Healthcare can record every detail and still lose the plot. Before we buy more AI, we should ask why our systems keep forgetting what they already know.
“Is there an economic or operational impact that’s going to translate to dollars?” — Nassib Chamoun, founder and CEO of the Health Data Analytics Institute, at the HIMSS AI in Healthcare Forum
That is an excellent question for healthcare AI. It is also an uncomfortable question for the companies selling it.
A model can impress in a demonstration. A chatbot can produce a convincing answer. An automation platform can complete a task in seconds.
But did the technology improve the outcome? Did it eliminate unnecessary work? Did it make the process more reliable? Did the financial benefit exceed the cost of implementation and maintenance?
Or did we simply teach the existing problem to move faster?
In medical billing, that last possibility deserves more attention than it gets.
A Man Survived a Heart Attack. Then He Had to Reconstruct His Own Story.
In December 2025, Amritpal Virdee was rushed to Midland Metropolitan University Hospital in England after suffering a life-threatening heart attack.
Doctors inserted a stent to restore blood flow to his heart. But his condition deteriorated. His heart stopped for several minutes, and he was admitted to the intensive care unit.
For a period of his recovery, much of his experience was missing from his memory.
While Amritpal was sedated, nurses documented his journey in an ICU diary. They recorded medical developments, family visits, conversations, and ordinary moments that might otherwise have disappeared.
His wife, Pam, and son, Ruben, were part of that journey. So was a visit from his brother.
During the visit, a nurse turned on the television so the brothers could watch the news and Match of the Day. The team discovered that Amritpal supported Newcastle United.
It was a small detail in an extraordinary situation.
Yet it helped preserve something a medical chart alone might not have conveyed: the person behind the patient.
When Amritpal later read the diary, he could begin to piece together what had happened while he could not remember it. He could understand not only the medical events but also the care, attention, and human moments surrounding them.
Catherine Beddowes, Senior Sister for Critical Care Follow Up Support at Midland Metropolitan University Hospital, explained that ICU diaries can preserve both medical milestones and personal details that help patients make sense of their experience.
Amritpal expressed his gratitude to the staff and doctors who cared for him.
The hospital published his story on October 23, 2026.
The lesson is not that a diary is a substitute for a medical record. It is that recording information and preserving its meaning are two different things.
Sometimes, caring for someone means remembering the things they cannot.
Now consider a very different setting.
A physician finishes seeing patients. The clinical work is done. The documentation is in the EHR. The billing team submits the claim.
Weeks later, someone discovers a problem.
An authorization number cannot be found. A payer requests additional information. A claim is denied. A staff member opens the EHR, checks the clearinghouse, searches email, logs into a payer portal, and asks a colleague whether anyone remembers what happened.
The question is familiar:
“Does anyone know what happened with this claim?”
No ICU alarm sounds. Nobody is fighting for their life. The consequences are not comparable to Amritpal’s medical emergency.
But the operational question is worth asking:
Why does healthcare repeatedly have to reconstruct information that should already be available?
In critical care, a diary helped a patient reconnect with his experience.
In medical billing, better operational memory could help a practice reconnect the dots between an encounter, the information supporting a claim, the actions already taken, and the work still required.
Different circumstances. Different stakes. One useful systems question: are we preserving information in a form that helps people understand what happened and decide what to do next?
Healthcare Has a Memory Problem. It Just Calls It an Administrative Expense.
Imagine a highly experienced medical biller.
She knows which payer requires an additional document. She remembers which physician routinely misses a particular field. She knows where the authorization number is usually stored, which claims need attention before Friday, and which payer representative might actually answer the phone.
Her knowledge is valuable.
But how much of it exists in the system—and how much exists only in her head?
She may use an EHR, a practice management system, a clearinghouse, payer portals, spreadsheets, email, and a collection of notes.
Each tool has a purpose. The problem emerges between them.
The authorization is in one place. The appointment is in another. The claim is somewhere else. The reason a staff member made a particular decision may be buried in an email from three weeks ago.
