I Reverse Engineered Dental Insurance AI. Denial Is the Design.

They're not trying to process your claims accurately. They're trying to deny them efficiently.

I spent days reverse engineering the AI architecture that dental and health insurers use to adjudicate claims. What I found wasn't a system with flaws. It was a system where denial is the design.

Let me explain to you exactly how it works.

The Machine. How dental insurance AI processes claims

When you submit a claim, it enters a pipeline built to reject it.

First, your claim gets dissolved into data. Diagnosis codes, procedure codes, your prescribing history, your patient's demographics—all fed into a model trained to spot patterns across millions of claims. The system groups similar procedures mathematically—crowns with onlays, extractions with surgicals—so denial patterns learned from one get applied to all of them automatically.

Then comes the kill score. Before any human touches your claim, a machine learning model trained on historical denials calculates the probability yours should be denied too. Score too high? You're flagged as "high-risk" and routed to a reviewer.

Here's what they don't tell you: routing to human review is the denial mechanism. The only path to payment is algorithmic approval. Human review is where claims go to die.

Finally, the rubber stamp. When Cigna's PXDX system denied over 300,000 claims in two months, their doctors signed off in batches—1.2 seconds per case. They never opened the patient files. The policies promised "medical director review." The algorithm delivered assembly-line rejection.

A Senate committee confirmed it: AI-driven denial rates run 16 times higher than manual review.

Sixteen times.

That's not a bug. That's the product.

The Targeting. How AI profiles dental practices

These systems don't deny randomly. They hunt.

• They profile you. Submit above-average crown recommendations? The algorithm learns your pattern. Now every claim you file gets flagged before you click submit.
• They flag combinations. Certain diagnosis-procedure pairings trigger automatic suspicion—based on historical patterns that may encode decades of discrimination, not clinical reality.
• They exploit frequency. Patient needs a second root canal? Flagged as "potentially excessive." Clinical necessity is irrelevant. The model sees a pattern it was trained to reject.
• They punish demographics. These systems train on historical claims data. Whatever bias existed in that data—against certain zip codes, certain ages, certain populations—the algorithm inherits it. Then it hides that bias behind the clean language of "risk scoring."

They're not assessing your claims. They're predicting which ones you won't fight for.

The Math They're Betting On. Why dental insurance denials work

Here's the number that explains everything: 0.2%.

That's the percentage of patients who appeal denied claims.

UnitedHealth's nH Predict algorithm has a 90% error rate—nine out of ten denials get overturned when challenged. But UnitedHealth doesn't care. They're not optimizing for accuracy. They're optimizing for the 99.8% of patients who give up.

The lawsuit states plainly: insurers "knew that only a tiny minority of policyholders will appeal denied claims, and the vast majority will either pay out-of-pocket or forgo care."

Each denial costs your practice $25 to $118 to fight. They're betting you won't. They're usually right.

UnitedHealth made nearly $300 billion last year. Their own CEO admitted 85% of denials could be prevented with better processes.

They're not preventing them.

The Systems They Don't Want Named. Cigna, UnitedHealth, and EviCore exposed

Cigna's PXDX: Batch rejection without opening files. 1.2 seconds per claim. Not disclosed to patients. Policies explicitly promised human review.

UnitedHealth's nH Predict: 90% error rate. Targets elderly patients in rehabilitation. Internal pressure to keep stays within 1% of AI predictions—forcing patients to leave facilities early or drain their savings. Publicly called a "guide." Exposed in litigation as the actual decision-maker.

EviCore: Calibrated algorithms that can dial denial rates up or down. They call it "simple sorting technology." They're very careful never to call it AI.

The language is deliberate. They describe the systems one way in court, another way in marketing, and not at all in your policy documents.

What This Means for Your Practice. Fighting back against dental insurance AI

Every radiograph you submit becomes evidence in a system you'll never see. Your treatment patterns are profiled. Your claims are scored before you hit send.

The asymmetry is total: they have the algorithms, the data, the behavioral models. You have a fax machine and a prayer.

Three things change the math:

• Document like you're being prosecuted. Because algorithmically, you are. Every clinical note should assume a machine is scanning for denial triggers. Watch our podcast on defensible documentation →
• Appeal religiously. A 90% overturn rate means most denials are bluffs. The 0.2% appeal rate is the entire business model. Break it.
• Move as a group. One practice leaving a network is a rounding error. Ten practices threatening to leave gets the CEO on the phone.

The Point. Understanding the dental insurance denial system

I didn't reverse engineer these systems to scare you. I did it because you can't fight what you can't see.

Insurance companies have spent years building AI to deny claims at scale, suppress appeals through exhaustion, and extract profit from the gap between what they owe and what they pay.

The 16x denial rates. The 90% error rates. The 1.2-second reviews. The quotas. The profiles. The hidden algorithms.

This is not dysfunction. This is the architecture.

They built the machine. Now you know how it works.