Article Summary: Penstock is using AI and machine learning to accelerate how new payment integrity concepts are discovered, validated, and deployed. In this Q&A;, President Steve Palma explains how AI identifies subtle trends in claims data, detects nuanced anomalies, enhances concept precision, and strengthens prevention strategies. Learn how Penstock blends AI-driven analysis with human clinical expertise—and why agentic AI is opening new possibilities for data quality, audit support, and end-to-end payment accuracy.

‍

For more on Steve Palma’s leadership perspective, see our spotlight interview

‍

1. What role is AI playing in Penstock’s current payment integrity strategy?

Our team is deeply focused on advancing how we discover and validate new audit concepts—and AI is playing a powerful role in that evolution. Penstock has always been strong at surfacing high-impact concepts through clinically driven audits and complex chart reviews. Now, with AI and machine learning, we’re scaling that capability to uncover patterns and payment anomalies that might otherwise go undetected. We’re using data mining to examine large volumes of structured claims data—looking for subtle shifts in billing behavior, emerging trends, or leakage that may not surface through traditional rules engines. These capabilities help us move beyond detecting known issues. Instead, we’re letting the data guide new hypotheses and smarter interventions.

‍

2. What does that look like in practice—from signal to audit concept?

The process typically begins with a signal—maybe a small deviation in code usage or a payment spike that doesn’t align with historical trends. From there, our team runs it through a series of model iterations, logic tests, and validation loops. The goal is to determine whether it’s a one-off anomaly or something more systemic, using our structured concept development process to guide decisions. For example, a seemingly minor shift in modifier usage across a narrow provider set can become a meaningful audit concept when paired with historical trends and clinical logic. That’s where AI accelerates our ability to refine and test ideas with speed and precision.

‍

3. How is AI helping Penstock detect more nuanced issues?

While large language models are well suited for unstructured data, structured claims data typically requires more traditional machine learning techniques. Our team is using those tools to detect everything from outlier claims—such as a $25,000 payment that should have been $250—to broader patterns like fee schedule misapplications that quietly inflate reimbursement. These are exactly the kinds of issues that, if not caught early, can cost plans millions. AI doesn’t replace the integrity process—it sharpens and scales it.

‍

4. What’s the impact for health plans?

The impact is measurable. With AI-enhanced discovery, we’re able to deliver concepts that are more precise, clinically grounded, and resilient against provider gaming. That means better savings, fewer unnecessary disputes, and stronger compliance alignment. It’s not just about catching more—it’s about catching what matters.

‍

5. How is Penstock balancing AI automation with expert judgment?

Our view is simple: AI is a force multiplier, not a substitute. The models surface the patterns, but our clinicians, coders, and analysts bring the experience and context needed to interpret them. That human judgment remains essential—especially when the goal is not just recovery, but credibility with providers and auditors alike.

‍

6. Are there any AI developments you’re particularly excited about?

One of the most promising areas we’re exploring is agentic AI—systems that can reason, act, and collaborate with a level of autonomy. For example, we’re piloting use cases where agentic AI can assess the quality of incoming data feeds before they enter analytic systems—flagging issues like missing provider IDs or sudden shifts in field population. We’re also exploring how these agents could support auditors by evaluating coding logic, referencing policy rules, and suggesting audit flags in real time. This kind of intelligent collaboration between agents, coders, and analysts isn’t hypothetical. It’s a direction we’re actively building toward.

‍

7. Final thoughts for health plans evaluating AI-driven partners?

There’s a lot of hype around AI—and not all of it is helpful. The question health plans should be asking is: Does this improve how we identify, validate, and act on issues that impact our bottom line? For our team, AI is not about replacing people or chasing trends. It’s about applying the right tools to recover more, reduce abrasion, and operate smarter across the payment integrity lifecycle. The goal is sharper insights, faster execution, and ultimately better outcomes for both plans and providers.

‍