Our biller knows how the pieces fit together because she has spent years learning the exceptions.
She is not merely processing claims. She is compensating for the architecture.
Now imagine she takes two weeks off.
Suddenly, the practice discovers that its operating manual has been walking around wearing comfortable shoes.
The system may contain most of the information. It simply does not preserve enough of the context to make that information easy to use.
This is the hidden cost of fragmented operations.
The most expensive information in a medical practice may be the information everyone assumes someone else remembers.
The consequences include repeated searches, duplicate work, delayed follow-up, unclear ownership, and avoidable errors. Some work is necessary. Payer requirements can be complex, and legitimate exceptions will always exist.
But there is a difference between doing necessary work and repeatedly rediscovering information because the system failed to make it available.
The first is part of healthcare administration.
The second is digital archaeology.
The tools are digital. The excavation is manual.
The Contrarian Argument: More Automation Is Not Always More Progress
The usual pitch for healthcare technology is straightforward:
Find a repetitive task. Automate it. Save time.
Repeat until the practice becomes more efficient.
There is nothing wrong with this approach when the task is necessary, the automation is reliable, and the results are measurable.
The problem is the assumption underneath it: that the work being automated should exist in the first place.
Imagine a claim is denied because authorization information was not attached correctly.
One solution is to automate the process of finding the authorization number and attaching it to the claim.
That might help.
But what if the real problem is that the authorization was never connected to the relevant service? What if nobody owned the handoff? What if the information was captured in a system that the billing workflow could not access? What if the number was present but its scope or validity had not been verified?
Automating the final step might address the symptom while leaving the underlying failure untouched.
Congratulations. We have industrialized the workaround.
The same pattern appears in many business processes. Organizations add a dashboard to monitor a spreadsheet, then automate the spreadsheet, then build an alert to warn staff when the spreadsheet is wrong.
Eventually, someone asks whether the spreadsheet was solving the right problem in the first place.
Healthcare is not immune to this cycle.
The issue is not that automation is bad. It is that automation can make a poorly designed process more efficient at producing unnecessary work.
A useful principle from management thinker W. Edwards Deming is to examine the system in which people work rather than automatically blaming individuals for recurring failures.
If staff repeatedly miss the same step, training may help. But if the process hides critical information, leaves responsibility unclear, and offers no warning before submission, training alone may not solve the problem.
The person who touches the claim last is not necessarily the person—or system—that created the failure.
Before automating a task, ask three questions:
- Is this task necessary?
- Why does it exist?
- What would prevent the underlying problem from recurring?
The goal is not to make people better at working around broken systems. It is to need fewer workarounds.
That is a much more interesting AI problem.
A Claim Is Not an Isolated Transaction. It Is the Ending of a Story.
Medical billing is commonly divided into functions: eligibility verification, referral management, prior authorization, documentation, coding, claim submission, payment posting, denial management, and appeals.
Those functions matter. But treating them as isolated activities can obscure the larger picture.
A claim is the financial expression of a clinical and administrative journey.
Consider the sequence:
Patient → Insurance → Eligibility → Referral → Authorization → Encounter → Documentation → Coding → Claim → Payer → Payment → Denial or Exception → Resolution → Feedback
Every stage can create information the next stage needs.
An authorization is not just a number. It has context: patient, service, provider, payer, date, and applicable conditions.
A clinical note is not just a document. It supports the medical record and may support the services billed, subject to applicable documentation, coding, and payer requirements.
A denial is not just a negative payment outcome. It is evidence about a transaction that needs to be understood.
The system should preserve the relationships among these pieces of information.
Otherwise, the practice receives fragments and must reconstruct the story.
Think of a claim arriving at the end of the process like a stranger carrying a stack of papers and announcing, “I believe you owe me money.”
The billing team has to investigate who the stranger is, what happened, whether the paperwork is complete, and whether the request is valid.
That is not a particularly efficient way to run a revenue cycle.
A more reliable process makes it possible to trace how a claim came to exist, which information supported it, what requirements applied, which actions were completed, and what remains unresolved.
