Applying AI Where It Matters Most in Pharmacy Benefit Management

Key Takeaway: AI creates the most value in pharmacy benefit management when it identifies clinical risk before a medication is dispensed, not after a claim is paid. The strongest approach combines AI’s ability to detect patterns at scale with pharmacist review and clinical judgment.

I have spent the past several years studying how artificial intelligence is evolving and what it actually means for pharmacy benefit management, including formal executive training in generative AI. More importantly, it has meant understanding where these tools are useful and where they’re not.

In many conversations, AI is framed through features, speed, or automation. But those conversations often stop short of a more important question: whether these tools are improving the decisions that ultimately affect patients and plans.

At PharmPix, we’ve approached AI differently. Rather than applying it broadly all at once, we started by asking where it could have the greatest impact. For us, that meant focusing first on the decisions that directly affect patient safety and plan performance.

Where AI Matters Most: The Point of Decision

Pharmacy decisions don’t all happen in one place. A prescription may start with one provider, be adjusted by another, and ultimately be filled without anyone having full visibility into the complete therapy. Each step is clinically valid on its own. But taken together, it can create gaps.

A report surfaces an issue. A trend line raises a question. A retrospective review explains what happened. By then, the decision has already been made.

We believed AI would be most valuable here, at the point where a decision is still in motion. Within our OneArk platform, AI continuously evaluates claims as they are processed, identifying patterns and risks that would be difficult to detect manually across large volumes of activity. When something does not align clinically, alerts are generated, whether for duplicate therapies, dosing issues, or combinations that introduce unnecessary risk.

These signals are identified before the medication is dispensed, allowing our clinical team to step in and work directly with pharmacists and prescribers. Each alert is reviewed by a pharmacist, ensuring that clinical judgment determines whether and how to act.

At PharmPix, that detection has demonstrated roughly 90% accuracy across drug-drug interactions, duplicate therapy, and dosing alerts. Quality is built into the decision itself, not layered on after the fact.

Extending AI Earlier: Strengthening Quality Before Implementation

As we saw the impact AI could have at the point of decision, we began applying that same approach earlier in the process.

One of the most practical applications of AI in our model is how we test and validate benefit design before it reaches a member. Within OneArk, we run thousands of test claims across different scenarios, simulating real-world conditions to ensure the benefit behaves as intended.

We are evaluating whether rules apply correctly, whether therapies are assessed appropriately, and whether scenarios could lead to unintended outcomes. AI allows us to test at scale, identify inconsistencies, and refine configurations before a decision is made.

This isn’t about replacing expertise. It’s about extending it. By using AI to process and validate at scale, our teams spend less time on manual review and more time refining strategy, supporting clients, and ensuring that benefit designs align with specific client and member needs.

Extending AI Beyond Clinical PBM Workflows: Improving How We Operate

As we’ve expanded our use of AI, the same principle continues to guide us: apply it where it strengthens execution and allows us to better serve our clients.

We now use AI to streamline operational work, including improving the consistency and efficiency of RFP responses and internal workflows. We’ve also begun incorporating AI into our programming environment, including tools like GitHub, allowing our development teams to work more efficiently and continue enhancing our platform.

The impact isn’t just speed. It’s our ability to stay focused on what matters most. By reducing time spent on administrative work, our teams are able to spend more time on what sits at the heart of our mission: supporting the clients who rely on us, and the members who rely on them.

The Role of AI and Clinical Judgment in Pharmacy

There’s a tendency to view AI as a replacement for human decision-making. That’s not how we use it.

AI is highly effective at identifying patterns. It can process large volumes of claims, surface anomalies, and prioritize where attention is needed, often with a high level of accuracy. But it never replaces clinical judgment or context. A flagged interaction may be appropriate depending on the member. A dosage that appears incorrect may reflect a deliberate decision by the provider. These aren’t errors. They’re decisions that require judgment.

That’s why our model is built around a partnership between technology and clinicians. AI brings forward the signal, and our clinical team determines what it means and whether action is needed. Used this way, AI does not distance us from the work. It allows us to focus more of our time and attention on the decisions and relationships that matter most.

What This Means for Employers and Brokers

When I think about the role of a PBM, I never start with pricing. I start with protection: Are we protecting the member from avoidable risk? Are we protecting the plan from unnecessary cost? Are we ensuring that each therapy is appropriate in the context of the whole member?

At its core, that requires attention. Attention to the full picture. Attention to how therapies interact. Attention to how decisions are made across a complex and fragmented system.

That’s where AI has made the biggest impact for us. It allows us to operate with a level of consistency and strategic focus that would not be possible otherwise, streamlining how we work so our teams can stay centered on the people and plans we’re responsible for protecting.

For employers and brokers, that distinction matters. Because two PBMs can look similar on paper, but deliver very different outcomes depending on how closely they are actually paying attention.

FAQs

How is AI used in pharmacy benefit management?

AI can evaluate large volumes of claims and benefit-design scenarios to identify patterns, inconsistencies and potential clinical risks that warrant closer review.

Can AI identify medication safety risks before a prescription is dispensed?

Yes. AI can help surface potential duplicate therapies, dosing concerns and drug interactions while a claim is being processed, when a clinical team can still assess what action is appropriate.

Does AI replace pharmacists in pharmacy benefit management?

No. AI can identify patterns at scale, but pharmacists provide the clinical judgment and member-specific context needed to determine whether a flagged issue requires action.