Legal For A Price: How Algorithmic Denials Are Stealing Your Clinical Autonomy

Two weeks ago, we talked about the operational autonomy you trade when private equity buys your practice. Last week, we talked about the personal autonomy you trade when you sell and stay on as an associate.

This week, we are talking about the clinical autonomy being taken from you right now — whether you sell your practice or not.

UnitedHealth, the largest health insurer in the country, recently got caught running an algorithm that flagged mental health patients for “too much” therapy. If a patient had more than 30 sessions in eight months, or went twice a week, the algorithm automatically denied coverage to save money.

Regulators in California, New York, and Massachusetts called it illegal under federal mental health parity laws. They pointed out that UnitedHealth does not apply these arbitrary algorithmic caps to physical health claims.

So, UnitedHealth settled. They paid the fines. And then they kept right on running the exact same algorithm in Medicaid plans across dozens of other states.

A colleague of mine summed up the corporate healthcare reality perfectly: “We like fines in this country. A fine means something is legal for a price.”

When a multi-billion-dollar insurance company sits in a boardroom, a regulatory fine is not a punishment. It is a line item on the P&L. They calculate the ROI of breaking the law: if the denial algorithm saves them $500 million in payouts this quarter, and the state fines them $50 million next year, the algorithm just generated a 10x return. They budget for the fine the way a dental practice budgets for gloves.

That is a massive policy failure. But it is not just a medical problem.

The exact same algorithmic playbook is being weaponized against dentists every single day.

Practicing Math, Not Medicine

My brother, James DeLuca, published a detailed breakdown of the algorithmic denial machine in March 2026. His data is not theoretical. He pulled it from active client engagements, and the pattern he found was not isolated. It was structural.

Initial claim denial rates climbed to 11.8% in 2024, up from 10.2% in prior years. Forty-one percent of healthcare providers now report denial rates above 10%. That jump is not dentistry being practiced differently. That is payers acting differently — at scale, with machines.

Here is the mechanism driving it.

Payers have deployed computer vision platforms — Pearl, Overjet, and their proprietary equivalents — to conduct automated radiographic audits of clinical diagnostics. These are not human reviewers working through a queue. These are machine learning systems processing thousands of claims per minute, engineered for a single objective: cost containment through industrialized friction.

Here is what that looks like in your operatory.

The Buildup Threshold. Your standard policy for crown coverage requires that 50% or more of the supragingival tooth structure is missing before a core buildup (D2950) is medically necessary. For two decades, this threshold was interpreted by human reviewers who evaluated radiographs with clinical context and professional judgment. A tooth at 47% structural loss with visible undermining? A reasonable reviewer approved it. That reviewer has been replaced by an algorithm. The payer’s AI analyzes your radiograph at the pixel level. It calculates structural loss at exactly 48%. The system does not evaluate clinical context. It does not consider undermining, fracture risk, or your 20 years of restorative experience. It executes a binary comparison: 48% is less than 50%. Denied.

The Perio Algorithm. You diagnose active periodontal disease. The gums are bleeding, the pockets are 5mm. You submit the claim for Scaling and Root Planing (SRP). But the insurance company’s algorithm scans the attached 2D radiograph and decides there is not exactly 2mm of crestal bone loss visible at the specific angle the software requires. The result is an auto-denial. The algorithm does not care about the clinical reality in the chair. It does not know the patient is bleeding. It only cares about the mathematical threshold programmed by an actuary to save the payer money.

The Exam Frequency Cap. A patient comes in with acute pain, so you perform a Limited Exam (D0140). But the patient just had their Periodic Exam (D0120) four months ago. The algorithm is hard-coded to deny any additional exams within a six-month window. The result? You either work for free, or your front desk has to have an agonizing conversation with a patient in pain about why their “great insurance” will not cover the doctor looking at their tooth. Some doctors instinctively waive the D0140 fee or apply it to the needed treatment to help the patient — a generous impulse that can actually violate provider agreements and trigger a different kind of audit exposure. Every option has a downside. The algorithm wins regardless.

