How actuarial AI could tell physician groups where a value-based contract stands midyear

The 476 accountable care organizations (ACOs) in the Medicare Shared Savings Program in 2025 won't learn until November whether they earned shared savings or owe losses for that year. The Centers for Medicare & Medicaid Services (CMS) delayed 2025 financial reconciliation until after it finalizes the 2027 Medicare Physician Fee Schedule, so it can apply a proposed retroactive guardrail to a projected spending trend it uses to update ACO benchmarks.

CMS also reopened 2024 reconciliation for ACOs whose agreements began that year, replacing results first published in September 2025 with updated figures in July.

As of January 2026, CMS estimated that 14.3 million Medicare beneficiaries received care through ACOs, up from 13.7 million a year earlier, and the Shared Savings Program grew to 511 ACOs. The CMS Innovation Center's 10-year Long-term Enhanced ACO Design (LEAD) Model replaces the ACO REACH Model on Jan. 1, 2027.

Groups in value-based contracts often don't know what they owe until months after a performance year closes, and sometimes years later, said Najib Jai, M.D., MBA, who worked on the provider side at Oak Street Health and at Conduce Health, the company he co-founded and led as CEO through its acquisition.

Jai began his career in private equity while earning his medical degree and MBA at the University of Chicago. In October 2025, he became chief operating officer of Arbital Health, a San Francisco company that provides software and actuarial services to providers and payers in risk-based contracts, where he oversees operations and implementation. The company's artificial intelligence (AI) assistant, Arbital AI, answers plain-language questions about contract performance, runs multistep analyses and saves results to shared workspaces. Arbital says its platform has completed more than 1,500 contract reconciliations.

Jai spoke with Medical Economics about why payers and providers end up disputing settlement numbers, how Arbital tracks a contract while the performance year is still underway and what it takes for a group to get started.

Our conversation, lightly edited for length and clarity, follows.

Tell me about Arbital. What does the company do, and what is actuarial AI?

Najib Jai: Taking a step back, I think it's important to recognize that health care in the United States is incredibly challenged. As a developed nation, we are spending copious amounts on health care, and unfortunately, people are not materially healthier as a byproduct of it, or at least not healthier relative to other developed countries.

One of the solutions that's been posed is this concept of value-based care. How do we compensate doctors and health care organizations for actually improving patient outcomes, not simply the volume of patients they see? That sounds great in concept. It turns out, contractually, it's incredibly complicated, and navigating the complexities of those contracts has been a major pain point in the value-based care ecosystem. That's where actuaries have historically been very helpful: measurement, understanding, analytics and everything in between. So what we're solving here at Arbital is really building the appropriate infrastructure for payer and provider contracts, with an emphasis on value-based care.

Now, actuarial AI is the buzziest of buzzwords within the Arbital ecosystem. But the way I think about it, actuarial AI is ultimately democratizing access to actuarial expertise, and what that looks like varies depending on the persona in question. For an actuary, as an example, I would think about it as an augmentative tool. How do you enable that actuary to be that much more efficient in their work and have access to additional benchmarks, data querying information and so on, to be that much more accurate in the insights they're providing?

For a layperson like myself, who has zero actuarial training, it enables me to understand the minutiae of contractual performance. What can I do to actually improve the way my medical trend is going? Where are the opportunities to impact patients differently? Those are the types of insights I would receive from an actuary but may not be sophisticated enough to produce myself, and actuarial AI supports that. So it really is this idea of democratizing access to actuarial expertise.

When a provider group signs a value-based care contract with a health plan, what is it actually seeing during the year? What does it know about how it's doing, and when does it find out whether the contract was profitable?

You would think, from a business standpoint, you'd always want to know how well you're performing financially. But the reality is, and I say this in part having spent time on the provider side, both at a company called Oak Street Health and at my own business, Conduce Health, back in the day, that even though we were in a particular value-based care contract, it was incredibly hard to know how we were performing throughout a given performance year.

