Pharma companies are deploying AI tools across a broad range of market access functions, from payer account strategy to contracting and beyond. According to our 2026 State of Patient Access research, 65% say that AI and RWD are ‘very important’ or ‘mission-critical’ to their access models today.
We sat down with two MMIT experts, Liz Rolla and Carolyn Zele, for a candid conversation about what AI can actually do for market access teams, how it’s changing the role of analysts, and why the data behind the AI matters as much as the AI itself.
There’s a lot of AI entering the market access space right now. What makes purpose-built AI different from the general-purpose AI tools pharma teams are already experimenting with?
Carolyn Zele, Solution Consulting Advisor: We’ve talked to many pharma companies that have tried to layer AI into their workflows by pulling data and feeding it into a general-purpose AI tool. The problem is that they don’t then trust the answers they get, because they aren’t grounded in verified, current market access data—which means they can’t act on them.
Think of Claude or ChatGPT as someone who’s not from the U.S. trying to make sense of what they’re seeing in the U.S. market. They don’t have the basis to understand the insurance market here, so they don’t know what they’re looking at. The word ‘covered’ or the word ‘preferred’ or the word ‘restricted’ might mean something completely different to ChatGPT than it does to a trained market access expert, who knows that restricted still means covered.
A purpose-built AI tool knows how to research, extract, and interpret across particular datasets. For example, our Ella AI has been trained on MMIT’s formulary and coverage intelligence, so when it answers a question about payer behavior or step therapy trends, it’s drawing on the same gold-standard data source pharma teams already rely on. A general chatbot can tell you what step therapy is, and maybe how prevalent it is in a particular TA. But Ella can tell you which payers are requiring it for your specific product, in your specific indication, right now.
Liz Rolla, VP of Product Management: I’ll add that it’s not just about whether the AI understands market access terminology. It’s also about whether the AI is driving a workflow. Think about it like the difference between embedded and bolted-on AI. I can go into any AI platform and feed it a deck, a text document, and a bunch of resources, and then have it try to make some sense of them. That’s bolted on. It means the AI is learning that source for the first time.
But embedded AI is built into the core architecture of a solution, which means it understands the layout, function and features of all the datasets. It’s been trained on these tools and this data, and knows how to research and extract the right elements. It’s a much more thorough analysis, and it points you in a direction.
Lots of times, organizations put AI in solutions just to say they have it. But the AI isn’t driving users to an actual decision or providing the information manufacturers need to take action. To me, the AI has to be intentional; it has to make the team more efficient and able to do their jobs better and faster. Otherwise, it’s just a feature that looks great in a demo, but it’s not really adding value.
“Speed to insight” has become a common phrase. What does it actually mean in practice for a market access team?
Carolyn Zele: If you’re trying to find an answer yourself—let’s say by going through 725 payer policies and trying to understand the different step requirements—it’s going to take you a very long time. Even if you’re putting the data in a pivot table, you still have to define everything and make sure you’re capturing what’s different and categorizing it correctly. When you ask a purpose-built tool like Ella that same question, AI does all that work in the background. The same steps happen, but at lightning speed, and you do not have do any of that work yourself. What you get back is the answer you needed in the beginning: where is the step problem?
Liz Rolla: Essentially, having a purpose-built AI tool is like having a team of analysts at your disposal. If you have a complex question, it will look across the entirety of the dataset to generate your answer. The person asking doesn’t have to pull reports or do pivot tables or Excel joins, because the agent is doing all of that work for you based on a question. It’s going to do all that data synthesis, look across that data, and put together a clear and comprehensive response. It’s like having an analyst at your fingertips, anytime.
So is the goal to replace the analyst? How do you think about the relationship between AI and access analysts?
Carolyn Zele: We never want to replace the analyst, because frankly, the boss shouldn’t be the one asking questions of AI directly. They won’t ever ask the right question or phrase it the right way. Think about what an analyst is actually supposed to do: consume and understand and make sense of huge amounts of data. They work slowly today because when you ask your analyst a question, they do everything manually.
But if the analyst is using AI in the background, two things happen. First, your analyst — who’s smart, knows how to ask the right questions, how to farm what they’re looking at — they’ll serve as the Quality Control for the AI. Did I ask the right question? Did I set it up correctly? And then they’ll synthesize responses much faster, which frees up their time to focus on strategy. AI isn’t really replacing the analyst layer; instead, it’s making analysts a hundred times more productive.
