Penguin Ai, a startup that has constructed an AI-based platform designed to streamline administrative inefficiencies in healthcare, has pulled in $29.7 million in enterprise funding to pay for product growth and to scale up deployments with payers and suppliers. Fawad Butt, founder and CEO of Penguin Ai and former chief knowledge officer of UnitedHealthcare, Kaiser Permanente and Optum, not too long ago spoke with Healthcare Innovation concerning the transition going down to the brand new world of agentic AI.
The Palo Alto, Calif.-based firm says its flagship platform combines task-specific small language fashions (SLMs), digital employees and brokers, with a healthcare-specific AI platform to streamline processes corresponding to prior authorizations, claims processing, medical data summarization, and appeals administration.
Healthcare Innovation: May you discuss a little bit bit about your background as a chief knowledge officer and the way it helped you type the thought for this firm?
Butt: Earlier than beginning Penguin, I spent a little bit little bit of time within the VC world at Canvas Ventures, being an working associate, and earlier than that I used to be the chief knowledge officer at Optum, answerable for all knowledge and analytics capabilities throughout all three strains of enterprise at Optum. Earlier than that, I used to be on the insurance coverage facet of the home at United Healthcare because the chief knowledge and analytics officer. Previous to that, I used to be in the identical function at Kaiser Permanente. So when you have got that distinctive vantage level, you see patterns. One of many patterns I noticed was that the information may be very disorganized. That’s why they created roles for for individuals like me.
The info is disorganized as a result of over the course of a number of improvements, we purchased best-of-breed options, and unexpectedly we have a spaghetti mess of 10,000 programs making an attempt to push knowledge forwards and backwards, in order that was at all times a difficult factor.
Additionally, after I was in these positions, I might purchase any platform within the business, as a result of we had the budgets to do it. I might go to AWS, Google Cloud or Azure, however the issue was that these are horizontal, generic platforms. They did not perceive healthcare. If you happen to purchased that expertise, you needed to assemble a workforce and determine learn how to use these generic instruments and apply them towards healthcare issues. So to me, the thesis was at all times that healthcare deserves its personal platforms, proper? And why aren’t there extra of these? And if I ever had the chance, I used to be going to construct one.
HCI: Isn’t it additionally a difficulty that historically the payers had the claims knowledge and the suppliers had the EHR knowledge, they usually did not actually need to share?
Butt: The fact is that for our well being system to function, that knowledge does get shared. A supplier has to supply medical data should you’re submitting a declare or a previous authorization, in order that knowledge does get shared. The foundational argument that I’ve heard all through my profession is that the incentives are misaligned. I agree with that, however the actuality is that the incentives are misaligned, and we share the information nonetheless.
The larger problem I see is that quite a lot of these firms have grown via acquisitions — United Healthcare, for instance. Each time you buy a plan, you get 13 core programs that include it. If you happen to try this 20 instances, unexpectedly you have got spaghetti programs doing too many issues.
The fact is that there hasn’t been a expertise evolution, till not too long ago, that would take that spaghetti mess and in reasonably priced method give you a brand new structure that might be deployed. However now the world goes agentic for essentially the most half, which implies the codecs do not matter anymore. Brokers can just about perceive and eat each structured and unstructured knowledge and context round that.
So the following model of healthcare is agentic healthcare. If we imagine that premise, and you are a CIO or a CEO at a hospital system or a payer, you may say, I’ve to construct 1,000 brokers, proper? You want a platform to have the ability to try this on. You could possibly use the Azure platform or AWS and attempt to train it healthcare. Or you may come to Penguin and discuss to us, as a result of we constructed the elements which can be required for an enterprise to undertake a platform. Now we have the connectors to your knowledge, whether or not it is in Epic or Cerner or athena or Oracle or Information Bricks or Snowflake — you title it. Now we have that in-built. Now we have a privateness engine in-built that de-identifies that data. Now we have a bias mitigation and scoring algorithm that checks to see should you’re introducing bias into your fashions. We constructed our personal small language fashions for prior auth, danger adjustment, and claims adjudication, after which we provide you with our brokers out of the field. That is what a platform is meant to do. It’s supposed to present you what you want so you will get to ROI in 90 to 120 days.
HCI: So are you envisioning each the well being programs and payers as potential clients?
Butt: 100%. As a result of, to me, it is a false dichotomy in some ways. Issues begin on the payer facet and finish on the supplier facet, or begin on the supplier facet and finish on the payer facet. They simply name them various things, proper? One could be referred to as declare scrubbing and one could be referred to as claims adjudication, nevertheless it’s basically a really related processes. So why not construct it as soon as, after which let’s convey the information and the intelligence collectively in a single place, so it isn’t payer towards supplier, it’s payer and supplier.
HCI: I used to be speaking to a well being system CIO who mentioned they had been beginning to construct brokers in home, and the payers are constructing their brokers. And it is nearly like it will turn into this battle between the 2 to see whose are higher. And she or he mentioned that agent warfare shouldn’t be the place they need to be.
Butt: No, however that warfare has began. The agent wars are right here, proper? It isn’t this futuristic factor that is going to occur. It is occurring right this moment. I sat with the CEO of one of many largest regional well being plans within the nation. He mentioned what they’re seeing is that, in some methods, the suppliers have adopted brokers quite a bit faster than the payer facet, as a result of the payers’ processes are extra advanced. In a single situation, he mentioned, a small community of suppliers that used to do 5% appeals on denials is now doing 100% appeals on each denial the well being plan is sending them. He believes the supplier group has an agent on their facet, and the well being plan has eight individuals on its facet. So how are they going to win that?
