The most expensive mistake is the one you could have prevented. Healthcare keeps buying smarter tools to fight denials—but the real breakthrough may be stopping preventable errors before they become claims, rework, and lost revenue.
“On our current trajectory, AI is set to exacerbate existing system failures.” — Caroline Pearson, executive director of the Peterson Health Technology Institute
A faster way to fix a preventable mistake is still a way
to fix a preventable mistake.
Imagine a small physician-owned clinic on an ordinary
Tuesday.
A patient arrives. The front desk has checked eligibility.
The authorization has supposedly been handled. The physician provides the
service. The documentation is completed. The biller submits the claim.
Then the denial arrives.
The authorization number is missing. The approved service
does not match the service performed. Or a payer requires documentation that
never reached the right place.
Now the office becomes a detective agency.
The biller searches the system. The practice manager calls
the payer. Someone asks the physician to clarify a note written weeks ago.
Another employee resubmits the claim. Everyone waits.
Eventually, the claim may get paid.
The team celebrates—not because the process worked, but
because it recovered from the process failing.
Healthcare has become remarkably good at cleaning up
messes. Why are we so reluctant to ask who keeps making them?
Nobody deliberately schedules rework. Nobody hires a medical
biller because they love chasing missing information. And no physician opens a
clinic hoping to spend the afternoon explaining an authorization that should
have been documented correctly in the first place.
Yet many practices spend valuable staff hours correcting
problems that began before billing ever touched the claim.
It is the administrative equivalent of mopping the floor
while the faucet is still running.
The mop may be excellent. The problem is upstream.
The industry may be solving the wrong half of the problem
Healthcare has invested heavily in revenue-cycle management,
coding tools, denial analytics, automated appeals, and increasingly
sophisticated AI.
These tools can be valuable. Claims still need to be
processed accurately. Denials still need to be investigated. Payers still make
decisions that require follow-up and, sometimes, appeals.
But a question deserves more attention:
What if the biggest opportunity is not fixing claims
faster, but preventing avoidable problems from reaching the claims department
at all?
Consider three common situations.
An authorization exists, but the information is not
available to the person preparing the claim.
A payer requires a particular document, but nobody
identifies the missing item until the claim is denied.
A clinical record contains the relevant information, but it
does not flow reliably into the next operational step.
In each case, the information may exist somewhere. The
failure occurs because it is incomplete, inconsistent, inaccessible, or
disconnected from the action that depends on it.
This is why I believe healthcare billing is not merely a
tooling problem. It is also an information-quality problem.
A new tool may help a team work around that problem. A
better-designed process may prevent it from recurring.
The distinction is important because an organization can
automate a bad workflow without improving the workflow itself.
Imagine hiring a robot to run in circles. The robot may run
faster, never complain, and produce an impressive activity report.
It will still be running in circles.
Three stages. One expensive chain reaction.
Revenue-cycle management is often divided into three broad
stages.
The front end: Registration, eligibility, benefits,
referrals, and prior authorization.
The middle: Clinical documentation, coding, charge
capture, and claim preparation.
The back end: Claim submission, payment posting,
denial management, appeals, and collections.
The problem is that these stages depend on one another,
while the information connecting them may not be dependable.
A demographic discrepancy can affect eligibility. A missing
authorization detail can delay reimbursement. Incomplete documentation can
create coding questions. A payer-specific requirement can be overlooked. A
charge can fail to match the documented service.
Each problem looks like a separate task when it reaches the
employee responsible for fixing it.
But the tasks may share a common cause: information did not
arrive at the right place, in the right form, at the right time.
The final person to touch an error is not necessarily the
person—or process—that caused it.
That is why blaming the billing department for every denial
is not just unfair. It can prevent the organization from finding the actual
failure.
If the same issue appears repeatedly, the right question is
not simply, “Who needs to work harder?”
It is, “Why does this issue keep surviving our process?”
The latest evidence: more technology does not
automatically mean less work
Recent developments offer a useful warning—and an
opportunity.
Prior authorization: a deadline is not the same as a
solution
In a September 2026 poll, the Medical Group Management
Association reported that 44% of responding medical group leaders said
prior-authorization turnaround had become slower in 2026 than in 2025. Only 7%
said it had become faster. The poll included 178 applicable responses.
