Why independent medical practices should stop automating billing chaos—and start preventing it upstream.
“The problem isn't volume. It's coherence.” — Thea
Campbell, MBA, RHIA, FAHIMA, writing about healthcare revenue-cycle
intelligence
What if your most expensive billing problem began weeks
before your biller ever saw the claim?
The $100 Question Hiding Inside Every Denial
Imagine hiring someone to fix your front door.
Every morning, they arrive to repair the same broken lock.
They work faster every week. They install better tools. Eventually, you give
them artificial intelligence.
The repairs become spectacularly efficient.
There is just one small problem: nobody has fixed the door.
Welcome to the strange economics of healthcare
administration.
A patient arrives. Insurance information is incomplete. An
authorization detail is missing. Documentation doesn't clearly support the
service. A payer requirement has changed, but the relevant team never receives
the update.
The claim goes out anyway.
Weeks later, a denial arrives. A medical biller
investigates, gathers records, corrects the claim, drafts an appeal, follows
up, and documents the result. The practice waits for payment while staff spend
valuable time reconstructing information that should have been available
earlier.
Then someone buys another AI tool to automate the appeal.
The technology gets faster. The original problem gets
another opportunity to happen.
That is the distinction independent physicians, clinic
owners, and medical billing professionals need to understand: automating the
correction of a problem is not the same as preventing the problem.
AI can help with both. But the order matters.
The next competitive advantage in medical billing may not
come from the practice that processes denials fastest. It may come from the
practice that consistently prevents avoidable errors before claims leave the
building.
The Contrarian Thesis: More Automation Is Not Always More
Progress
Healthcare has a complicated relationship with
administrative technology.
When a process becomes inefficient, the natural response is
to automate it. If claims are denied, automate denial management. If
documentation is incomplete, introduce documentation AI. If authorization
requests are delayed, automate authorization workflows.
Each solution may be useful.
But there is a question we often skip:
Why is the same problem entering the system in the first
place?
The revenue cycle is not a collection of isolated
transactions. It is a chain of connected decisions and information.
Eligibility affects authorization. Authorization affects
scheduling and service delivery. Clinical documentation supports coding. Coding
informs claim submission. Claim submission interacts with payer-specific rules.
Payment outcomes should inform future workflows.
When those connections break, the practice pays for the same
information gap multiple times.
First, someone must locate the missing information. Then
someone must correct it. Another person may need to review it. The claim may
need to be resubmitted. Payment is delayed. And the same defect may affect the
next patient.
This is why the distinction between downstream automation
and upstream intelligence matters.
- Downstream
automation helps process, correct, appeal, and recover claims after
problems appear.
- Upstream
intelligence helps identify missing information, inconsistencies, and
known risks before they become preventable billing failures.
Neither approach makes the other obsolete. A practice will
still encounter legitimate disputes, changing payer rules, and situations that
require human judgment.
But if a problem can be prevented reliably before
submission, correcting it afterward should not be the preferred strategy.
A faster appeal is useful.
A correct claim that never needs that appeal is usually
better.
What the Latest Industry Discussion Reveals
In an October 20, 2026 article, Thea Campbell of Solventum
argued that healthcare's challenge is not simply the quantity of information
but its coherence. Her discussion of revenue-cycle intelligence emphasized
connecting fragmented data, identifying documentation gaps earlier, and
shifting from reactive correction toward prevention.
That perspective matters because healthcare organizations
already generate enormous amounts of information. Yet having data somewhere in
an electronic record does not mean the right person can find it, trust it,
interpret it, and act on it at the right time.
A payer requirement buried in one workflow may never reach
the person preparing the claim. A correction made after a denial may never
inform the next similar encounter. An authorization number may exist but remain
disconnected from the service it was intended to support.
The problem is not necessarily a lack of software.
Sometimes, the problem is that the software does not
preserve and propagate the right information across the work that depends on
it.
The uncomfortable counterpoint: AI can create new
problems, too
The Peterson Health Technology Institute raised a related
concern in a report covered by Becker's Hospital Review on October 14, 2026.
Its analysis argued that, under current fee-for-service incentives, some
AI-enabled documentation, coding, billing, and authorization tools could
reinforce inefficient workflows and increase costs rather than reduce waste.
Executive Director Caroline Pearson warned: “On our current trajectory, AI is
set to exacerbate existing system failures.”
That is a warning worth taking seriously—not proof that all
healthcare AI increases costs, and not an argument against automation itself.