This is the difference between storing data and preserving operational context.
Data answers: “What information do we have?”
Context answers: “What does this information mean for this case?”
Operational memory answers: “What happened, who did what, why did they do it, and what needs to happen next?”
That final question is where the opportunity lies.
The Denial Is the Crime Scene, Not Necessarily the Crime.
A denial is visible. Its cause may not be.
When a claim is denied, the practice reviews the reason, checks documentation, investigates the payer’s decision, and determines whether to correct, appeal, or write off the claim.
That work is often necessary.
But the denial may be the last visible symptom of an earlier failure.
Perhaps eligibility was checked too early and not reverified after a change. Perhaps a referral was received but never connected to the appointment. Perhaps an authorization was obtained but not associated with the service. Perhaps the documentation was complete but the claim contained a mismatch.
Or perhaps the payer made an error despite the practice having followed the applicable requirements.
These causes are different. They require different responses.
A useful system should not treat every denial as proof that the practice made a mistake. Nor should it assume that every denial is the payer’s fault.
The better question is:
At what point did the conditions for a successful claim stop being true—and what evidence supports that conclusion?
That changes the work.
Instead of repeatedly correcting individual claims, the practice can identify patterns, investigate root causes, and determine which failures can be prevented.
A recurring eligibility problem might call for a change in verification timing.
An authorization problem might require better tracking of scope, expiration, or service matching.
A payer-specific pattern might require a contract review or a documented escalation strategy.
A documentation issue might require clinician education or a better information-capture process.
A system defect might require a technical change.
The correct response depends on the evidence.
The faster the practice can identify a recurring failure, the less time it spends rediscovering the same lesson.
The objective is not to eliminate every exception. It is to stop treating every exception as if it were the first one ever encountered.
The Human Cost of Reconstructing Information
Fragmented information affects more than the billing department.
A physician may be asked to clarify documentation long after the encounter.
A front-desk employee may repeat an eligibility check because the previous result cannot be found or trusted.
A billing specialist may spend time locating information another team member already collected.
A patient may receive an unexpected bill while the practice investigates an unresolved payer issue.
A practice owner may see revenue delayed without knowing whether the cause is payer behavior, an internal process failure, or both.
The American Medical Association’s prior authorization survey, reported in May 2026, illustrates the broader administrative burden.
The survey found that physicians reported an average of 40 prior authorization requests per week, consuming an average of 13 hours of physician and staff time. Ninety-five percent said prior authorization delayed access to necessary care, while 94% said it contributed to physician burnout.
These findings concern prior authorization specifically. They should not be misrepresented as measurements of all billing inefficiency.
But they raise a relevant question for independent practices: how much administrative effort is unavoidable, how much is driven by external requirements, and how much is caused by the way information and responsibility are organized internally?
A practice cannot eliminate every payer requirement.
It can investigate whether its own systems make compliance harder than necessary.
That distinction matters. It prevents practices from blaming themselves for problems they cannot control—and from overlooking problems they can fix.
Healthcare Is Buying More AI. Is It Getting More Value?
On October 23, 2026, MobiHealthNews reported on a HIMSS AI in Healthcare Forum discussion about measuring AI by adoption, outcomes, operational performance, and cost.
Nassib Chamoun, founder and CEO of the Health Data Analytics Institute, posed the question featured at the beginning of this article: is there an economic or operational impact that translates to dollars?
The point is important because technical performance is only one part of the equation.
A system can perform well in a test environment and still struggle in daily operations. Integration takes time. Staff need training. Data may be incomplete. A tool may require ongoing maintenance. A workflow may be technically automated but still require substantial human intervention.
The demo is not the business outcome.
For an independent clinic, the right questions include:
- Has the number of manual touches per claim declined?
- Is less time spent searching for information?
- Are avoidable corrections decreasing?
- Are recurring errors being prevented or merely processed faster?
- Can staff identify the next action without opening several systems?
- Does the benefit justify implementation, maintenance, and oversight costs?