The federal decision-time requirements introduced in 2026
were intended to improve timeliness for certain impacted payers and services.
Yet the practice's workload extends beyond the formal decision window.
Staff may need to determine whether authorization is
required, gather documentation, locate the correct submission channel, check
status, respond to requests for additional information, and follow up on
unresolved cases.
A payer's decision clock does not measure every minute a
clinic spends getting that decision.
A shorter deadline does not necessarily create a shorter
workflow.
For practice owners, the lesson is to measure the entire
process—not just the time between submission and the payer's response.
Oracle Health: AI is moving upstream
In September 2026, Oracle Health announced five planned AI
capabilities across revenue-cycle workflows, including prior authorization,
documentation quality, charge capture, professional coding, and appeals
management.
The announcement reflects a broader shift: using technology
to identify reimbursement problems earlier, rather than concentrating
exclusively on downstream correction.
That is a promising direction. But an announcement is not
proof of a realized financial outcome. Practices still need to evaluate
availability, accuracy, integration, implementation costs, and measurable
results.
For a small practice, the useful question is not, “Does this
product have AI?”
Ask instead: Which recurring problem will it prevent, and
how will we prove that it worked?
The warning from Peterson Health Technology Institute
Caroline Pearson's October 2026 analysis argues that AI's
potential depends on better evidence, appropriate incentives, safeguards, and
redesigned workflows.
The warning is not that AI has no value. It is that
technology can reinforce existing system failures if organizations automate
activity without improving outcomes.
That should resonate with every independent practice
considering a new platform.
If a tool generates more alerts, more documentation, more
reviews, and more appeals—but does not reduce the total burden, the practice
may have purchased a more sophisticated way to remain busy.
Automation is an amplifier. It does not decide which
problem deserves to be amplified.
The practical answer: move the intervention closer to the
source
Prevention does not mean eliminating every denial. Some
denials arise from payer decisions, coverage limitations, contract disputes, or
circumstances outside the practice's control.
It means identifying the problems that can be
prevented and addressing them earlier.
Here is a practical approach for a small or medium-sized
physician-owned practice.
First, choose one recurring failure. Do not start by
trying to transform the entire revenue cycle. Select a problem that occurs
frequently, consumes staff time, or delays payment.
Second, establish a baseline. Record how often it
happens, why it happens, how much staff time it consumes, and what financial
impact follows.
Third, trace the problem backward. Was the
information missing? Was it incorrect? Did it exist but fail to reach the next
person? Was the requirement misunderstood? Was a check performed too late?
Fourth, add a control at the earliest practical point.
This might be a pre-visit verification, a required-field check, an
authorization-to-service match, a documentation checklist, or a clear
exception-routing process.
Fifth, test whether the change works. Compare
performance with the baseline. Look at rework, staff time, claim acceptance,
denial reasons, and total cost.
Sixth, improve the process before expanding it. If
the same failure continues, determine whether the control is poorly designed,
the data are unreliable, or the root cause has not been addressed.
The point is not to add more checkpoints until staff need a
map to get through the day.
The point is to put the right checkpoint in the right
place.
Where AI actually helps
AI may help identify missing information, compare data
across documents, summarize payer requirements, flag inconsistencies, and
direct exceptions to the appropriate person.
But AI should support accurate work—not manufacture
certainty.
A useful system should be able to show which information
supports its recommendation, identify conflicts, acknowledge uncertainty, and
route ambiguous cases for human review.
It should not invent clinical documentation, recommend
unsupported coding, or treat an authorization as a guarantee of payment.
Before adopting a tool, ask the vendor:
- Can it
explain why it flagged a problem?
- Can it
identify the source information behind its recommendation?
- What
happens when information is missing or contradictory?
- Can
staff override an incorrect recommendation and document why?
- Does
it integrate with the systems the practice already uses?
- How
does the product perform across different payers and service types?
- What
are the total implementation, training, maintenance, and review costs?
- How
will we measure whether the tool improves outcomes?
Do not buy a promise of efficiency. Buy a measurable
improvement in a defined problem.
