Nikhil Mendhi: All right. Good day, everyone. This is Nikhil, President and Chief Operating Officer of Exponential AI. Welcome back to our podcast, a podcast for Payer AI. Today we have our esteemed guest Deepan, who is an EVP of Solutions at a very well-known company, Firstsource, a leader in BPAS and BPO for Payers. Deepan, welcome to our podcast.
Deepan Vashi: Thanks for having me, Nikhil.
Nikhil: The topic for today is one of the hot topics that we have heard many times. It is claims AI. If you look at the history of payers, the two things that payers care the most about are paying claims and answering calls. Claims are the one transaction where all the healthcare dollars flow through.
That’s why we chose this topic to address. We received a popular demand to talk about how AI can be applied in claims. Deepan, my first question is, what’s different about AI in claims today? Because if you look at the history of the past 50 or 70 years, payers have been considering paying claims. The auto adjudication rate is almost top of the mind for all the payers. How is AI different in claims today?
Deepan: I would say claims must be processed timely and accurately. That’s the basic advice. If we were to look at it, if a claim were to be processed accurately for the first time, the net promoter score is plus 30. If we were to adjudicate the claim and it gets into an error loop, the net promoter score would go south, minus 25. Every time we miss pay and delay the payments, we invite calls from both members and providers.
Therefore, it’s very important that we deploy AI in claims operations and make it an AI-first, claims operation. Today, we continue to struggle to pay claims on time, resulting in interest payments, resulting in penalties, and that runs into millions of dollars for healthcare. We also spend a fortune on recovering overpayments, as you know. As you just mentioned about auto adjudication, it runs about 80% to 85%, and every percentage point improvement in auto adjudication then unlocks significant value. That’s where I feel AI needs to be deployed into claims.
You can even go from the perspective of Health Plans, where there are three major processes: code to card, card to care, care to pay. AI can fit into all three of them. The significant value that it can unlock is in the operations area, in claims to start with, and contact centers because these are the largest shops, the largest cost centers within the Health Plans. From that perspective, absolutely right, I think claims are the area where we can deploy.
Nikhil: Awesome. Great thoughts, Deepan. You actually pointed out a very important thing in your opinion. It’s not just administrative cost. When you think about applying AI, even though administrative costs are an area where we evaluate the investment in AI, there are other intrinsic values that are unavailable. One is the overall experience for the provider, which reflects in your net promoter score. Also, on the quality side, as we know, especially on the Medicare Medicaid side, there are significant bonuses tied to improvement in quality.
Deepan: Absolutely.
Nikhil: When executives are thinking about investing in AI, other than administrative costs, what other things should they consider in their ROI analysis?
Deepan: Most importantly, it’s not just about looking at value from claims operations. As you improve claims, the number of calls that you have to respond to also comes down. There is a downstream impact of it. The data quality is also critical as you look at these.
As you just mentioned, quality is important. It’s not just limited to this; deploying AI across Health Plans and making it AI-first health plan operations has three major areas that you could even look at. For example, it’s not just about reducing costs, but it’s about creating better financial, clinical outcomes. It’s about improving member and adding experience.
Today, Healthcare lags, and Health Plans lag in terms of providing experience to consumers, which is here to stay. If you look at the way retail operates, the way other experiences that we get from an airline or travel or other industries, healthcare significantly lags. That’s where AI will be a way to jumpstart into that aspect as well, Nikhil.
Nikhil: Awesome. That’s a really good point. That brings up a very important topic, the inherent nature of the healthcare business. You pointed out financial services. When you talk about financial services, it’s again a highly regulated industry like healthcare. When you look at the variability in the nature of the transactions that we handle, there are trillions of claims that fall through the system every day, but each claim is very different.
What do we need to explore, or what can AI offer to adapt to these transactions? That is one unique thing about healthcare: the entire dynamic nature of the business, which leads to the variability in every single call or every single claim that we get. How can AI stand up to this variability?
Deepan: There are many ways to eliminate inefficiencies. I would link to that as we have more variability, that is more inefficiency. There are ways to eliminate that. For example, significantly reduced handling time by identifying claims that are likely to append, they are likely to suggest, they’re likely to duplicate, and claims that might miss pay or result in an appeal. Those are the areas where we can reduce inefficiencies.
We can identify patterns in the claims that are likely to be delayed and then for provider green channel, that we can pay them accurately and avoid the penalties. For providers on the other side of the table, the revenue cycle depends completely on the way claims get paid by payers. There are linkages, and therefore variability equates to inefficiencies in that sense.
