concept focused on helping small and medium-sized physician-owned
clinics reduce administrative friction by improving information quality,
workflow coordination and revenue-cycle predictability.
His work focuses on a simple premise:
Most of the problem starts upstream.
Instead of continually optimizing the downstream
administrative maze, healthcare technology should identify where unnecessary
complexity begins—and remove it.
Continue the Conversation
What is one administrative task in your practice that
everyone accepts as “normal” but probably shouldn't be?
Comment below.
I am especially interested in the workflows that nobody
questions anymore because they have been around for so long.
Free Resource
For physicians and clinic owners interested in reducing
administrative burden and understanding the upstream information problems
behind revenue-cycle friction:
Check the Featured section of my LinkedIn profile for a
free resource. No signup required.
References
Michigan
Medicine — “Talent show brings joy to kids at children’s hospital,” September
11, 2026.
The source for the Donovan “DJ” Sheedy story, the C.S. Mott seventh-floor
talent show, Cassandra Sheedy's observation, Landen Silveus, Matthew Bessette
and Gregory Yanik.
American
Medical Association — “Equip your physician practice to meet new ADA mandates,”
September 11, 2026.
Source of Taylor Johnson's statement about making reasonable changes so every
patient is served in the right way.
American
Medical Association — “AMA survey: Prior authorization reform pledge falls
short with physicians,” May 13, 2026.
Source for the current physician-reported administrative burden associated with
prior authorization, including workload, time, burnout and denial statistics.
American
Medical Association — “Physician burnout rates are falling, specialty gaps
remain,” April 16, 2026.
Source for the 2025 physician burnout figure and broader context surrounding
administrative burden and physician well-being.
Disclaimer
This article represents the author's perspective on
healthcare technology, administrative workflows and medical billing. It is
intended for educational and discussion purposes and does not constitute legal,
regulatory, financial, medical or coding advice.
Healthcare organizations should obtain appropriate
professional advice regarding specific compliance, reimbursement, privacy,
security and operational requirements.
Final Question
If AI could give your practice back 10 hours every week,
what would you want your people to do with those 10 hours?
That answer may tell us more about the future of healthcare
than any AI benchmark.
#OnnX #HealthcareAI #MedicalBilling #HealthcareTechnology
#PhysicianPractice #RevenueCycleManagement #HealthcareInnovation
#AIinHealthcare #AdministrativeBurden #PhysicianBurnout #HealthTech
#DigitalHealth #PracticeManagement
No comments:
Post a Comment