HCI: Nicely, as a part of your pitch to those execs, have they got to ditch quite a lot of this funding they’ve in all these costly legacy instruments that they’ve already made a wager on?
Butt: No, if in case you have a beginning place that claims, let’s name that sunk price and begin afresh, you have not been in healthcare lengthy sufficient, proper? That is simply not how it will work. The funding when it comes to expertise could be legacy, however when it comes to data, it’s gold. All the enterprise’ processes are captured in knowledge all throughout these enterprises. So in case you are envisioning an agentic world, then you do not have a look at it as previous expertise or new expertise. You have a look at it as helpful knowledge or not helpful knowledge, proper? And I can guarantee you that there’s some very helpful knowledge in these legacy programs that’s going to counterpoint the intelligence. If you happen to’re constructing small language fashions or refining giant language fashions, that data goes to be essential. We have been listening to that knowledge is the brand new oil for a very long time, however the sincere fact is that now’s the time the place, should you can construct the suitable refinery, you may flip that oil into gas to your group.
HCI: Does generative AI have potential to alleviate quite a lot of the issues that individuals describe about knowledge requirements or semantic interoperability?
Butt: I feel it does. It is already taking part in a significant function. It was that the healthcare knowledge world operated in buildings or in knowledge fashions. So I’ve an information mannequin and you’ve got an information mannequin. If I must eat your knowledge or share my knowledge with you, our knowledge fashions must be aligned by constructing a bridge within the center which says your x means my y. So that you spent quite a lot of time and vitality constructing these bridges as a result of that requires material specialists to weigh in, and people are laborious to seek out and costly to retain. However now as a result of we’re not making an attempt to normalize knowledge earlier than we are able to apply intelligence to it, we are able to eat that data in each codecs and have the ability to analyze it in very related methods and derive insights in very related methods, in order that step that used to take 18 to 24 months of simply aligning the information…… I am not going to say ETL [extract, transform, and load] is lifeless, however it’s much less related within the agentic world.
HCI: UPMC Enterprises is one in every of your buyers. That looks like a corporation that might be an apparent potential buyer.
Butt: We hope so. We can not touch upon that. They have been a incredible associate for us. They had been one of many early ones to to understand what it’s that we had been doing, they usually’ve been supportive persistently all through the method. I might say in our earliest days, the establishment that acknowledged what we had been doing was UPMC.
HCI: Is a part of what buyers the workforce you’ve been in a position to construct when it comes to execs with quite a lot of expertise in healthcare?
Butt: Sure, three of the technical leaders — head of AI, head of supply, and head of engineering — are ex-Optum individuals. Most of the workforce that now we have both labored at Kaiser, United, Optum, CVS and others on the technical facet. Then now we have our chief technique officer, Mark Caron, who was the CIO at three completely different Blues plans and a CTO at Catholic Well being Initiatives, in addition to the CIO of the Collaborative Care Division at Optum.
And Missy Krasner, who’s been an advisor for us from the start. She’s been at Amazon, Google Well being, and Redesign Well being. I feel that exhibits within the completeness of the imaginative and prescient and and it exhibits within the understanding of the nuance on governance and safety and privateness that’s specific to healthcare.
HCI: Are there some issues on the coverage and regulatory facet, or involving market pressures, which can be providing you with tailwinds to get extra engagement?
Butt: I feel there are two large tailwinds. One is that payers have been decimated over the course of the final 18 months. The Medicare Benefit enterprise that was extremely worthwhile has turned and their medical loss ratios have gone via the roof. So there’s large strain on them to scale back prices. And the price that they will management shouldn’t be the MLR, it is the working price of operating the enterprise.
Then CMS has some rules round interoperability and particularly round prior authorization that persons are making an attempt to recalibrate round. With an agentic world turning into actuality, lots of people are selecting brokers to try this automation vs. one other utility that they used to do. So we’re seeing an amazing quantity of demand and numerous engagements on that use case on the payer facet. Danger adjustment is one other use case on the payer facet that’s fairly strong for us. Claims is creating as one other space of of demand. On the supplier facet, I might say it is round income cycle administration — the claims scrubbing, the extraction of medical codes. Medical coding itself is a giant use case for us. After which appeals is the opposite one. We will be launching some extra capabilities that we’re piloting round sufferers and suppliers that might be coming down quickly.
HCI: Do some agentic effectivity features to well being programs or payers come from lowering administrative worker head counts? Or are they seen as instruments for these workers to make use of?
Butt: I do not see any alternative of head rely anytime quickly, frankly, and it is as a result of the system is capacity-constrained proper now. The nurses or docs or admin staffers are being requested to do increasingly more. An instance is prior authorizations. One of many large payers informed us their pharmacists used to must evaluate 40 circumstances a day for prior authorization. Then that went as much as 70 circumstances a day. Now they’re asking their pharmacists to do 100-plus case evaluations a day. And guess what’s occurring? They’re burning out. They’re making errors. They’re quitting and leaving the sector. I’ve a private story round this, as a result of my spouse was a previous authorization pharmacist, and I noticed her go from working eight hours a day to 14 hours a day making an attempt to make quota. So I feel it is a capability concern. It’s going to simply relieve the individuals who have the accountability to do that work to concentrate on crucial a part of the work and work on the high of their license.
