The impact of AI on today's pharmacy benefits

Scott Draeger: Santiago, there's just so much discussion about AI right now. To help frame this discussion, can you talk about what we mean when we use the term AI at Optum Rx and also give us an overview of the different types of AI?

Santiago Abraham: Happy to do that, Scott. Artificial intelligence is really revolutionizing the way we work. It is both enhancing experiences, boosting the performance of our teams and driving efficiencies. And as I think about artificial intelligence and machine learning (AI and ML), there are three broad categories that we think of. 

The first category is predictive AI, which is driven by pretty sophisticated models to do things like helping us forecast trends and identifying key focus areas in areas like fraud, waste and abuse, for example. 

The second I would classify as generative AI, which includes things like ChatGPT and other large language models that excel at handling large sets of unstructured data, making them invaluable in a lot of different domains. Whether that's streamlining administrative tasks or powering personalized conversational chat bots and voice or even serving as knowledge assistants for us. Just a lot of untapped potential there that we're just getting at.  

Finally, I will add a new category that's more nascent, which is called agentic AI. That's an area where we are taking complex tasks and starting to break them up into sets of discrete problems leveraging multiple bots and multiple agents functioning almost like a personal assistant. 

So those are three major categories of AI that we’re focused on here at Optum Rx.

Scott: Yeah, I think to many people AI does seem like a new technology, but this has actually been a long time coming. Santiago, can you walk us through you and your team's journey with AI in recent years?

Santiago: Absolutely. To your point, AI has been in motion for a number of years. We began our journey as far back as 2019, with the development of what we called machine learning predictive models. Those included things like identifying opportunities for adherence or non-adherence for chronic conditions like diabetes, depression, hypertension, etc. Also, things like pharmacy fraud models. 

In addition to those, we've identified opportunities to really understand how we can improve quality during dispensing. This includes AI pill validation that was put in place with visual processing at the dispensing fulfillment lines. It looks for quality by making sure that the pill vials are dispensed correctly at the point of fulfillment. That’s already been rolled out to all six of our dispensing sites and being used today more than 5,000,000 times each month.  

And most recently we've established what we call Innovation Councils that are focused on making sure that we're identifying those opportunities and use cases as it's such a diverse emerging environment. We are really trying to drive the benefits of AI through all of our major opportunities, whether they be experience or efficiency opportunities into all of our strategic areas.

Scott: Sid, you know, while preparing for this discussion, it struck me that the last time we spoke, you specifically mentioned how you can't become too enamored with the shiny new object when creating new solutions. How do you avoid that temptation to use AI with everything and make sure you know our use of AI is truly impactful?

Sid Sahni: Great question, Scott. So, we apply the same discipline to AI that we apply to every new opportunity that we evaluate. It really starts with the use case. 

• What is the problem we are trying to solve? 
• What is the benefit that we are trying to derive?  
• And once you can solve for that, you can then ask, where can AI assist? 
• How could AI do this better? 
• How could AI turbocharge what it brings? 

So, we have applied that same rigor to bring AI into various disciplines. It’s helped us not only improve the existing solutions but also made us think about potential improvements that we might not have otherwise even considered. Let me give you a couple of examples.

For one, there are things that we do on a regular basis and have done for a long time such as preventing fraud, waste and abuse in the system. We’ve been able to apply AI to that, and this year we’ll recover almost $100 million more as a result of the increased intelligence that AI was able to afford us. 

Then we think about the areas of member experience. We've used AI to be able to predict when a patient or a member will have a problem and then proactively go out and improve that situation or be able to mitigate it. It will be completely invisible to the members that there was even potential for something to go wrong. 

We've also used AI to better serve our members in our call centers, to predict what the member is calling about. We use ambient listening to actually understand and document it, and then at the end we use generative AI to create a summary of the call for the individual to review. This drives satisfaction and improves their experiences. 

We then want to go beyond that. These are things that have been improved by AI, but what else could we do?  Where could we go next and what new solutions can we create? Those are the areas where we're bringing the same discipline and rigor.

Scott: You know, many of the common concerns we hear about AI center on privacy and security. Santiago, I think you're the perfect person to answer this question. Can you speak about the practices we have around data privacy and security?

Santiago: You're absolutely right. This is an area that I'm keenly focused on. 

Healthcare data is highly sensitive and making sure that we provide the right safeguards to ensure that we are ultimately compliant with all elements of data security is priority one. I would say further that there are additional elements and guardrails around AI itself, and for us that centers around governance models making sure that we're going through the appropriate review cycles. 

Within UnitedHealth Group we have stood up an AI review board that is charged with looking at elements of:

• Fairness
• Accountability
• Oversight
• Transparency
• Privacy
• Security

The AI Review Board goes through all of these considerations to ensure that any model that we put in place is appropriately governed and appropriately approved to safeguard the data that it leverages. So, from a AI data privacy and security perspective, it’s making sure that we're staying compliant with applicable laws, regulations and client contracts. All are key elements to where we're going.  

