One nurse turned pandemic grief into an act of compassion. Her story raises a provocative question: In an increasingly automated healthcare system, are we getting better at processing information while forgetting the people behind it?
“I believe it should be a physician-led endeavor.” — Jennifer Goldsack, inaugural leader of the American Medical Association’s Center for Digital Health and AI
What can 850 flowers teach a billion-dollar healthcare
technology industry?
On October 8, 2026, Sarah Skinner, a nurse practitioner and
critical-care coordinator at UMass Memorial Medical Center in Worcester,
Massachusetts, helped deliver 850 homegrown dahlias to hospital patients.
No new algorithm. No billion-dollar technology platform. No
complicated implementation roadmap.
Flowers.
Skinner’s annual Dahlia Day began with a personal hobby that
helped healthcare workers cope with the stress and grief of the COVID-19
pandemic. What started as a small idea grew into a community effort involving
nearly 50 participants, including gardeners, local groups, and members of the
New England Dahlia Society.
One nurse’s garden became something much larger: a reminder
that patients are cared for not only by their clinical teams but also by a
community that sees them as people.
Skinner has described the difference a flower can make to a
patient facing a difficult hospital stay. Her initiative has grown from a
handful of blooms into hundreds delivered across the medical center.
The flowers are the story. The context is the lesson.
A patient lying in a hospital bed is not simply a diagnosis,
a room number, or a collection of observations. That person may be frightened,
exhausted, lonely, or worried about the family waiting at home.
A flower does not solve the underlying medical problem. It
does something different: it acknowledges the human experience surrounding it.
And that distinction matters far beyond a hospital corridor.
The uncomfortable question: What happens when healthcare
remembers the record but loses the person?
Healthcare has become remarkably sophisticated at collecting
information.
We have electronic health records, clinical decision
support, predictive analytics, automated coding, eligibility systems, prior
authorization portals, claims platforms, and artificial intelligence that can
summarize a clinical encounter before a physician has finished taking off a
pair of gloves.
And yet, healthcare professionals still spend substantial
time navigating administrative processes that compete with patient care.
The contradiction is difficult to ignore.
We can capture thousands of data points about a patient
while failing to carry one crucial piece of context from one workflow to the
next.
A clinical encounter may be documented correctly, yet the
information needed to support a claim may be incomplete or inconsistent.
Coverage may be verified, but the supporting evidence may not follow the
encounter. An authorization may exist, but its scope, conditions, or dates may
not be reflected where the next person needs them.
The information may exist somewhere.
The problem is whether it remains meaningful, connected,
and usable when it matters.
That is not simply a data-storage problem. It is a context
problem.
And if that sounds abstract, consider the administrative
journey of a patient.
The patient receives care. The clinician documents the
encounter. Someone checks coverage. Someone else verifies authorization
requirements. A coder interprets the documentation. A billing team prepares the
claim. A payer processes it. If something does not align, another person
investigates.
Each step may be functioning according to its own rules.
But does the entire sequence preserve the same understanding
of what happened, what was required, what was verified, and what evidence
supports the next action?
A collection of individually efficient steps does not
automatically create an intelligent system.
Sometimes it creates an exceptionally efficient way to pass
confusion from one department to another.
With a tracking number.
Healthcare has a strange relationship with efficiency
When an administrative problem appears, the conventional
response is predictable.
Add another tool.
Add another dashboard.
Add another work queue.
Add another layer of automation to process the exceptions
generated by the previous layers of automation.
Soon, everyone has a dashboard, nobody has enough time, and
the physician is still trying to figure out why a claim related to a completed
encounter is stuck in a queue.
The software may be working exactly as designed.
That is the uncomfortable part.
A workflow can operate correctly and still produce a bad
outcome.
If eligibility information is incomplete, automating the
eligibility workflow may accelerate the handling of incomplete information.
If authorization requirements are misunderstood, faster
claim submission will not correct that misunderstanding.
If clinical documentation and administrative requirements
are disconnected, an AI coding tool may identify a documentation gap after the
opportunity to resolve it at the point of care has passed.