It highlights a more important principle:
Automating a flawed process can make the flaw more
productive.
If an AI tool generates documentation faster but fails to
preserve accuracy, the practice has not necessarily improved. If it submits
more claims without resolving recurring information gaps, it may simply move
the workload to another department. If it identifies more billable details
without adequate clinical and compliance safeguards, it can create additional
audit and financial risk.
The objective should not be to maximize the number of tasks
AI performs.
The objective should be to improve the quality of decisions,
reduce preventable rework, protect appropriate reimbursement, and preserve
accountability.
The Economics of Fixing Things Twice
Medical billing costs are not limited to the invoice from a
billing company or the salary of an in-house biller.
They also include the less visible cost of administrative
repetition.
Consider a hypothetical independent practice that handles
2,000 claims each month. Suppose 10% require additional work because of
preventable information or process problems.
That means 200 claims need attention.
If each takes an average of 20 minutes to investigate and
correct, the practice spends approximately 66.7 staff hours per month on this
work.
If the average loaded labor cost is $35 per hour, that
represents about $2,333 in monthly labor.
These figures are illustrative assumptions, not industry
benchmarks. Actual costs vary by specialty, payer mix, staffing model, denial
category, and complexity.
Now imagine the practice reduces those preventable cases by
25% through better intake, information validation, and workflow coordination.
That would eliminate 50 cases per month, or about 16.7 staff
hours. At the assumed labor cost, the direct labor value would be roughly $583
monthly.
That calculation does not include the potential value of
faster payment, reduced write-offs, fewer patient billing surprises, or less
staff frustration. Nor does it account for implementation costs, false alerts,
or the fact that some denials are unavoidable.
The point is not that every clinic will achieve these
results.
The point is that prevention has an economic value that
practices can measure rather than merely assume.
And there is a second advantage: preventing a recurring
problem may eliminate future work across multiple teams, not just the person
who would otherwise correct the claim.
The Upstream–Downstream Distinction
|
Question |
Downstream approach |
Upstream approach |
|
When does it act? |
After a problem appears or a claim is submitted |
Before or during the workflow that creates the risk |
|
Primary purpose |
Correct, appeal, recover, and resolve |
Identify gaps, validate information, and prevent avoidable
failures |
|
Typical examples |
Denial queues, appeal drafting, claim correction |
Eligibility checks, authorization validation,
documentation completeness |
|
Useful measures |
Appeal success, recovery rate, days to resolution |
Preventable error rate, first-pass acceptance, rework
avoided |
|
Main limitation |
The original work and delay may already have occurred |
It cannot prevent every denial or replace expert judgment |
The strongest operating model combines both.
Use downstream tools to resolve problems that remain. Use
upstream intelligence to learn from those problems and reduce the likelihood
that the same preventable issue recurs.
A denial should not be the end of a workflow. It should be a
signal that can improve the next workflow.
Six Steps Toward Upstream Billing Intelligence
A small practice does not need to replace its entire
technology stack to begin. It can start with one recurring problem, a clear
owner, and a measurable baseline.
1. Capture the right information
Start with the information that materially affects whether a
claim can be supported and processed correctly.
Depending on the service and payer, that may include:
- Current
eligibility and coverage details.
- Required
authorization or referral information.
- Complete,
accurate clinical documentation.
- Correct
patient and provider identifiers.
- Applicable
coding and billing requirements.
- Relevant
payer-specific rules and supporting records.
Do not confuse collecting more data with collecting better
data.
A hundred fields that nobody uses are not necessarily more
valuable than ten fields that prevent a recurring failure.
The aim is to capture the right information at the right
time, from an appropriate source.
2. Structure it so people and systems can use it
Information that exists only in free-text notes, email
threads, scanned documents, or someone's memory is difficult to operationalize
reliably.
Where feasible, establish standardized fields, consistent
terminology, clear ownership, and validation rules.
For example, a practice may require a defined authorization
status before a particular service proceeds. The exact rule must reflect the
payer, service, applicable exceptions, and clinical realities—not a universal
assumption that every service requires prior authorization.
Structure turns information from something people search
for into something a workflow can use.
3. Preserve what has already been learned
When a biller discovers that a particular payer requires
additional documentation for a recurring service, the insight should not
disappear when the case is closed.
Record the finding, its source, the date it was confirmed,
its scope, and who is responsible for validating future changes.
A past denial can inform a future check. But it should not
automatically become a permanent rule: payer policies change, exceptions exist,
and historical patterns can be misleading.
