Transforming Program Integrity with Agentic AI

Autonomous agents aren’t anything new. Every day, our economy is supported and driven by software systems purchasing items, settling transactions, and sending marketing messages. For decades, most Wall Street trading has been driven by automated software. Just as we’re getting used to GenAI, the era of agentic AI has descended upon us.

What Are AI Agents or Agentic AI?

The industry hasn’t agreed on a standard agentic AI definition. If GenAI is used for generating text, images, and videos, agentic AI combines GenAI technology with the ability to make decisions. According to professors at MIT’s Sloan School of Business, AI agents are, “autonomous software systems that perceive, reason, and act in digital environments to achieve goals on behalf of human principals with capabilities for tool use, economic transactions, and social interaction.” Practically, this means AI agents are software systems that may use GenAI, machine learning models, and other software to interact with other agents and people.

In our industry, an AI agent may use an API to retrieve medical claim details and feed that information into a GenAI tool, like TENEX, to summarize the claim so it’s fed into a report for review. The AI agent operates autonomously for the retrieval and summarization leaving the decision to a PI professional.

Human Driven, AI-augmented Decision Making

That idea is a powerful one for the PI professional. In a world of connected MMIS systems, things like prepay claims analysis, provider enrollment, and post-pay adjudication can all be turbocharged beyond rules-based, monolithic systems into modular agentified tasks. In a world of agents communicating with humans and other agents, program integrity can finally focus on actual integrity and less on the operations of the systems in use.

In the Medicaid arena, every state is 1-of-1. Each state’s rules are unique to the culture and needs of its citizens. Each state has its own set of rules governing how providers and beneficiaries are enrolled and how benefits are administered. Each state also operates with the core fundamentals of accuracy, reliability, accountability, and trust in their Medicaid enterprise. For a long time, technical complexity around rules and modeling hindered the ability to apply these core principles. Agentic AI systems provide a way for PI professionals to express each states’ rules as they see them instead of having to translate compliance into computer code.

For a long time, Medicaid enterprises and vendors focused on the easiest parts of PI technology: big data wrangling, interfacing systems, and report generation. Accuracy has remained elusive. Complex policies and rules were implemented in rules engines which are hard to translate into programming languages. This resulted in sophisticated (and buggy) if-else and switches allowing fraudulent claims to squeak by.

Agentic AI Spurs the Integrity in Program Integrity

Agentic AI is an integrity enabler. Agents can understand rules written in English (hopefully plain English).  Finally, computers can be used for what they’re good at: handling massive amounts of data, executing rules, and generating results. In-the-loop professionals can finally get to do what they’re good at: spot recognition, creativity, and final decision making when presented with accurate information.

We’re witnessing the beginning of the next transformation of program integrity. In our next RIViR Reads, we’ll take the concept further and explore our part, the PI professional’s piece of the puzzle.