Automation can reduce repetitive work. It can also make a
flawed process move faster.
A faster wrong turn is still a wrong turn. It simply gets
you lost ahead of schedule.
The strategic question is not merely, “How can we automate
this task?”
It is:
“Why does this task keep becoming necessary, and what
information would prevent the problem from recurring?”
That question shifts attention from downstream correction to
upstream prevention.
It also changes how we should think about AI in healthcare.
A timely warning from the healthcare AI conversation
On October 12, 2026, the American Medical Association
published an interview with Jennifer Goldsack, its inaugural leader of the
Center for Digital Health and AI.
Goldsack argued that AI should be physician-led. She also
emphasized that technology is not a silver bullet and that adoption should be
grounded in evidence, meaningful clinical benefit, and the needs of physicians
and patients.
That perspective deserves attention.
The healthcare AI debate often gets trapped between two
extremes: technology will solve everything, or technology will undermine
everything.
Neither position is especially useful to a physician trying
to run a practice.
The practical question is whether a tool improves the work
that matters, preserves appropriate human judgment, and produces outcomes that
can be demonstrated.
For independent practices, the stakes are particularly
concrete. Technology must fit the workflow, work with available information,
earn staff trust, and justify its cost. A sophisticated system that requires a
second full-time employee to manage it is not necessarily an efficiency
improvement.
And an AI tool that produces an answer without preserving
the evidence behind that answer may create a new problem: confidence without
accountability.
Goldsack’s physician-led principle is therefore relevant
beyond clinical AI. It is relevant to the administrative systems that determine
how clinical work is documented, authorized, coded, billed, and paid.
The people responsible for care should help define what
the technology is supposed to accomplish.
Not merely what it can do.
The hidden human cost of administrative friction
Administrative complexity is often discussed as though it
were an inconvenience to the business office.
It is more consequential than that.
The American Medical Association has reported that
physicians spend an average of approximately 13 hours per week on prior
authorization-related tasks in its physician survey findings. The AMA has also
documented widespread physician concern about delays to necessary care and the
contribution of prior authorization to burnout.
These findings concern prior authorization specifically;
they should not be mistaken for a measure of all medical billing work. But they
illustrate how administrative requirements can consume professional capacity
and interfere with care.
For a physician-owned clinic, time spent resolving
administrative inconsistencies has an opportunity cost.
It is time not spent seeing another patient, reviewing a
complicated case, supporting a colleague, or improving the practice.
For staff, recurring exceptions mean repeated investigation,
phone calls, portal searches, documentation requests, and follow-ups.
For patients, administrative friction can mean uncertainty,
delayed treatment, confusing bills, or the exhausting experience of becoming
the messenger between organizations that already have their information.
The impact is not always dramatic enough to make headlines.
It is often a hundred small interruptions.
And a hundred small interruptions can become the operating
model.
The question is not whether every administrative task can be
eliminated. Healthcare requires verification, documentation, controls, and
accountability.
The question is whether the same avoidable work has to be
repeated because the system failed to preserve information that was already
known.
A well-designed healthcare workflow should not require
every participant to rediscover the same facts.
The denial is not always the problem. It may be the clue.
Consider a denied claim.
The instinct is to fix the claim, appeal it, resubmit it, or
assign someone to investigate it.
Sometimes that is exactly what is needed.
But a denial is also evidence that something did not align
with the payer’s requirements or processing rules. The reason may involve
coverage, authorization, documentation, coding, claim construction, payer
policy, or another factor. The denial itself does not prove which party made
the mistake.
That distinction matters.
If a practice treats every denial as an isolated billing
problem, it may repeatedly correct the same symptom without identifying the
underlying cause.
Imagine a clinic discovers that several claims have been
denied because authorization information is missing or inconsistent.
The immediate task is to address those claims.
The more valuable investigation asks:
- Was
authorization required for these services?
- Was
the correct authorization obtained?
- Were
the service dates and scope consistent with the authorization?
- Was
the evidence captured and available to the billing workflow?
- Did
the relevant information change after the initial verification?
- Could
the mismatch have been detected before the claim was submitted?