A practice could adopt five new tools and still require staff to copy information between them, reconcile conflicting records, check multiple portals, and determine which system contains the authoritative answer.
The technology inventory would look impressive.
The employee’s day might look exactly the same.
That is why a better question is not, “How much of our revenue cycle have we automated?”
It is:
“How much unnecessary work has disappeared?”
That is the measure that matters.
Three Perspectives That Challenge the Status Quo
Atul Gawande: Expertise Needs Reliable Systems
In The Checklist Manifesto and Better, surgeon and writer Atul Gawande explored how structured practices can help professionals manage complexity and improve reliability.
The lesson is not that healthcare workers should be replaced by checklists or software.
It is that expertise alone is an unreliable substitute for systems that make critical information and steps consistently available.
A highly experienced biller may know how to handle a difficult claim. But the organization should not need that person’s memory to rediscover the same requirements every time a similar case appears.
Practical implication: Make important steps visible, make exceptions explicit, and preserve the information needed to repeat successful work.
W. Edwards Deming: Fix the System, Not Just the Last Person in the Chain
Deming’s work on quality management emphasized improving the system in which people work.
Applied to medical billing, that means examining recurring failures as patterns rather than treating every one as an isolated staff error.
If authorization information repeatedly fails to reach the billing team, another reminder email may not be the answer.
Investigate the handoff.
Who owns the information? Where is it recorded? How does the next person know it is complete? What happens when the information is missing or inconsistent?
Practical implication: Measure recurring failures, identify their causes, and redesign the conditions that produce them.
Nassib Chamoun: AI Needs to Prove Its Value
Chamoun’s October 23 comments at the HIMSS AI in Healthcare Forum emphasized the need to look beyond model performance to operational and economic impact.
For an independent practice, that is a useful standard. A tool is not valuable simply because it uses AI, offers an impressive interface, or completes a task quickly.
It is valuable when it solves a real problem at a cost and risk the practice can justify.
Practical implication: Evaluate technology against measurable outcomes, not feature lists or marketing claims.
Together, these perspectives point toward a common principle:
Reliable operations require capable people, useful technology, and a system that preserves information, clarifies responsibility, and makes the next appropriate action visible.
A Record Is Not the Same Thing as a Memory
Let us return to Amritpal’s ICU diary.
The nurses did not simply collect disconnected observations. They preserved a sequence of events in a form that helped him understand his experience later.
The diary connected details to a timeline and a human context.
Medical billing needs a different kind of operational record, but the principle is relevant.
A folder containing an authorization, a clinical note, a claim, a remittance advice, and an appeal letter may contain all the necessary information. Yet a staff member may still have to determine how those documents relate to one another.
A useful operational system should help answer:
- What happened?
- When did it happen?
- Who was responsible?
- What information supported the action?
- Which requirement or policy applied?
- What remains unresolved?
- What is the next action, and by when?
- If the decision is challenged, can the practice reconstruct the reasoning?
That last question becomes especially important as AI enters revenue-cycle operations.
It is not enough for a system to produce an answer. The practice needs to understand the basis for consequential actions, verify the supporting information, correct mistakes, and determine who is accountable.
A confident answer without a traceable basis can create a new problem while appearing to solve an old one.
Healthcare does not need automation that merely sounds certain.
It needs automation that can show its work.
The Next Generation of AI Should Not Just Answer Questions. It Should Help Finish Work.
Much of the public conversation about AI focuses on chat interfaces, generated text, summaries, and faster responses.
Those capabilities can be valuable.
But a medical practice does not collect revenue because someone wrote a clever paragraph about a denial.
Revenue-cycle work requires a sequence of connected actions: identify the problem, retrieve relevant information, check applicable requirements, determine the next appropriate step, route the work to the right person, and verify that the outcome occurred.
A system that identifies an authorization mismatch but leaves the staff member to locate the authorization, determine ownership, set a reminder, and follow up has solved only part of the problem.
A more useful system would connect detection to action while preserving appropriate human oversight.
For example, imagine a claim-preparation process that identifies missing information before submission.