The objective is reusable knowledge, not institutional
folklore disguised as an algorithm.
4. Propagate the information to the next decision
This is where many organizations lose value.
A correct authorization detail is not enough if scheduling
cannot see it. A documentation requirement is not enough if the relevant
clinical or billing team never receives it. A denial insight is not enough if
it never reaches the workflow that created the error.
Map the handoffs.
Ask who needs the information, when they need it, what
action it should trigger, and how the practice will know that the action
occurred.
5. Act before the problem becomes expensive
Use rules-based validation where straightforward checks are
sufficient. Consider AI when information is unstructured, patterns are complex,
or context must be synthesized across sources.
Examples include flagging missing required fields,
identifying inconsistencies, prioritizing exceptions, or surfacing a known
documentation risk before submission.
Not every problem requires an AI model. A deterministic rule
may be cheaper, easier to explain, and more reliable for a clearly defined
requirement.
AI should earn its place by improving a real workflow—not by
appearing in the product description.
6. Learn from outcomes
Track what happened after the intervention.
Was the claim accepted? Was the alert correct? Did staff
resolve the issue? Did the same defect recur? Did the workflow create
unnecessary interruptions? Did a change improve compliance without slowing
care?
Feed verified lessons back into the process.
This creates a practical learning loop:
Capture → Structure → Preserve → Propagate → Act → Learn.
That sequence is not a promise that every error can be
eliminated. It is a disciplined way to turn recurring administrative problems
into opportunities for improvement.
What Should a Physician-Owned Practice Measure?
If a vendor promises that its AI will “transform the revenue
cycle,” ask for the baseline, the denominator, the measurement period, and the
operational definition.
Start with these metrics:
1. Preventable denial rate. Track the proportion of
claims denied for clearly defined, addressable causes. Separate these from
medical-necessity disputes, payer policy conflicts, and other categories that
may not be preventable at the point of capture.
2. First-pass acceptance rate. Measure how often
claims are accepted on initial submission. Define acceptance consistently;
acceptance for processing is not the same as final payment.
3. Rework hours per 1,000 claims. This helps reveal
administrative effort that headline denial rates can hide.
4. Days to payment. Monitor whether cleaner
submissions and faster resolution are associated with shorter payment cycles.
Account for payer mix and other factors that affect timing.
5. Avoidable write-offs. Distinguish preventable
losses from legitimate contractual adjustments, noncovered services, and other
appropriate write-offs.
6. Alert precision. Determine how often the system's
warnings identify genuine, actionable problems. Too many false alerts can
undermine adoption and create alert fatigue.
7. Cost per resolved problem. Include software, staff
time, vendor fees, implementation, maintenance, and oversight—not just the cost
of processing a task.
8. Compliance and patient impact. Monitor
documentation accuracy, audit findings, patient billing complaints, and
inappropriate delays in care.
Establish a baseline before introducing a new tool. Compare
similar periods and claim populations where possible. Avoid crediting the
technology for changes that may have resulted from staffing, payer behavior,
seasonality, or other process improvements.
The question is not simply whether AI saved time.
It is whether the practice achieved a meaningful,
sustainable improvement without creating a new problem somewhere else.
Five Mistakes That Can Turn Prevention Into Another
Expense
Mistake 1: Automating before understanding the workflow.
If nobody can explain why a claim repeatedly fails, automation may reproduce
the same failure at scale.
Mistake 2: Treating historical patterns as permanent
payer rules. Historical data can reveal useful signals, but current
policies and documented requirements still matter.
Mistake 3: Assuming every denial is preventable. Some
denials involve disputed medical necessity, changing policies, contractual
disagreements, or facts that could not reasonably have been known beforehand.
Mistake 4: Removing experienced billers from the decision
process. Medical billing professionals carry context that may never be
fully represented in a dataset. Use technology to support that expertise, not
erase it.
Mistake 5: Measuring activity instead of outcomes. Faster
claim submission, more automated tasks, and more alerts do not automatically
mean fewer errors, better collections, or lower total costs.
The most dangerous implementation is one that produces
impressive dashboards while leaving the underlying process unchanged.
Five Myths About AI and Medical Billing
Myth 1: More AI means fewer denials.
Reality: Results depend on the quality of the information,
the design of the workflow, the type of denial, and whether staff can act on
the system's findings.
Myth 2: A denial is simply a billing department problem.