The answers may reveal different problems requiring
different interventions.
Perhaps the authorization was never obtained. Perhaps it
existed but was recorded incorrectly. Perhaps the information was available in
one system but never propagated to another. Perhaps the payer applied a rule
that needs further review.
Those are not interchangeable explanations.
The objective is not to assume every denial is
preventable. It is to learn which failures are preventable, identify them
earlier, and stop repeating them.
A mature system does more than clear a work queue. It learns
from the work entering the queue.
Otherwise, the queue becomes an institutional memory that
nobody has time to read.
The real AI opportunity may be upstream
Much of the current healthcare AI conversation focuses on
tasks that happen after information has already been generated.
AI summarizes notes. AI suggests codes. AI identifies claims
that need attention. AI predicts payment outcomes. AI helps process denials and
communicate with patients about bills.
These applications can be useful.
But there is another opportunity: improve the quality,
structure, and continuity of information before downstream processes depend on
it.
Call it upstream intelligence.
In a medical billing context, that means helping the
practice establish and preserve relevant administrative context as early as
possible.
For example:
Before care: Confirm applicable eligibility
information, identify authorization or referral requirements, and preserve the
evidence and timing of verification.
At the point of scheduling or service: Make relevant
restrictions, unresolved requirements, and status changes visible to the people
who need them.
During documentation and claim preparation: Check for
inconsistencies between available information and the requirements that apply,
without confusing an administrative rule with clinical judgment.
After submission: Analyze denials and exceptions to
identify recurring patterns, distinguish root causes from symptoms, and feed
those lessons back into earlier workflows.
This is not a promise that every denial can be prevented or
every claim can become deterministic. Payer rules change. Information can be
unavailable. Clinical circumstances are complex. Some disputes are legitimate
and some decisions require human review.
It is a different design philosophy.
Rather than asking AI to clean up an endless stream of
downstream problems, ask what must be captured, checked, preserved, and
communicated so that fewer avoidable problems reach those downstream stages.
Capture → Structure → Preserve → Propagate → Act → Learn.
That is a more useful starting point than simply adding AI
to every step.
The flower and the claim have something in common
At first glance, Sarah Skinner’s Dahlia Day and medical
billing have almost nothing in common.
One involves flowers and human connection. The other
involves eligibility rules, documentation, claims, and payment.
But there is a revealing parallel.
A flower delivered to a patient is more than a physical
object. Its meaning depends on context: someone thought about the patient, took
the time to grow it, and chose to offer a small gesture of care.
Likewise, a piece of information in healthcare is not
valuable simply because it exists. Its usefulness depends on whether the right
person can understand it and act on it at the right time.
A recorded authorization that cannot be found when needed
has limited operational value.
A diagnosis detached from the documentation supporting it
can create confusion.
A claim without the context needed to interpret it may
become an exception waiting to happen.
Healthcare does not need to romanticize administrative work.
A claim is not a bouquet, and an authorization portal is unlikely to inspire
poetry.
But both examples point to a broader principle:
Value is not created by collecting information alone.
Value depends on preserving meaning and enabling the right action.
Skinner’s initiative makes that principle visible because
the human purpose is obvious. The patient receives a flower and understands
that somebody cares.
In administrative healthcare, the purpose can disappear
beneath layers of systems and transactions.
The work continues. The meaning gets harder to see.
That is precisely why upstream design matters.
Five myths that keep healthcare stuck downstream
Myth 1: More data automatically means better decisions.
Reality: More data can create more noise if
information is incomplete, inconsistent, outdated, or disconnected from its
purpose.
The goal is not to collect everything. It is to preserve the
right information, its source, its timing, and its relevance to the next
decision.
Myth 2: AI will fix a broken workflow.
Reality: AI can help improve a workflow, but it does
not automatically correct unclear ownership, missing evidence, conflicting
rules, or poor process design.
Before automating a task, understand why the task exists and
what outcome it should produce.
Myth 3: A denial is a billing department problem.