Instead of merely displaying an error, the system could identify the missing item, locate relevant existing records where authorized, explain what remains unverified, assign the issue to the appropriate role, and track its resolution.
If the information cannot be confirmed, the system should say so.
If the situation requires clinical judgment, payer interpretation, or human approval, the system should escalate rather than improvise.
If an automated action occurs, the practice should be able to inspect what happened and why.
This is not a promise that every claim can be made error-free or every denial eliminated. Payers make decisions, rules vary, clinical circumstances differ, and some disputes are unavoidable.
It is a different definition of progress.
The objective is not to automate the appearance of work. It is to reduce the amount of work required to reach a reliable outcome.
That is where AI can move beyond being another interface and become a meaningful part of operational infrastructure.
A Practical Plan for Independent Practices
A practice does not need to rebuild its entire technology stack to start improving information flow.
It needs to identify where information is lost, where work is repeated, and where responsibility becomes unclear.
Step 1: Choose One Painful Process
Do not begin by attempting to automate the entire revenue cycle.
Choose one recurring problem, such as eligibility verification, authorization tracking, missing documentation, claim-status follow-up, or denial resolution.
Select a process with enough volume to measure but a manageable scope.
A narrow pilot makes it easier to identify whether a change actually improves performance.
Step 2: Map What Really Happens
Document what staff actually do—not merely what the procedure manual says they do.
Where is information captured? Which systems are consulted? When is information copied manually? Who owns each step? What happens when a required item is missing?
Where do staff rely on memory, informal messages, or personal relationships to move work forward?
The purpose is not to criticize staff. It is to reveal the hidden work required to keep the process functioning.
Step 3: Establish a Baseline
Measure the process before introducing a new tool.
Useful measures include:
- Manual touches per claim.
- Time spent locating or reconciling information.
- Percentage of claims requiring avoidable correction.
- Denial rate by category and payer.
- Days in accounts receivable, interpreted alongside payer mix and contract terms.
- Time from identifying an exception to resolving it.
- Percentage of tasks with clear ownership and a documented next action.
Do not assume every denial is preventable. Separate payer-driven, coverage-related, documentation-related, coding-related, eligibility-related, and other causes where the available evidence supports those categories.
A useful baseline distinguishes what the practice can change from what it cannot.
Step 4: Fix the Information Handoff
Before buying more software, ask whether existing systems can preserve the necessary information and make it available to the next person.
Can authorization details be linked to the relevant appointment or service?
Can claim status and follow-up responsibility be made visible?
Can staff see which information is confirmed, missing, outdated, or conflicting?
Can the practice establish a reliable source of truth for each critical field?
The answer may involve configuration, training, integration, process redesign, or new software. The right intervention depends on the cause.
Step 5: Define Ownership and Escalation
Every unresolved issue should have a clear owner, a next action, and an appropriate deadline.
A status label that says “pending” is not a plan.
Pending with whom? Pending for what? Since when? What happens if nothing changes?
A well-designed process makes those questions answerable without requiring a staff member to remember the history.
Step 6: Automate the Repetitive Work That Remains
Once the process is understood, automate appropriate repetitive tasks.
Examples may include routine status checks, reminders, routing, data validation, and identification of missing information.
Preserve human review for ambiguous cases, clinical decisions, exceptions requiring judgment, and actions with material consequences.
Do not allow automation to silently invent missing data, alter clinical documentation, or represent an unverified requirement as satisfied.
Step 7: Review Results and Adjust
Compare performance against the baseline.
Did manual touches decline? Did staff spend less time locating information? Did the targeted error category improve? Did unresolved exceptions decrease? Did the system create new work elsewhere?
A pilot that fails to improve the intended outcome still provides useful information—provided the practice examines the result honestly.
The objective is not to prove that the software was a good purchase.
The objective is to discover what actually works.
The Metrics That Matter More Than an AI Demo
A polished demonstration can make almost any workflow look simple.
Real practice operations are less cooperative.
Before purchasing an AI or revenue-cycle product, ask how it will affect measurable outcomes.