The Appeal Paradox

Here is the number that should make every practice owner stop and read it twice.

Roughly 90% of AI-driven denials are eventually overturned when they reach an administrative law judge. The denials are not clinically accurate. They are engineered for deterrence.

But only 1% of denials are ever formally appealed by providers.

The economics explain why. A payer denial engine processes claims in milliseconds. Your billing team processes appeals in hours. When an algorithm can generate ten thousand denials in a minute, a practice relying on manual processes cannot mathematically survive the volume. The administrative burden of investigating the denial, pulling the chart, drafting the narrative, assembling supporting documentation, and filing the formal appeal often exceeds the monetary value of the individual claim.

A $150 buildup denial is not worth an hour of a billing coordinator’s time. So the practice writes it off. Multiply that write-off across hundreds of claims per year.

This is not a system failure. This is a system performing exactly as designed. The payer profits from the delta between the volume of automated denials and the fraction that are contested. Your silent compliance funds their margin.

The Hidden Analytics

The individual denials are frustrating, but they are only the symptom. The disease is the analytics running quietly in the background.

Most dentists have never heard of platforms like Cotiviti, but these payer-side analytics engines know exactly who you are. They establish baseline treatment frequency benchmarks by procedure code, by geography, and by provider type. They track your SRP frequency. They track your D0140 utilization. They track your core buildup ratio.

If your diagnosis rate sits above their statistical baseline, you are flagged. Not because your clinical judgment is wrong, but because your pattern deviates from the norm the algorithm expects. The insurer is not auditing your notes because they think you are committing fraud. They are auditing you because the data said you were an outlier.

This creates a perverse reality: a dentist who is genuinely excellent at diagnosing periodontal disease will naturally have a higher SRP rate than a dentist who ignores it. But the algorithm cannot tell the difference between clinical excellence and over-treatment. It only sees the deviation.

The dentist who does their job well is, paradoxically, the one most likely to be flagged. Understanding your own clinical production data — and how it maps against these algorithmic baselines — is the first line of defense.

The Cost of Doing Business

When a payer weaponizes an algorithm, they shift the financial burden directly onto your overhead.

It is no longer just the lost revenue of the denied claim. It is the 45 minutes your front desk coordinator spends on hold trying to appeal it. It is the clinical time you spend writing narratives — because the doctor writes the narrative, even if the front desk coordinates the submission. It is the administrative burnout that drives your best team members out the door. The operational cost of turnover compounds on top of the denial losses.

And here is where it gets worse.

As James documented in his analysis, the regulatory environment is beginning to shift. California’s SB 1120 now explicitly prohibits insurers from denying coverage solely based on an AI algorithm, requiring licensed physician review of all medical necessity denials. The CMS Interoperability Rule, taking effect in 2026, mandates that payers provide specific, documented, transparent reasons for all claim denials.

The practices that have already built defensible clinical documentation will have a distinct advantage as these laws force human review back into the denial process. The practices still operating reactively will continue absorbing the losses while the regulatory machinery slowly catches up.

The Bridge to What Comes Next

Here is the detail that should keep you up at night.

The clinical coding patterns flagged by payer AI are the same patterns a buyer’s Quality of Earnings team will analyze when you bring your practice to market. If a payer’s algorithm has already determined that a meaningful percentage of your buildups do not meet pixel-level radiographic proof, a forensic auditor will find the same pattern in your historical data.

As James put it in Phantom EBITDA: the buyer does not need to hire Cotiviti. The payer has already done the audit for them.

Every denial in your ERA is a data point. The denial pattern becomes the QoE finding. The QoE finding becomes the multiple compression. The multiple compression becomes a six-figure reduction in your exit value.

That connection is worth understanding before you ever sit across from a buyer — because by the time the QoE team pulls your ERA data, the window to address it has already closed. Run your free EBITDA Leakage diagnostic →