By Arun Rangamani

The use of generative artificial intelligence (GenAI) by health plans and pharmacy benefit managers has stirred a great deal of controversy, recently, as fears have grown that complex, highly nuanced coverage decisions were being automated by “black box” technologies. In fact, the more common, though less sensational, scenario that’s playing out as healthcare payers incorporate GenAI into their operations is improved communication with providers and, in many cases, better patient outcomes.

The key to making that possible: data-driven algorithms that are helping insurers spot and address billing inaccuracies, overpayments and incorrect coding, often before claims are even paid. This ability to get out in front of inaccuracies in near real-time is making it possible for healthcare payers to take a hands-on, collaborative approach with providers that simply was not possible in the old days of retroactive payment integrity audit and recovery efforts.

Coding Guideline Complexity Demystified

Contrary to popular portrayal in the press, healthcare billing inconsistencies are not all malicious. In fact, studies have shown that the largest source of waste in the U.S. healthcare system is administrative complexity. All told, this one variable contributes an estimated $260 billion in wasted healthcare expenditures annually.

In our work with the world’s leading health plans, we find that the most common real-world sources of administrative complexity are billing errors stemming from fee schedule updates and changes to guidelines and contracting details. Invariably, what happens is that frequent changes in the way different health services are coded and entered into billing systems can often result in small, but persistent errors that can accumulate into large sums over time.

For example, in one recent project, we found that a single coding guideline change for a common cardiovascular treatment was being consistently missed by providers, leading to a 75% rise in payer costs before the anomaly was identified. In another example, we found that a state fee schedule rate change was not correctly updated by an insurer, causing $800,000 in over-payments to providers.

Today, we’re able to use GenAI to instantly scour hundreds of thousands of submitted claims and cross-check each variable within those claims against the latest Medicare, Medicaid and private insurance fee schedules, guideline sand contracting details to identify these inaccuracies earlier – sometimes before a claim is even paid – to allow insurers and providers to rectify the issue before it becomes a much more significant problem.

GenAI Shifts Coverage Decisions from Adversarial to Educational

Similarly, when it comes to the prior authorization and coverage decisioning process, the introduction of GenAI into health plan workflows has allowed payers to identify nascent signals in the data by scouring through a wide variety of data sources including Centers for Medicare and Medicaid Services (CMS) guidelines, medical newsletters, and regional regulatory data to quickly flag potential issues. In one recent example, EXL identified a provider request for a rheumatoid arthritis drug for which the patient was not eligible to receive based on FDA guidelines for that drug. As a result of this immediate red flag, the health plan was able to engage directly with the provider to resolve the issue and get the patient the care they needed.

This represents a seismic shift in the way insurers have traditionally communicated with providers around payment and coverage decisions. By immediately surfacing a potential contra-indicated therapy and making that information available to both clinical reviewers at the health plan and the providers who prescribed the treatment, the technology turns a potentially adversarial coverage denial into an opportunity for payers and providers to collaborate and find a better solution. Importantly, for patients, it helps to avoid a potentially dangerous treatment.

Domain Expertise is Critical When Applying AI to Complex Healthcare Operations

While EXL’s use of GenAI with health insurers has shown incredibly promising results, not all GenAI solutions were created equal. Critical variables for any users evaluating these solutions are whether or not they have been developed using specialized models that have been fine tuned to the unique needs of the health insurance domain, and whether or not they provide full transparency into every output.

Some GenAI solutions operate as “black box” technologies that simply spit out results and decisions without showing the data lineage behind those outputs. That kind of opacity is a problem in healthcare-related use cases where transparency is critical. Similarly, off-the-shelf large language models (LLMs) developed for general intelligence use cases often struggle to extract and analyze highly specialized health insurance data.

Used correctly, however, the right technology, paired with the right expertise in understanding clinical coverage decisions and payments is creating a breakthrough opportunity to improve communication between payers and providers, and improve outcomes for patients.