Operational reliability
Measure the percentage of tasks completed correctly the first time, the number of manual touches per claim, and the frequency of missing or conflicting information.
Resolution speed
Track the time from identifying an exception to assigning it, acting on it, and resolving it. Distinguish delays caused by the practice from delays caused by payers or other external requirements.
Financial performance
Monitor denial rates by category, avoidable rework, days in accounts receivable, collection performance, and write-offs. Interpret each measure in context; payer mix, specialty, contract terms, and case complexity matter.
Staff capacity
Measure time spent on repetitive follow-up, information retrieval, duplicate entry, and manual reconciliation. The purpose is to determine whether staff gain capacity for higher-value work—not merely whether a task has moved to another screen.
Auditability and safety
Measure whether the system can show the information used, actions taken, changes made, and approvals obtained. Review how it handles uncertainty, incorrect inputs, access controls, and exceptions.
One particularly useful measure is touches per claim.
If the number of touches falls while accuracy and appropriate oversight remain strong, the practice may be removing unnecessary work.
If the software reduces clicks but creates more corrections, escalations, or downstream denials, the improvement may be cosmetic.
The most important question is not how quickly a system can process a claim.
It is how reliably the practice can move from the patient encounter to an appropriate financial resolution.
Seven Myths That Deserve Retirement
Myth 1: “We need more automation.”
Sometimes. But if the process is poorly defined, automation can replicate the same problems at greater speed.
Better question: What recurring failure are we trying to prevent?
Myth 2: “Our EHR already contains the information.”
Perhaps. But the information may not be structured, current, connected to the relevant transaction, or available to the person who needs it.
Better question: Can the next person find and use the right information without reconstructing the case?
Myth 3: “Our experienced biller knows how to handle it.”
That expertise is valuable. It is also vulnerable if it exists only in one person’s memory.
Better question: Can the organization preserve the knowledge and process so the team can use them consistently?
Myth 4: “AI will eliminate denials.”
No responsible vendor should promise that. Some denials reflect payer decisions, coverage limits, legitimate documentation requirements, or disputes that require human judgment.
Better question: Which categories of avoidable errors can we reduce, and how will we prove it?
Myth 5: “A higher automation percentage means a better operation.”
Not necessarily. A highly automated process can still produce errors or require extensive human intervention.
Better question: What work disappeared, what errors declined, and what new risks emerged?
Myth 6: “The cheapest tool is the most affordable.”
A low subscription price may conceal integration expenses, training, support, duplicate workflows, or additional manual work.
Better question: What is the total cost of operating the process before and after implementation?
Myth 7: “The problem is always the payer.”
Payer complexity is real. So are internal process failures. Treating every problem as the payer’s fault prevents a practice from fixing failures it can control.
Better question: What evidence identifies the cause, and which part of the problem is within our control?
What to Look for When Evaluating a Revenue-Cycle Tool
An independent practice should not need a technology department the size of a hospital’s to evaluate whether a product is useful.
Start with practical questions.
Does the product fit the existing environment?
Ask which EHRs, practice management systems, clearinghouses, and data formats it supports. Clarify what requires integration, what remains manual, and what happens when an interface fails.
Does it preserve context?
Determine whether the system connects information to the relevant patient, encounter, authorization, claim, payer, and task. Ask how it handles conflicting or outdated data.
Can it explain its actions?
Ask whether staff can inspect the source information, decision rules, action history, and approvals associated with an automated outcome.
Does it handle uncertainty responsibly?
A system should distinguish verified information from inference. It should escalate ambiguous or consequential situations rather than manufacture certainty.
Does it reduce work across the process?
Request a pilot with baseline measurements. Count manual touches, time spent, corrections, and unresolved exceptions—not just the number of automated tasks.
Can the practice control access and protect information?
Review role-based permissions, authentication, encryption, logging, retention, vendor access, and incident response. Confirm whether a business associate agreement is required and in place before protected health information is handled by a vendor acting as a business associate.
What happens when the product is wrong?
