How to Overcome AI Roadblocks When Patient Data Is Scattered

Some experts argue that most health plans are still in the early stages of implementing artificial intelligence (AI). Not because technology isn't ready, they say, the real challenge is data. AI depends on consistent and reliable data, but healthcare is known to have patient data scattered across multiple systems.  

Kevin Deutsch, SVP of Health Plans at Softheon, a Software-as-a-Service (SaaS) company that facilitates health insurance enrollment, administration, and renewal, believes that for health plans to deliver real value with AI, they must first unify their data by prioritizing data governance and regulation. Recently, Healthcare Innovation spoke with Kevin Deutsch about best practices.

Are health plans ahead of providers right now in implementing AI around the member experience?

I don't think so. I think healthcare in general is very slow right now in AI adoption, but I don't think that the health plans are necessarily ahead. There's this talk of AI, but there's no strategy aligned with why we're talking about AI. There's a discrepancy between the strategy of what is trying to be solved, where it can be helpful, versus just needing to get into AI. Data architecture is the next biggest issue.

Where are the key areas of challenge right now in terms of members?

I think for members, they're demanding a more personalized experience. There are a lot of things that are happening right now in the industry. There's ICHRA, the Individual Coverage Health Reimbursement Arrangement, and this convergence of the employer market, which is 150 million individuals, and the individual exchange market. Essentially, these members are looking for a more personalized experience.

There is no one-size-fits-all for how you're handling members. There's a real opportunity in AI to deliver a more personalized experience to members. That's where I think the expectations are from the members' perspective. It's now up to the health plans and payers, how they're able to deliver that personalized experience utilizing something like AI.

What have you learned so far?

Where we're seeing this have the most impact right now is in customer service. We're seeing a lot more AI chatbots being developed and being presented to members, as opposed to the traditional enrollment experience. There are more guided experiences through things like chatbots. We have developed an internal LLM (Language Learning Model) with different models that is making our team more efficient in the work that we're doing. By becoming more efficient and getting to more consistent answers on a timely basis, we're able to better serve our customers and their members for different inquiries that they may have. They are getting access to real-time information, which is allowing them to provide a consistent experience and get to the more challenging issues quickly, because they're not dealing with some of those standard inquiries that come up.

That's the first application that I've seen as it relates to healthcare, which is the implementation of more chatbots and chat interfaces that could take care of a lot of those simpler tasks, so that the more complex things can be escalated to humans.

Are you involved in all three AI branches: Algorithmic, Generative, and Agentic?

We're involved in all of them. Softheon has been ahead of the agentic AI curve. Our entire system is built on thousands of unique microservices that are intended to do a specific task and do that task really well. When you think about the concept of Agentic AI, Softheon has been doing that for many years.

Where we've been able to pivot a bit is utilizing generative AI to speed up the activities that we're doing internally, something as simple as testing the platform, as opposed to humans going in and manually executing different test cases. Now we have AI that can generate test cases based on new features that have been developed. Now we're delivering higher-quality code, which then positively impacts our clients and the members who are utilizing the platform.

We've covered all aspects of AI, and I think we've been able to do that because of how much data we have access to from the data that we're processing for our members. Some people don't think of data as their own standard operating procedures and documentation within their organizations.

One area for the health plans to possibly focus on is forgetting the membership data for a second. What are the standard operating procedures and desktop manuals that you have? Have AI sit on top of it to help get to answers more quickly. That's one in particular that I think could be effective for health plans, and it's something that I believe is often overlooked.

How are you working with providing organizations on this data, if at all?

We host all the data in our private cloud, but we expose APIs for carriers and health insurance companies to access that data and store it on their own platforms. One of the challenges that's based on the payer side is that we're giving them access to all the data that we're capturing in our system, but all the other vendors that they work with, do they have the same access to that data so that it could sit in a centralized location and then have AI on top of it? We're providing that data via flat files, standard EDI-type transactions. We also have APIs that they can tap into to access the data at any point in time, serving any specific purpose.

Is this changing the interactions with providers?

I think it's becoming more and more real-time. We're seeing less communication via the standard EDI interfaces that I mentioned earlier, and everyone is requesting real-time interfaces that stream data. Healthcare has to be real-time, and every interaction should be real-time, because the moment that a file is sent, it's already outdated.

There are changes that are happening every second, from demographic data to financial payment data to eligibility information. If that's not happening in real time, I think we're all missing the mark, and then anything like AI on top of it is then ineffective because it's operating on outdated information.

How do you see plan members from an AI standpoint?

Our first wave of AI has been empowering our staff to be more effective in the work that they do, which is then powering the experience for the members. I would suggest any organization that's thinking about AI to start with their internal team. How do you make your internal team more effective in the job they're doing?

There's a lot of hesitation when you talk about automating individuals' work; am I still going to have a job? What our CEO says all the time is that AI is not replacing humans. AI is replacing humans who don't use AI. How do you take tools like AI to make yourself more effective, so that everything else that is more complex, you're able to focus on.

What can organizations do, considering that data is spread over different platforms?

Starting by taking an inventory of how many different integration points you have, and knowing where the data lives. The first and foremost step is taking an inventory. The second step is to ensure that critical information is stored in a structured, centralized location. Not everything is critical information. There are systems like ours that are processing certain sensitive data that the health insurance company doesn't need in its own platform. But if I'm a healthcare payer, I'm looking at that critical data set that defines my membership, that defines my service areas that I'm operating in, and make sure that I have access in real-time to that information, so that when the right AI opportunity comes in to solve a specific problem, I'm not reaching into all these different systems to try to figure out what I have.

Taking an inventory is the first step in making sure that all of the critical data that exists is absolutely always there in real-time. Then think about your strategy for AI.

Could you advise on practices for member acquisition and billing?

What I would say for member acquisition is, know who these individuals are. With AI, you can very easily segment populations more than you ever have before and deliver up personalized experiences to these individuals. Through AI, you can know a lot more about people than you did previously. There is no one-size-fits-all, so I think for member acquisition, it has to be personal. And with things like AI, there are certain ways to make that more personalized.

On the billing side, it's all about how consistent you can deliver an experience for somebody who's receiving a bill. They want to make sure that the bill is correct, the bill is timely, and then they can pay that bill. If any of those things are missed, there's a major challenge in billing. How do you put something like AI in place that's going to deliver a consistent experience and make sure that the balance that's being billed to the individual is appropriate each and every time.

Looking ahead, what do you foresee regarding AI implementation within healthcare?

The line is going to become more and more blurry on whether you're dealing with a human being or you're dealing with an AI robot agent. I don't think we're going to know in a few years. As a consumer, I don't think I'm going to care. As long as I'm getting the content that I need, it's not going to be as much of a challenge. We're going to very quickly start to see more personalization, and we're not going to know what the human is doing in the background, if at all, in these interactions.

I believe that there's this common misconception that I need to be doing AI. It's as broad a statement as that. Several steps should be taken before you talk about being in AI. Step one would be to determine where all the data is, whether it be membership data, demographic data, or financial data. Step two is determining what the processes and procedures would be. What problems am I trying to solve?

In summary: Stop talking about AI and focus on where your data is and what problem you are solving. AI is probably a great way to solve it going forward.