Reality: Some failures originate in registration,
eligibility, authorization, clinical documentation, scheduling, coding, or
handoffs between departments.
Myth 3: Upstream prevention eliminates the need for
appeals.
Reality: Even accurate, well-prepared claims can face
disputes and denials. Downstream expertise remains necessary.
Myth 4: Structured data makes human judgment unnecessary.
Reality: Standardization improves consistency, but
exceptions, ambiguous requirements, and clinical context still demand
professional judgment.
Myth 5: The most advanced AI is the best investment.
Reality: A simple validation rule that prevents a recurring
error may deliver more value than a sophisticated model that identifies the
problem only after it occurs.
Legal, Ethical, and Operational Guardrails
Prevention cannot become an excuse for overbilling,
inappropriate automation, or unsafe shortcuts.
A responsible upstream intelligence program should include:
- Clinical
and coding integrity: Documentation and codes must accurately reflect
the services provided and the clinical record. Technology should not
invent diagnoses, embellish medical necessity, or encourage unsupported
coding.
- Privacy
and security: Protect patient information, restrict access according
to role, and evaluate vendors' data handling, security controls, and
contractual obligations.
- Human
accountability: Define who reviews uncertain recommendations, approves
consequential changes, handles exceptions, and remains accountable for the
final decision.
- Auditability:
Preserve the relevant source information, system recommendation, material
changes, human decisions, and reasons for overriding or accepting a
recommendation where appropriate.
- Change
management: Monitor payer requirements, validate rules, document
updates, and retire outdated logic.
- Fair
and appropriate access: Ensure automated administrative checks do not
create inappropriate barriers to clinically necessary care.
Applicable requirements depend on the workflow, data
involved, payer contracts, and jurisdiction. Practices should obtain qualified
compliance and legal advice when needed.
The goal is not to automate every decision.
It is to make routine decisions more reliable while ensuring
that consequential or uncertain decisions receive the right level of review.
Where OnnX Fits: From Billing Automation to Upstream
Intelligence
This is the problem OnnX is being developed to address.
OnnX's thesis is straightforward: medical billing AI can
fix claims, but upstream intelligence aims to prevent the problems that break
them.
Built around physician and medical-billing experience, the
concept is to help small and medium-sized practices identify missing
information, inconsistencies, workflow risks, and recurring revenue-cycle
problems earlier—before they become avoidable rework.
The ambition is not to promise that every denial will
disappear, that every payer will behave predictably, or that human expertise
can be replaced.
It is to improve the quality of information moving through
the practice.
The proposed approach follows six stages:
- Capture:
Obtain relevant clinical, operational, and payer information.
- Structure:
Make the information consistent and usable.
- Preserve:
Retain verified knowledge and important changes.
- Propagate:
Deliver the right information to the people and workflows that need it.
- Act:
Surface risks and support appropriate decisions.
- Learn:
Use verified outcomes to improve future performance.
Over time, practice-specific context and validated outcomes
could help the system distinguish recurring patterns from isolated events.
That is a development objective, not a claim that the
product has already demonstrated a particular reduction in denials or operating
costs. Those outcomes must be tested and measured.
The broader principle is what matters: the revenue cycle
should not have to rediscover the same problem every time a new claim arrives.
The Future: From Reactive Revenue Recovery to Preventive
Intelligence
The next phase of healthcare automation will not be
determined by who can put the most AI agents into a workflow.
It will be determined by whether those agents have reliable
information, useful context, appropriate authority, and measurable
responsibilities.
Three developments are worth watching.
First, payer and provider intelligence will become more
connected. Organizations will increasingly seek to understand
payer-specific patterns and operational requirements before problems
accumulate. Historical patterns will remain useful, but they must be validated
against current requirements.
Second, AI will increasingly operate as a layer across
workflows rather than as a collection of isolated tools. A system that
connects eligibility, authorization, documentation, coding, and claim outcomes
may provide more value than disconnected automation at each stage—provided the
integrations are reliable and secure.
Third, accountability will become a competitive
advantage. Practices will need evidence that their systems improve
outcomes, preserve audit trails, protect patient information, and escalate
uncertainty appropriately.
The winners will not necessarily be the organizations with
the most automation.
They may be the ones that know which problems should never
have reached automation in the first place.
Frequently Asked Questions
Is upstream billing intelligence the same as denial
prevention software?
Not necessarily. Denial prevention software may focus on
particular checks or claim risks. Upstream billing intelligence is a broader
operating concept: connect relevant information, preserve verified knowledge,
identify problems earlier, and use outcomes to improve the workflows that
create claims.