Reality: Some denials are appropriately resolved by
billing. Others may reflect earlier failures in eligibility verification,
authorization, documentation, coding, or communication. Some arise from payer
decisions or rules outside the practice’s control.
Treat the denial as a signal to investigate—not an automatic
verdict about fault.
Myth 4: The best system is the one with the most
features.
Reality: Independent practices need tools that fit
their workflows, integrate with their existing systems, produce reliable
results, and earn the trust of the people expected to use them.
A feature that creates another manual handoff may be less
valuable than a simple check that prevents a recurring error.
Myth 5: Automation is successful when the work moves
faster.
Reality: Speed matters, but it is not the only
measure. A faster process that increases errors, rework, inappropriate billing,
or patient confusion is not a success.
Measure the entire outcome—not just the time saved at one
step.
A new concern: What happens when AI makes billing more
efficient at the wrong thing?
A Guardian report published October 12, 2026, examined the
growing use of AI in medical billing and debt collection, including tools that
communicate with patients about bills and systems that help determine payment
strategies.
The development illustrates an important distinction: automation
can optimize a process without necessarily improving the outcome for the person
affected by it.
A system may become more efficient at generating bills,
pursuing balances, or predicting payment behavior. Those capabilities do not,
by themselves, establish that the underlying bills are accurate, that patients
understand their obligations, or that the process treats people fairly.
The same principle applies on the provider side.
AI may help generate documentation or suggest codes, but
efficiency alone does not prove that the documentation accurately reflects the
encounter or that the resulting claim is appropriate.
The responsible question is not simply whether AI can
increase collections or reduce labor.
It is whether the system improves accuracy, transparency,
accountability, and the experience of the people on both sides of the
transaction.
For independent practices, that includes protecting the
integrity of claims and avoiding unsupported coding or billing. For patients,
it includes understandable communication and appropriate handling of disputed
or unaffordable bills.
The purpose of healthcare technology should not be to make
every transaction more aggressive or every decision more automated.
It should be to make the system more accurate, more
accountable, and more useful.
What independent practices can do in the next 30 days
A clinic does not need to launch a massive AI initiative to
begin improving its information workflows.
It can start by investigating a few recurring problems.
Week 1: Find the friction.
Choose one recurring administrative issue: authorization
delays, eligibility mismatches, missing documentation, or a common category of
claim denial.
Review a manageable sample of recent cases.
Document what happened, where the problem was first
detectable, what information was available, and how much additional work was
required.
Do not begin by purchasing software. Begin by understanding
the process.
Week 2: Identify the earliest useful checkpoint.
Ask where the problem could first have been detected
reliably.
Could the team have identified a missing authorization
before the service? Could a discrepancy have been flagged before claim
submission? Was a required piece of evidence available but inaccessible to the
person who needed it?
Distinguish what the practice can control from what depends
on a payer, external system, or clinical decision.
Week 3: Establish ownership and a feedback loop.
Assign responsibility for the specific check and define what
happens when the check fails.
Record the reason for each exception in a consistent way.
Give staff a route to escalate unclear or conflicting information instead of
forcing them to guess.
Review the results with the people who perform the work. A
workflow designed without the people using it is likely to create another
workflow that needs explaining.
Week 4: Measure whether the change helped.
Compare the results with the baseline.
Useful measures may include:
- The
frequency of the selected error or exception.
- The
percentage of cases resolved before claim submission.
- The
amount of staff time spent on repeat investigation.
- The
number of claims requiring rework.
- Time
from identifying an issue to resolving it.
- The
accuracy and completeness of the evidence supporting each action.
Use a defined measurement period and account for differences
in case mix or volume. Do not declare victory because one week looked better
than the previous week.
The purpose is to learn whether the intervention made a
measurable difference.
The first AI project may not need to be an AI project at
all. It may need to be a better-defined workflow.
What should healthcare leaders measure instead of
celebrating automation alone?
Automation rates and processing speed are useful, but
incomplete.
A system can automate a task while leaving the underlying
error rate unchanged. It can reduce handling time while increasing the number
of exceptions sent to another department.
For practice owners and healthcare technology leaders, a
more balanced scorecard should include four dimensions.