And then finally, I'll add that the element of human oversight is important as well. We call it “human in the middle,” which is making sure that the models are supported by the appropriate guardrails to ensure that decisions are ultimately made by the appropriate experts in any given area.  So, while the AI may support it, the AI is not making the decisions that we need to make and are ultimately accountable for.

Scott: Santiago, can you build on that a little bit? Can you talk a little bit about how we interface with external partners?

Santiago: You know, that's a great question, Scott. I will say there is a great opportunity right now as there are a lot of vendors in the marketplace coming up with really, really innovative solutions around AI. 

For us, it is a “buy versus build” question. I would say we're in what I call a mixed model. So, where appropriate, we're leveraging industry frameworks, but ultimately crafting internal solutions. These are areas where we believe there’s an opportunity to differentiate ourselves. We have a lot of really talented developers and data scientists who are supporting this work and who are leveraging all the AI disciplines around this. But in addition to that there are partner opportunities.  

When we are leveraging partners, again, we’re following the same guardrails we talked about earlier. This means going through the AI review board and making sure that we're doing the appropriate things to safeguard data with those partners to build out the best experiences for our clients and for our members and patients. 

And this is an area where I would say we've done a lot of assessments and reviews and we've evolved quite a bit in this space. We are super happy to help clients with this as well.

Scott: Sid, as I've listened to what both you and Santiago have had to say, I think one question our clients are bound to ask us is how we can use AI to help them with the challenges that they see around pharmacy benefits. Can you think of any specific use cases where you think we can have an immediate impact?

Sid: Scott, one of the primary responsibilities we have as a PBM is to help our clients predict, understand and then ultimately manage drug spend. We think AI has a significant impact on that. We are thinking through:

• How could AI predict client-by-client trend more accurately?
• How could it be more precise?
• How could it be more unique and tailored to the unique situation and the unique population of a client?
• Then, what solutions could emerge from that?

These insights can be quite powerful. We've seen AI go through the journey to now get us to a place where we could actually develop new solutions for our clients. And so, we're really excited about leveraging AI in this core responsibility that we have as a PBM.

Scott: Santiago, in addition to the examples Sid gave, are there other solutions and developments that that really excite you?

Santiago: Yeah, absolutely, Scott.  I’ll point back to my original comments around the consumer. I really get excited about improving their experience with AI. I'll give you two examples. 

One went live very, very recently and we’re still in learning mode. It addresses the problem where about one third of the time, we don't have clarity on member intent when they call us. And so, one of the things that we’ve been looking at is how to better guide our members as they are navigating the system. If we’re not sure what they're looking for, they may land in a place where they still need to be guided to the right place. This could involve a call transfer, for example. Leveraging AI, we are on a journey toward solving this.  

We just launched the tool this past month and already have been able to get about 50% of unclear calls landing in the right zone to be guided to the right agent who can help them address their needs. So that's super exciting and we're just getting started. The model is actually learning as we speak. And we've already seen about 10% improvement even from the start of the journey. 

Another example that’s coming soon is in the digital realm. It’s the same challenge, really understanding where do I go to find a lower cost medicine option, or if I need help with the refill. The ability to do that is coming with what we call an Optum Companion capability. That’s going to allow you to state your need in plain speech, and we'll bring to you the experience that will support that need in a digital form. We're super excited about this, and this is something that is coming later this year.

So, it’s about helping you navigate that complexity in a way that is appropriate for you. These are just a few examples of how AI is really helping to improve how we're navigating the consumer through their needs. And we're super jazzed to bring it to life here at Optum.

Scott: It's really, really fascinating. Sid, final thoughts?  What does the future of artificial intelligence look like from your perspective?

Sid: So, to say that we're excited about artificial intelligence would be an understatement. The potential it holds has been compared to the dawn of the industrial revolution, where machines were able to do more physically than humans could. 

Could AI represent that level of generational change in our industry and our society at large?  Of course, we need to think about this very responsibly. We need to be aware of the risks. We need to make sure that we're doing this in a manner that we understand, especially in an area like healthcare, which we know is personal and very, very meaningful to people. 

But at the same time, we want to be excited about what the future brings and all the potential that AI can deliver for us. I don't think we fully comprehend it yet. This space is moving fast and we are moving with it. But as we go through, we will uncover things and learn things and think of things that are beyond our comprehension at this point. 

In summary, we are incredibly stoked about what AI can do, what AI could do, and also at the same time stoked about doing it in a manner that is responsible and careful. So, I'm incredibly excited about this, as is all of Optum Rx.