Liz Rolla: That’s a huge point. I’d add that what we’ve heard from clients is that analysts get a lot of these super basic questions — and even though they’re basic, they still take time to resolve. So for some of those block-and-tackle questions, purpose-built AI can get those answered quickly. That frees the analyst to focus on the more complex business questions that require AI plus human judgment. It shifts the analyst’s time toward the work that actually requires their expertise, not just harvesting data to answer the same foundational question for the fifth time that week.
Pharma companies already invest heavily in data. Why hasn’t that investment already translated into faster answers for market access teams?
Liz Rolla: The data problem in market access has never really been about access to data. It’s really about the distance between the data and the person who needs to act on it. We talked to dozens of clients during the development of Mosaic, which is our AI-powered analytics experience within the MMIT Platform. We kept hearing the same three issues pop up.
First, everything took too long to do. Slow load times interrupted their workflow, and uncovering key insights required lots of clicks on lots of screens. Secondly, their work processes were too manual. You might have formulary data in one place, PAR data somewhere else, competitive trend data in another tool, plus a separate BI platform. Stitching things together across multiple tools and solutions was time-consuming. Those two issues result in the third and most pervasive complaint: by the time clients had the insights they needed to move forward, the window to act had already closed.
Carolyn Zele: One client actually described their current tools as “brutal.” That word stuck with all of us throughout the whole Mosaic development process. Nobody wants to spend more money on data to get a slower answer!
So how did you figure out what market access teams actually needed when building Mosaic?
Liz Rolla: Before we began development, we asked several clients to explain what they lacked, what would help them drive their business. What was surprising for me was how much clients actually need this technology. We all know AI is important, but there’s a lot of hype out there. To actually hear from clients that they need AI to accelerate their work, and that they’re getting expectations from their leadership that they should be using it, was a turning point for me.
Carolyn Zele: I’d add that many of our teams, like our PAR Insights teams, are constantly answering client questions. When we looked back at the work they’ve done for clients over the last five years, we got a good picture of what clients actually ask us. This helped us understand the types of questions that Ella AI needed to be able to answer. In a way, we’ve been functioning as our own AI long before we had AI. Once Mosaic was in beta, we saw it spread like wildfire as people shared it with more and more of their colleagues. That really validated that we were on the right track, because once people started using it, they didn’t want to stop.
There’s a real trust problem with AI in pharma right now. How do you think about that in the context of market access decisions?
Liz Rolla: The tolerance for error in our client base has always been zero. If clients ever think our data is wrong, they will not hesitate to tell us. Mosaic doesn’t change that expectation. We think about it as a continuous commitment to our customers to provide trustworthy information. We spent a lot of time testing Mosaic with Ella AI before releasing it, and the accuracy bar is high — as it should be.
With Mosaic, market access teams are able to act more cohesively, because everyone has access to the same information in real time to make better decisions. The insights derived from your data are no longer siloed within a small number of individuals, because everyone can access and interpret that data now, from your leadership on down.
Carolyn Zele: I could separate AI users into two groups. The first is those people who’ve been doing market access research for many years, looking at policies and restrictions, and they are legitimate subject matter experts (SMEs). The other category are people who are new to market access. And if those folks don’t work with the SMEs to fully understand their question and phrase it correctly, they’re going to get an answer back from AI that may not be complete. It’ll be a partial answer because they asked a partial question…and that’s when people start to distrust AI.
So is there a governance risk? Yes. But using Ella AI presents a lot smaller governance risk than using ChatGPT, or Claude, or any other AI platform built by people who are not SMEs in market access. They might be SMEs in AI, but they’re not SMEs in market access, and that’s the crucial element.
Where does this go from here? What does “good AI” for market access look like in the next few years?
Carolyn Zele: Good AI in market access is AI that keeps learning. Right now, Ella can tell you which payers require your drug to step through a competitor before coverage, and she can tell you when United made a step change on a comparable brand, and which payers followed within a given window. Where this goes next is Ella learning how to ask back, to clarify your initial question. To say, “I want to make sure I understand what you mean by advantaged in this context before I give you an answer.” That feedback loop is what makes AI go from useful to indispensable.
Liz Rolla: Good AI in 2026 and beyond is AI that’s grounded in verified data, purpose-built for the domain, and continuously improving. The industry is getting better at distinguishing between “we have AI” and “our AI actually does something.” For market access, that something has to be faster answers on data you already trust. The teams that do the best are going to be the ones that move from reactive to anticipatory, using AI to spot what’s coming while they still have time to get in front of it and act. That’s the shift we’re building toward.
Learn how Mosaic provides the clarity and speed market access teams need to analyze, anticipate, and validate payer behavior