Ask how errors are detected, corrected, reported, and reviewed. Establish who approves consequential actions and who remains accountable for operational decisions.
The strongest vendor conversations are not demonstrations of what AI can say.
They are discussions about what the system can reliably do, what it cannot do, and how the practice will know the difference.
Efficiency Is Not Permission to Ignore Responsibility
Healthcare automation operates within legal, contractual, clinical, and ethical constraints.
HIPAA obligations may apply to the practice and to vendors handling protected health information on its behalf. Depending on the relationship, a business associate agreement and appropriate safeguards may be required.
Access should be limited to authorized users and legitimate purposes. Practices should evaluate how information is transmitted, stored, retained, and deleted, and how security incidents are handled.
Billing and coding must remain consistent with applicable documentation, coverage, and coding requirements. AI should not be used to manufacture documentation, upcode services, create unsupported diagnoses, or submit claims based on facts that have not been established.
Automation should also respect the distinction between administrative support and clinical judgment. A tool that assists with workflow should not quietly assume authority over decisions it is not qualified or authorized to make.
Finally, accountability must remain clear.
If an AI system identifies an issue, someone should know whether it is a recommendation, a draft action, or an executed action. Staff should be able to inspect consequential decisions and correct errors.
The more autonomous a system becomes, the more important it is to define its boundaries, oversight, and audit trail.
Automate work. Do not automate away responsibility.
Practices should obtain appropriate legal, compliance, and technical advice for their specific circumstances rather than treating a software vendor’s assurances as a substitute for due diligence.
Where Healthcare AI Is Headed
The next phase of healthcare AI will not be defined solely by which model produces the most impressive response.
It will increasingly be defined by whether systems can operate reliably inside real organizations.
That means connecting information across systems, maintaining state, respecting permissions, handling exceptions, and providing evidence of what occurred.
In revenue-cycle operations, this could mean identifying a missing authorization before claim submission, surfacing an unresolved eligibility issue, routing a task to the right person, monitoring a deadline, and preserving the history needed to understand the result.
It could also mean recognizing that a problem cannot be solved automatically and escalating it before a deadline is missed.
The opportunity is substantial, but it should not be confused with a promise of perfect automation.
Payer requirements will continue to change. Clinical documentation will remain nuanced. Contracts will differ. Some claims will require appeals, negotiation, or human review. Systems will occasionally fail.
The goal is to make those realities manageable without forcing staff to reconstruct the same information repeatedly.
The future may contain fewer disconnected portals, fewer manual handoffs, and fewer situations in which the only person who understands a process is the employee who has worked there for twenty years.
That would be progress.
Not because technology has replaced human expertise, but because it has made that expertise easier to use, preserve, and share.
What This Means for OnnX
At OnnX, we approach medical billing from a different starting point.
Instead of assuming the answer is simply more downstream automation, we focus on the relationship between data, workflows, and revenue.
The premise is that many operational problems become harder and more expensive when information is incomplete, disconnected, or unavailable at the moment a decision must be made.
A denial may be the first visible sign of a problem that began much earlier.
A claim may require repeated correction because a prior step did not preserve the information the next step needed.
A billing team may spend hours chasing status updates because ownership, dependencies, and deadlines are not visible in one coherent operational process.
The opportunity is to address those problems as connected systems—not as a growing collection of isolated tasks.
For independent practices, the distinction matters.
A large health system may have specialized teams and substantial IT resources to absorb operational complexity. A small practice has fewer people, tighter margins, and less capacity to absorb repeated administrative work.
When a process depends on a particular employee’s memory, the practice carries an invisible operational risk.
When a system preserves the relevant information, tracks the work, and makes the next appropriate action visible, the organization can become less dependent on informal workarounds.
That is the direction OnnX is working toward: using AI-enabled operational intelligence to help independent clinics make their revenue-cycle processes more consistent, traceable, and manageable.
This is a product philosophy, not a claim that every capability is already deployed or that all billing problems can be eliminated. Specific results must be established through implementation, measurement, and validation.
