Why AI’s Real Power Lies in the Claims Data Black Box

In the national conversation surrounding healthcare technology, a recurring narrative has taken hold: Artificial Intelligence is here to save the front office. We hear constantly about AI’s ability to automate patient scheduling, streamline clinical documentation, and reduce the administrative burden on nursing staff. While these improvements are undoubtedly valuable for point-of-care efficiency, they address only a fraction of the true financial crisis crippling our healthcare system.

The United States currently spends over $5.6 trillion annually on healthcare, yet research published in JAMA indicates that between $760 billion and $935 billion of that spending is pure waste. When accounting for the additional layers of overcharges, billing errors, and fraud that accumulate at the employer level, that figure swells to a staggering $1.6 trillion annually. This amount exceeds the GDP of over 180 countries.

“Front-office automation does not touch any of it,” says Jude Odu, Founder of Health Cost IQ and author of Model Optimal Care. “The real money, and the real waste, lives deeper in the system, inside claims data and provider pricing”.

The Pricing Black Box

For self-insured employers, who cover more than 160 million Americans, healthcare is often a “black box” of spending. These organizations collectively spend hundreds of billions of dollars each year, yet most lack basic visibility into where that money actually goes.

Traditionally, third-party administrators process hundreds of thousands of claims annually, but manual audits typically only review 5% to 10% of those transactions, leaving the rest unexamined. It is in this unexamined data that waste becomes measurable. Every duplicate charge, unbundled service, and instance where a provider bills three to five times fair market rates is visible in the claims data. If anyone is looking!

AI changes this equation by enabling machine learning systems to review 100% of claims in near real time. By comparing every line item against established benchmarks and Medicare reimbursement rates, AI can flag anomalies that human reviewers miss. Already, major institutions are seeing results; the CMS confirmed in 2023 that it now uses AI and machine learning to detect fraud patterns that were previously invisible to human eyes.

Common Distortions in Health Data

Once employers begin using AI to look closer, the scale of billing errors they uncover is often shocking. The American Medical Association has documented a 20% claims-processing error rate among commercial insurers, representing an estimated $17 billion in annual waste. Odu’s own analyses across plans covering over two million lives have found that as much as 50 cents of every dollar spent can be classified as wasteful or inefficiently allocated.

The most common issues uncovered include:

• Duplicate Claims: Identical services billed multiple times for the same patient on the same date.
• Upcoding: Billing for a higher-acuity service than what was delivered, such as a routine office visit being coded as a complex evaluation.
• Unbundling: Separating services that should be billed under a single care episode into individual line items to increase total charges.
• Excessive Facility Fees: Adding facility charges to services that could have been performed in lower-cost settings without clinical justification.

Perhaps most surprisingly, overpayments often stem from providers being reimbursed for more than they actually billed. In one analysis of a mid-size health plan, this single category represented over $5.5 million in unnecessary spending.

The Shift to Proactive Fiduciary Duty

The traditional model of healthcare cost control is reactive: processing claims throughout the year and hiring an auditor to review a sample after the money is already out the door. Recovering those funds is a difficult, administrative fight that most employers never fully win.

AI moves the intervention point forward. By analyzing claims (before payment is made), systems can pause or deny charges that trigger flags for upcoding or unbundling. With this idea in mind, predictive modeling can now analyze historical data to identify high-risk members, allowing care managers to intervene proactively before a $200,000 hospitalization occurs.

This transition is also increasingly driven by legal necessity. Under ERISA and the Consolidated Appropriations Act (CAA) of 2021, self-insured employers have a legal duty to ensure plan assets are spent prudently. In 2024 and 2025, several major corporations, including Johnson & Johnson and Wells Fargo, faced lawsuits alleging fiduciary failures related to drug pricing and vendor oversight.

That’s why AI matters. Odu concludes with, “The plan sponsor that can demonstrate continuous, AI-powered review of all claims, provider pricing benchmarked against market data, and regular pharmacy audits occupies a materially stronger fiduciary position than one relying on annual, random-sample reviews conducted by vendors with questionable financial interests.”

He continues, “AI applied to claims data turns a health plan from a fixed cost into a managed asset. The question is no longer whether employers can identify and prevent waste. It is whether they will take steps to do so”