A big part of that is that there are a series of data opacities, and the data lives in a variety of different places technologically. So having a holistic picture of how much your patient population is utilizing certain care that's driving dollars, or how well you're actually on track to hit certain benchmarks or measurements, is incredibly hard.

The reality is, during the performance year, you actually don't know how you're performing. You have several estimates. You perhaps work with an actuarial team to forecast based on historical information and anticipated spend, but actually having discrete, verified information on performance is incredibly challenging.

Oftentimes, and this is perhaps one of the craziest statistics to think about, you don't actually know how much you owe or are owed financially until anywhere from three months after that performance year has ended to, in some egregious cases, years after that performance period has ended. We've seen that on the Arbital side, where we've worked with payers and provider organizations that are still trying to navigate exactly what is owed to whom from 2022, and we're now in 2026. That's an extreme example. But unfortunately, not knowing how you're performing until after the performance year is a reality for many of these organizations.

Payers and providers often end up arguing over different numbers when a contract is settled. Why does that happen, and what changes when both sides are looking at the same data?

What it comes down to is that where the data lives, and how to best interpret and process that data, is incredibly challenging and requires a certain amount of expertise and skill. What that means is that you have two different stakeholders, payers and providers, operating under different information, either because the data is not necessarily available to both parties or because the ways in which they're interpreting that data are very different.

The other component that makes it really challenging is trying to reconcile that data with what's actually in the contracts. Going back to my prior point about the complexities of value-based care contracts, having the data is one thing. Having it translate into the minutiae of what's in the contract is another. On that contractual front, it's also important to highlight that, like many contracts, it's up for interpretation. So not only are you trying to reconcile contractual terms to disparate data that may not be readily available to both parties, but you're also trying to interpret what's in the contract and translate it to that data in a way that makes intuitive sense.

Having a universal source of truth creates visibility that's agreed upon by both parties. What that means is that the payer and the provider understand how performance is actually working, how certain contractual terms are being interpreted and then, ideally, at the end of a performance period, who's owed what and how that contract came to a conclusion.

A few years ago, CMS set the goal of getting every traditional Medicare patient and most Medicaid patients into accountable care relationships by 2030. For provider groups, the hard part is usually handling the operational and financial complexity at scale rather than a willingness to take on risk. How does Arbital help manage that complexity?

The way I would think about it, there are a variety of components any provider organization needs to have satisfied in order to navigate a risk-based contract. One is appropriately designing the contract. Two is measuring performance throughout the year to understand opportunities for impact. And three is actually reconciling it satisfactorily based on the relevant information and data that's available to you.

Where we support at Arbital is that entire life cycle of the contract. First, helping groups understand at the onset what's in the contract and what's relevant, which patients are attributed, what the carve-outs are, all of the minutiae associated with that, is critically important.

Second, we support understanding throughout the performance year: How are you doing? And more importantly, what can you do about it? Information on performance is necessary but insufficient to actually improve an outcome. Most provider organizations have their own clinical and operational models that, in theory, are designed to bend the patient cost curve in a way that leads to positive financial results contractually. But how to deploy those levers, and for which patient populations, can be incredibly hard to know without data, insight and information. That's a big part of what Arbital AI is able to do as an actuarial AI tool: surfacing with specificity where there are opportunities for impact and then providing suggestions around the next best operational action. Which lever should you pull, and for which patient populations?

And lastly, as a function of being this universal source of truth around performance, data and everything in between, we can support the reconciliation process so there's less likelihood of disagreement between the payer and the provider, and a true understanding of the most important part, which is how we actually improve patient outcomes and, in turn, enable better compensation for all parties involved.

How does Arbital's software figure out whether a contract is on track while the year is still going? You said it can take months or even years to get that data back now, so what's actually happening under the hood when Arbital looks at one of these contracts?