Should a small medical practice invest in upstream tools
before downstream AI?
Not automatically. Start with the practice's most costly and
frequent problems. If a denial queue is already creating significant financial
exposure, downstream tools may be justified. If recurring errors originate in
eligibility, authorization, or documentation, upstream interventions may offer
a stronger starting point. Many practices will need both.
Can AI eliminate medical billing denials?
No. Some denials are preventable; others involve payer
disputes, medical necessity, policy interpretation, or circumstances outside
the practice's control. A credible solution should specify which categories it
addresses and report results against a defined baseline.
Do small practices need expensive AI platforms?
Not always. Standardized intake, reliable eligibility
checks, clear authorization workflows, simple validation rules, and consistent
denial-cause tracking may address important problems. Add AI where it solves a
problem that simpler approaches cannot handle effectively.
Will upstream intelligence replace medical billers?
The objective should be to reduce repetitive work and
improve the information available to experienced professionals. Human expertise
remains important for exceptions, payer-specific context, compliance, appeals,
and decisions that require judgment.
How should a practice evaluate a vendor's claims?
Request results for comparable practices and claim types, a
defined baseline, measurement periods, the denominator used, implementation
costs, false-alert rates, and evidence that improvements persist. Ask what the
system cannot do and how uncertain recommendations are handled.
What is the best first step?
Review a recent sample of denials and rework. Identify the
recurring causes, estimate their frequency and cost, and trace each one back to
the earliest point at which the problem could reasonably have been detected.
Choose one high-impact, measurable workflow to improve.
The Question We Should Be Asking
Healthcare has spent years building increasingly
sophisticated ways to find, correct, appeal, and recover from billing problems.
That work remains necessary.
But what if our next breakthrough is not a better way to
repair the process after it fails?
What if it is a better way to stop sending the same
preventable problem through the process in the first place?
If you own a medical practice or work in medical billing,
where do you see the greatest waste today: correcting claims downstream, or
failing to capture the right information upstream?
Share your experience in the comments. If you know a
physician, practice manager, or medical biller wrestling with this problem,
share this article with them. The most useful insight may come from the people
doing the work every day.
For a free resource, see my LinkedIn Featured section—no
signup needed.
About the Author
Dr. Daniel Cham is a physician and founder of OnnX,
an early-stage initiative focused on upstream healthcare revenue intelligence.
His perspective brings together medical practice experience, healthcare
operations, and medical billing, with an interest in how better information and
workflow design can reduce preventable administrative friction.
His work explores a simple question: How can healthcare
organizations spend less time correcting avoidable problems and more time
delivering care?
Connect: LinkedIn | Website | YouTube
Disclaimer
This article is for educational and informational
purposes only. It is not medical, legal, coding, reimbursement, or compliance
advice. Examples and financial calculations are illustrative, not guarantees of
performance. Practices should verify current payer requirements and applicable
laws and consult qualified professionals for decisions specific to their
operations.
References and Further Reading
- Campbell,
Thea. “The next advantage in revenue cycle is context.” Solventum, October
20, 2026. Read
the article.
- Bruce,
Giles. “AI scribes, billing tools drive up healthcare costs: Report.”
Becker’s Hospital Review, October 14, 2026. Read
the article.
- Shryock,
Todd. “Medical billing denials are rising: Can AI help health care
practices get paid faster?” Medical Economics, March 6, 2026. Read
the article.
- Bala,
Purnendu. “Where artificial intelligence actually helps in medical billing
and revenue cycle management, and where it doesn’t.” Medical Economics,
April 1, 2026. Read
the article.
- The
Hidden Cost of Medical Billing: Why Clinics Are Losing Revenue in Silence
(and How AI Is Reshaping the Middle Layer of Healthcare). Explore how
fragmented workflows and administrative rework contribute to revenue
leakage. Read
the article.
- Jenn
Carbajal Asked for a Second Opinion. Healthcare Leaders Need One Too. Examine
why better-connected information and intelligence matter in healthcare
decision-making. Read
the article.
- Navigating
the Rising Tide of Medical Claim Denials: Expert Insights and Strategies
for 2025. Review practical approaches to understanding denial causes
and improving revenue-cycle workflows. Read
the article.
#MedicalBilling #HealthcareAI #RevenueCycleManagement
#PhysicianOwnedPractices #HealthcareOperations #OnnX
No comments:
Post a Comment