A system that saves five minutes on one task but adds ten
minutes of review elsewhere may not be an improvement. A system that reduces
avoidable errors but creates excessive false alerts may not be an improvement
either.
The relevant measure is the total effect on the workflow.
The metrics that matter more than a flashy dashboard
A polished dashboard can make almost any process look
impressive. Practice owners need measures that reflect operational reality.
Track:
- First-pass
claim acceptance: How often claims are accepted without correction or
resubmission.
- Avoidable
denial rate: How many denials arise from problems the practice could
reasonably prevent.
- Rework
hours: How much staff time goes toward investigating, correcting,
resubmitting, and appealing claims.
- Days
in accounts receivable: How long it takes to collect payment.
- Aged
receivables: How much outstanding revenue remains unpaid beyond 90
days.
- Authorization
cycle time: How long the full process takes, including preparation and
follow-up.
- Documentation
and charge discrepancies: How often issues are caught before
submission versus afterward.
- AI
performance: False positives, missed issues, staff overrides, and time
spent reviewing recommendations.
- Total
cost of ownership: Software, implementation, integration, training,
maintenance, and human oversight.
Compare like with like, and separate problems the practice
can prevent from those it cannot control.
For example, a lower denial rate is not automatically a win
if claims are being held too long before submission. Faster processing is not
automatically a win if it creates more corrections later.
A metric is useful only when it tells you whether the
system is getting better—not merely busier.
Five myths worth retiring
Myth 1: A clean claim guarantees payment.
It does not. Coverage, medical necessity, authorization,
documentation, contract terms, and payer policies can still affect
reimbursement.
Myth 2: More AI means fewer denials.
Not necessarily. Results depend on data quality,
integration, workflow design, accuracy, and oversight.
Myth 3: Every denial is billing's fault.
No. Problems may begin in scheduling, registration,
authorization, clinical documentation, payer processing, or other parts of the
system.
Myth 4: Conservative coding is always safer.
Accurate coding is the objective. Under-coding can fail to
represent legitimate services, while unsupported coding can create compliance
and repayment risks.
Myth 5: A secure-sounding product is a secure product.
Security depends on access controls, encryption, monitoring,
vendor practices, data governance, and incident response—not marketing
language.
The lesson is simple: labels do not replace evidence.
A small-practice ROI example
Suppose a practice spends 10 staff hours each month
correcting one category of preventable billing errors.
At an assumed fully loaded labor cost of $30 per hour, that
is $300 per month—or $3,600 per year.
If a workflow improvement reduces that rework by half, the
practice saves approximately five staff hours per month, worth $150 in direct
labor under these assumptions.
That is an illustration, not an industry benchmark.
The practice should also evaluate whether the change reduces
delayed cash flow, avoidable write-offs, overtime, repeated patient contacts,
or other measurable costs. Those benefits should be counted only when supported
by actual results.
Then subtract the cost of the intervention.
If the solution costs more than the problem it solves, it
may be the wrong solution—even if the demonstration is impressive.
The smartest investment is not the most advanced tool. It
is the intervention that delivers a worthwhile result at a defensible cost.
The bigger opportunity: upstream intelligence
This is the problem space that interests me as a
physician-founder.
I believe healthcare has an opportunity to move beyond
treating billing as a series of disconnected corrections and toward treating it
as a connected information process.
The sequence is straightforward in principle:
Capture → Structure → Preserve → Propagate → Act → Learn.
Capture relevant information when it becomes available.
Structure it so people and systems can use it consistently. Preserve its
context. Propagate it to the next workflow. Act when something important is
missing or inconsistent. Learn from recurring failures so the same problem is
less likely to happen again.
The hard part is making that sequence work in real
practices, where information is scattered across systems, payer requirements
vary, staff have limited time, and exceptions are unavoidable.
That is what I mean by upstream intelligence:
identifying preventable information failures earlier and making the next
correct action clearer.
OnnX is being developed around this approach to healthcare
revenue intelligence, with a focus on the needs of small and medium-sized
physician-owned practices.
The thesis is not that technology can eliminate every
billing problem. It is that some downstream problems can be reduced by
improving the information and decisions that precede them.