1. Accuracy: Are the relevant records, codes, and
claims correct and adequately supported?
2. Continuity: Does the information needed for the
next step arrive with enough context to be understood and used?
3. Operational burden: Are staff spending less time
on avoidable rework, repeated verification, and preventable handoffs?
4. Human impact: Does the process help clinicians and
staff focus on appropriate work, and does it reduce unnecessary friction for
patients?
The fourth measure is not a substitute for financial
performance. It is a reminder that healthcare exists to serve patients, not
merely to process transactions.
And the first three are not optional because a system has an
impressive AI label.
A practice should be able to explain what a tool does, what
evidence supports its use, where human review remains necessary, and how
success will be measured.
If the vendor can explain the model architecture but cannot
explain how the tool reduces a specific operational problem, the practice may
be looking at a technology demonstration rather than a business solution.
What this means for OnnX
OnnX is being developed around a specific thesis: healthcare
billing is not just a downstream claims-processing problem. It is also an
upstream information-quality and context problem.
Built by physicians and medical billers, OnnX is intended to
help small and medium-sized independent practices reduce avoidable
administrative friction by focusing on the information and requirements that
shape billing outcomes.
The focus is not to replace the physician’s judgment or turn
the practice into an automated factory.
It is to help reduce downstream chatter by improving how
relevant information is structured, checked, preserved, and made available to
the people who need it.
That includes the upstream context surrounding eligibility
and authorization: what was verified, what requirements apply, what evidence
exists, and what needs attention before the workflow advances.
The larger ambition is to make revenue-cycle operations more
predictable and less reactive.
That ambition must be tested against real workflows,
measurable outcomes, reliable integrations, and the realities of independent
practices. It is not a claim that every denial can be eliminated, every payer
rule can be anticipated, or every administrative task can be automated.
It is a design direction: solve the right problem earlier,
preserve the reasoning behind important actions, and learn from recurring
exceptions.
In other words, do not merely ask how to process the next
denial faster.
Ask why the same kind of problem keeps returning.
And ask whether the system could have recognized the
relevant context sooner.
The contrarian lesson from Dahlia Day
Sarah Skinner did not need an algorithm to recognize that
patients in a hospital might appreciate a gesture of kindness.
She saw an opportunity to bring something human into a
difficult environment. She acted on it. Other people joined her. The idea grew.
The technology industry often celebrates scale, and rightly
so when scale creates real value. But scale is not the same as significance.
A system can process millions of records and still fail to
preserve a crucial detail. A hospital can introduce sophisticated tools and
still leave staff struggling with fragmented workflows. A practice can automate
a large share of its administrative tasks and still burden its physicians with
preventable exceptions.
The answer is not to reject technology.
It is to be more demanding about what technology is for.
The most valuable healthcare AI may not be the system that
produces the most output. It may be the system that helps prevent unnecessary
work, makes uncertainty visible, preserves evidence, and allows people to make
better decisions.
That is a less glamorous sales pitch than promising to
automate everything.
It is also a more serious standard.
Because the ultimate objective is not to give physicians
more software to manage.
It is to help them spend more of their time doing what
brought them into medicine in the first place.
And if an initiative can deliver 850 flowers to patients by
bringing together a nurse, a garden, and a community, perhaps healthcare
technology leaders can also find ways to reconnect the pieces of a system that
too often operate in isolation.
The flower is not the cure.
The algorithm is not the cure, either.
The real opportunity is to build systems that preserve
what matters—and stop making people work so hard to recover what should never
have been lost.
Frequently asked questions
What is the main lesson of Sarah Skinner’s Dahlia Day?
A relatively small act of compassion can have a meaningful
effect when it recognizes the human experience of illness. The initiative also
demonstrates how an individual idea can grow through community participation.
What does this story have to do with medical billing?
The connection is a broader principle: information matters
when it preserves context and supports the right action. In billing,
information can exist in one system yet fail to reach the next person or
process in a usable form. That can contribute to avoidable investigation and
rework.
Can AI eliminate medical billing denials?