Nikhil: Good point, Deepan. That brings up one more thought. Here is how organizations can manage their own AI strategy to manage the variability, which means how can they manage their AI strategy. Their AI strategy would be front and center for their success with artificial intelligence and its application across the payer enterprise. If you look at the life cycle of a model, you train the model with historical data, build insights, and convert those insights into actions. As transactions and data regulation change, new CPT codes and procedure codes come in, the models need to be retrained. The front and center of this AI strategy is having a rigorous model management approach where organizations don’t have to spend a lot of money on talent, but on the technology that enables them to manage the models at the same time, the speed to market for any new solution. Any thoughts there, Deepan?
Deepan: Absolutely. Building and managing AI models is like building and managing the skills of a knowledge worker. Models are like AI agents. Therefore, deploying AI agents requires collaboration across both IT and business stakeholders, and it requires access to data. The AI-first approach also requires change management to be addressed because, how does this fit into the overall operation? Therefore, we must start small and look at how each solution can target a specific business outcome.
That’s the way you can deploy it and make it scalable. The AI-first model can start in one area and then quickly scale in different segments. Therefore, the platform that you select is also very important. The approach you select, the platform you select should be one that allows subject matter experts to play with the model. The model should be more like a citizen model. Everyone should be able to work with it, no longer a domain-only of data scientist working on some cool technologies. It should come in a way that subject matter experts in claims and context and data areas can play with it.
More and more, the platform that you select needs to be real-time. When you infuse AI in a way that it touches every transaction, infuse it in the transaction flow itself. The models that we build, the capabilities that we create should make these AI models in a way that they integrate with the claims engine itself. From that perspective, the AI capabilities will come from multiple different ways, from a health plan. Today, RPA vendors, click or claim vendors, claim administrative system vendors, system indicators. All of them bring AI capabilities.
What’s the right way to approach this? In my view, the way to approach this is to create models using your own data. Health Plans can create their own IP, and that would avoid bias. That would create explainability as well, making sure that in the long run, you maintain transparency. This is a probabilistic approach rather than a deterministic approach that we’ve used for ages to process claims. That’s the way one can scale and apply AI on every claim.
Nikhil: Great. You have some amazing thoughts, and I believe that we are seeing similar trends not only in healthcare but also in other industries. Some early adopters have put a lot of thought into this and have already started incorporating their ideas into their daily business processes and operations. Firstsource, for example, operates in the BPA BPO business, where they assume operational risk on behalf of their clients. How is AI transforming this?
How do you see the future of BPAs and BPOs evolving with the availability of more AI capabilities in the market?
Deepan: On one side, it enables creating a model for BPAs that allows for a PMPM transparent model for Health Plans, where we assume multiple capabilities and deliver them. AI allows us to significantly reduce PMPM. This is where the AI plus model creates significant value for Health Plans without having to worry about managing these models. These types of capabilities are in the middle segment, from 25 onwards in terms of Health Plans, and the top 15-20 Health Plans would prefer to keep these capabilities in-house. They want to integrate them using platforms they have developed over many years, which are difficult to abandon due to their millions of members, backend systems, and ecosystem. From this perspective, I think we take a slightly different approach when working with the top 10 Health Plans, where we deploy AI models in their environment using their data and managing their models.
Nikhil: That’s great. Any other parting thoughts for our payer clients who are considering implementing AI? What would be a good starting point?
Deepan: I think the approach requires integration with good admin platforms, and that’s critical. So the first solution that we deploy will test and ensure that integration is achieved. It’s important to select the right use cases to deliver tangible results, enabling you to scale quickly and deploy an AI-first approach. Change management is equally critical. You must choose a platform that enables subject matter experts, as I mentioned earlier so that even those without coding experience can work with these models. This will make it more successful and enable everyone in the enterprise to use it, rather than just a few people deploying models, leaving everyone else to figure out how to use it. It must be taken to a level where I can work with an AI model build.
Nikhil: Great. For our payer viewers, a platform-based approach, applying AI in real-time, and looking for early wins are the three things you should focus on to achieve the success of your AI-first strategy. Again, Deepan, thank you very much for sharing your thoughts and experiences with us. I look forward to more of these podcasts with you in the future. Thank you very much, Deepan.
Deepan: Thank you.
[music]
[00:14:32] [END OF AUDIO]