Our underlying belief is that independent physicians should not have to choose between caring for patients and building a financially sustainable practice.
Predictable revenue is not the purpose of medicine.
But without financial stability, independent practices have less freedom to invest in their teams, serve their communities, and remain independent.
The work of improving medical billing should therefore begin before the denial, before the appeal, and often before the claim itself.
It should begin with the information that makes the entire process possible.
The Question We Should Be Asking
Amritpal Virdee’s ICU diary mattered because it preserved details that could otherwise have disappeared.
It helped him reconnect with a part of his life that he could not remember.
No one should confuse that deeply personal experience with a delayed insurance payment. The human consequences are different, and the comparison has limits.
But the story reminds us of a broader principle: information has value beyond the moment it is first recorded.
Its value depends on whether it remains available, understandable, and connected to the decisions that follow.
Healthcare has invested enormous effort in digitizing records and automating tasks.
The next challenge is to make the information within those systems work together.
For medical practices, that means fewer mysteries about why a claim failed, fewer hours spent searching for information already collected, and clearer accountability for what happens next.
For AI developers, it means moving beyond systems that merely generate answers toward systems that support reliable, traceable work.
For practice owners, it means evaluating technology by the burden it removes—not the number of features it adds.
And for the healthcare industry, it means recognizing that a sophisticated system can still be operationally forgetful.
We do not need every medical practice to become a technology company.
We need technology to make running a medical practice less unnecessarily complicated.
The next breakthrough in medical billing may not be another tool that processes claims faster. It may be a system that makes fewer claims require rescue in the first place.
That is a more interesting goal than simply adding AI to the billing department.
And it is a much better reason to build technology.
One final question for physicians and practice owners: If your billing team disappeared for two weeks, would your systems preserve the work—or would your practice discover that its most important operating manual was stored in someone’s head?
I would love to hear how physicians, practice administrators, and revenue-cycle professionals are tackling this problem.
What consumes more time in your practice today: doing the work, or figuring out what happened to work someone already did?
Share your experience in the comments. If this perspective would help another physician or practice owner, consider reposting the article.
About the Author
Dr. Daniel Cham is a physician, medical consultant, and healthcare technology entrepreneur. He is the founder of OnnX, which focuses on AI-enabled approaches to medical billing and revenue-cycle operations for independent medical practices.
His work explores how better information architecture, operational workflows, and responsible automation can help physicians spend less time navigating administrative complexity and more time building sustainable practices.
Continue the Conversation
For a free OnnX resource, visit the Featured section of my LinkedIn profile. The resource is intended to provide practical insight without requiring a signup.
Knowledge creates opportunity for progress.
References and Further Reading
- Sandwell and West Birmingham NHS Trust. “ICU diary helps heart attack patient piece together missing memories.” October 23, 2026. The original story of Amritpal Virdee, his family, and the ICU diary.
- MobiHealthNews. “Healthcare AI’s value depends on adoption, outcomes and costs.” October 23, 2026. Reporting on the HIMSS AI in Healthcare Forum, including Nassib Chamoun’s comments about measuring operational and economic impact.
- American Medical Association. “AMA survey: Prior authorization reform pledge falls short with physicians.” May 2026. Survey findings on physician workload, delayed care, and prior authorization.
Disclaimer
This article is intended for general educational and informational purposes. It does not constitute medical, legal, coding, reimbursement, compliance, or financial advice. Healthcare billing requirements vary by payer, contract, specialty, jurisdiction, and service. Practices should consult qualified professionals regarding their specific circumstances. The discussion of OnnX describes its guiding approach and intended value; it is not a guarantee of financial, clinical, or operational results.
#Healthcare #MedicalBilling #RevenueCycleManagement #HealthcareAI #ArtificialIntelligence #HealthTech #IndependentMedicalPractices #PhysicianEntrepreneurs #HealthcareInnovation #DataQuality #WorkflowAutomation #MedicalPracticeManagement #ClaimsManagement #PriorAuthorization #OnnX
Leave a Reply