No. Denials have different causes, and some reflect payer
decisions, changing requirements, disputes, or circumstances outside a
practice’s control. AI may help identify inconsistencies and recurring
patterns, but its value depends on the quality of the information, the
workflow, and the human oversight surrounding it.
What is upstream intelligence in healthcare billing?
It is the practice of identifying, structuring, checking,
and preserving relevant information earlier in the workflow, so downstream
teams are less likely to encounter preventable gaps or mismatches. Examples
include eligibility context, authorization requirements, verification evidence,
and consistent exception tracking.
How should a small medical practice begin using AI?
Start with a clearly defined problem, a measurable baseline,
and a workflow that staff understand. Determine whether the problem is caused
by missing information, fragmented systems, unclear responsibility, or
repetitive manual work. Then evaluate whether a software or AI intervention
improves accuracy and reduces burden without introducing unacceptable risk.
What should practices ask AI vendors?
Ask which specific problem the product solves, what evidence
demonstrates its effectiveness, how it integrates with existing systems, how
errors are detected, what information is retained, when human review is
required, and how the practice can measure the outcome.
Final thought: What are we optimizing for?
We can optimize for more automation, faster processing,
larger dashboards, and higher volumes of transactions.
Or we can ask a more important question: Are we building
a healthcare system that helps people care for patients—or one that keeps
asking them to repair the system?
Sarah Skinner’s 850 dahlias offer a simple reminder.
Healthcare is ultimately a human endeavor. Technology earns its place when it
strengthens that endeavor rather than obscuring it.
What would change if we treated every recurring billing
exception not merely as another task to process, but as evidence that the
system may be missing context upstream?
I’d be interested in hearing from physicians, medical
billers, and practice owners: What is one recurring administrative problem your
team keeps fixing that should have been prevented earlier?
Share your experience in the comments. The answer may be
more useful than another dashboard.
About the author
Dr. Daniel Cham is a physician, entrepreneur, and
founder of OnnX, an AI-powered medical billing initiative focused on upstream
information quality, context, and reducing avoidable administrative friction
for independent medical practices.
His work explores the intersection of clinical practice,
healthcare operations, and technology—with a central question: How can
healthcare systems preserve the context needed to support better decisions and
allow physicians to focus more of their time on patient care?
Connect and follow:
- Website:
drdanielcham.com
- Spotify:
Listen to
the podcast
- YouTube:
Watch and subscribe
- X:
Follow updates
- Facebook:
Join the
conversation
- Substack:
Read more insights
- TikTok:
Explore short-form content
- Instagram:
Follow on Instagram
Disclaimer
This article is for general informational and educational
purposes. It does not provide medical, legal, coding, billing, or reimbursement
advice. Healthcare organizations should evaluate technology and workflow
changes against applicable laws, payer requirements, professional standards,
privacy obligations, and their own operational circumstances. The OnnX concepts
described here represent a product thesis and intended direction, not a
guarantee of product capabilities or outcomes.
Sources and further reading
- PEOPLE,
October 11, 2026. Nurse
Delivers Hundreds of Dahlias from Her Personal Garden to Patients at a
Mass. Hospital on “Dahlia Day”.
- American
Medical Association, October 12, 2026. How
the AMA Will Help Set the Agenda for Health AI’s Future.
- American
Medical Association, May 13, 2026. AMA
Survey: Prior Authorization Reform Pledge Falls Short with Physicians.
- The
Guardian, October 12, 2026. As
US Medical Debt Grows, AI Will Be Used More to Collect It.
Continue the conversation
If you are a physician or independent practice owner,
explore the free resources in the Featured section of my LinkedIn
profile. No signup is required.
What would healthcare look like if we stopped optimizing
the downstream noise and started fixing the information gaps upstream?
#Healthcare #Physicians #MedicalBilling #IndependentPractice
#PracticeManagement #RevenueCycleManagement #HealthcareInnovation
#ArtificialIntelligence #HealthTech #PhysicianBurnout #PatientExperience
#PatientCenteredCare #HealthcareLeadership #OnnX
No comments:
Post a Comment