Medical billing AI helps fix claims. Upstream
intelligence aims to prevent the problems that break them.
That proposition must be validated through real workflows,
real users, measurable outcomes, and appropriate safeguards. The goal is not
another dashboard that tells the practice what went wrong yesterday. The goal
is to help the practice catch more preventable problems while there is still
time to do something about them.
The future belongs to systems that learn before they
repeat
Denial management will remain necessary. Payer requirements
will change. Documentation will still require clinical judgment. Some disputes
will remain unavoidable.
The goal is not a fantasy world with zero denials.
It is a system that distinguishes preventable errors from
legitimate disputes, recognizes recurring patterns, and makes the correct next
step easier.
The best technology may not be the system that generates the
most alerts or processes the most claims. It may be the one that quietly
prevents a familiar problem from returning.
No dramatic dashboard. No celebratory meeting. Just fewer
errors, less rework, and more time for patients.
That may not sound as exciting as the latest AI launch. But
neither does a well-functioning plumbing system. People tend to notice plumbing
only when it fails.
Healthcare billing deserves a higher standard than
celebrating how quickly we repair what we could have prevented.
A resolved denial pays attention to the past. A corrected
workflow protects the future.
For independent practices, that distinction can mean fewer
administrative distractions, more predictable operations, and a better use of
limited staff time.
The real breakthrough is not making every correction faster.
It is making fewer corrections necessary.
One question for physicians, practice managers, and
medical billers
Are we investing too much in fixing billing failures—and
too little in preventing them?
What recurring billing problem do you believe should be
caught upstream, before it becomes somebody else's emergency?
Share your experience in the comments. If this reflects what
you see in practice, consider reposting it to bring more physician owners,
practice managers, and medical billing professionals into the discussion.
About Dr. Daniel Cham
Dr. Daniel Cham is a physician, healthcare entrepreneur, and
founder of OnnX. His work explores medical billing, practice management,
healthcare technology, and the opportunity to improve the quality of
information moving through healthcare operations.
Disclaimer: This article is for educational and
informational purposes only. It does not constitute legal, medical, coding,
compliance, tax, or reimbursement advice. Requirements vary by payer, plan,
service, and jurisdiction. Practices should consult qualified professionals and
applicable authoritative guidance for decisions affecting their operations.
Connect: LinkedIn | Website | YouTube
P.S. A free resource is available in the LinkedIn
Featured section—no signup required.
References
1. Centers for Medicare & Medicaid Services (CMS). Interoperability
and Prior Authorization Final Rule (CMS-0057-F). Explains federal
prior-authorization decision timeframes and requirements for impacted payers.
2. Medical Group Management Association (MGMA), September 2,
2026. Fewer
Than 1 in 10 Practices See Faster Prior Auth Turnarounds in 2026. Reports
survey findings on medical practices’ experience with prior-authorization
turnaround times.
3. KFF, August 13, 2026. Prior
Authorization Metrics Provide New Insights into Insurer Practices, but Gaps
Remain. Analyzes insurer-reported authorization data, including denial
rates, appeal outcomes, response times, and limitations in publicly reported
metrics.
4. The
Hidden Cost of Medical Billing: Why Clinics Are Losing Revenue in Silence (and
How AI Is Reshaping the Middle Layer of Healthcare). Related article
examining documentation quality, revenue leakage, rework, and AI-assisted
billing.
5. Healthcare
Saved a Stranger’s Life in Minutes. Why Does It Take Months to Fix a Broken
Claim?. Connects healthcare’s ability to respond to clinical emergencies
with its continuing challenges in preventing administrative failures.
6. Unpacking
the Medical Billing Process: From Visit to Payment—What Experts Really Say
& What You Should Be Doing Now. Provides additional context on claim
denials, payer complexity, workflow inefficiencies, and revenue-cycle
management.
#MedicalBilling #HealthcareAI #RevenueCycleManagement
#PhysicianEntrepreneur #PracticeManagement #HealthcareInnovation #ClaimDenials
#RevenueIntegrity #ClinicalDocumentation #PriorAuthorization #HealthTech
#HealthcareOperations #UpstreamIntelligence
No comments:
Post a Comment