The way I would think about it is through the lens of Arbital AI. In reality, Arbital AI is not a foundation model wrapper. It is our own proprietary data. But conceptually, it operates very similarly to any AI tool, in the sense of garbage in, garbage out, or quality information in, quality information out. So first and foremost, we're incredibly thoughtful about ingesting contractual information from our partners and then reconciling that with the relevant data fields a provider or a payer has, to ensure that what's surfaced on the platform is accurate and readily understood by both parties. That foundation is critically important, and having Arbital AI access that data and information is also critically important to demonstrate how performance is tracking.

From there, a series of additional calculations enables us to understand with some level of clarity how well performance is tracking throughout a performance year. Whether that's access to IBNR [incurred but not reported claims] calculations, benchmarking forecasts or propensity matching, saying that with this population, this is what the typical trend would look like, all of that gets fueled into Arbital AI. That enables Arbital AI in particular to provide insights into what the trajectory is for MLR [medical loss ratio] and, again, most importantly, what a group can do about it.

So, very concisely, we first ensure that we're ingesting the requisite data at the onset in terms of contract information, patient fields and attribution. Then we layer in these additional benchmarks and propensity matching to ensure we're able to pull the relevant information for Arbital AI. In turn, that enables us, one, to manage all of the incoming data throughout the performance year to have a much clearer picture and, two, to more accurately forecast performance beyond a given performance period to understand where there really are opportunities for impact.

Say a platform flags a patient who's likely headed for an avoidable hospital stay or an elective surgery. How do those alerts and patient chase lists help a group actually get ahead of rising costs?

I go back to my prior point that, in theory, every provider organization has a series of bespoke, personalized levers associated with its model. If you're a primary care group, perhaps it's a patient engagement arm, a medication adherence arm or a post-hospital discharge team that interacts with patients. Whatever it may be, the reality is that all of those resources, all of those levers, are incredibly finite. So what's really important is understanding how you actually deploy those resources throughout the performance year to impact rising costs with patients.

What Arbital AI does is two things. First and foremost, we're able to identify for provider organizations when costs are rising, and a big part of that is ensuring we've got the right data, whether that's ADT [admission, discharge and transfer] feeds, IBNR data or anything related directly to the hospitals within a given area. We can begin to identify where utilization is trending and tracking. That's critically important to understand throughout the performance year.

As an example, knowing that you have a patient population with congestive heart failure that is regularly hospitalized in a given geography you're contractually associated with makes a big difference, because in that instance, we can provide what we believe are the next best actions operationally. How can you deploy, as an example, your care managers to reach out to those patients after discharge for a congestive heart failure exacerbation, to ensure they're on their appropriate medications, they're going to their follow-up visits and so on?

We don't consider our role at Arbital to be actually executing those levers. That's not where our expertise is. It's really on that actuarial side. But what we think we're uniquely qualified to do is help those organizations with those finite, impactful levers understand when to use them, for which populations and under what circumstances, i.e., those rising costs you just referenced.

Arbital is built for large systems with a lot of patients. Could you walk me through what it looks like for a group to get started with the platform?

We're always in this process with new partners in terms of implementation. The way I think about it is, first and foremost, we want to understand the complexity of your contracts. Again, a lot can be up for interpretation in terms of patient attribution. What's in? What's out? What does performance mean? What dollars are involved in the dynamic? So we work very closely with our provider partners and our payer partners to understand exactly what the intention is behind various contracts.

Associated with that, we want to understand the requisite data fields we'll need to display the right information. We often work closely with either actuarial teams or med econ teams to understand the contractual component, and then we work very closely with technical teams to make sure we have the requisite data that actually reconciles with the contracts.

In reality, most groups have access to their contracts and can agree on interpretation. Data is a whole other story. So in some instances, we serve as a helpful lever to allow these organizations to better navigate where their data is and how they can understand and interpret it, and more importantly, how they can ensure it's accurate and clean, that there aren't duplications and everything in between. As a byproduct of that process of reconciling the contracts with the data, we're ensuring that everything displayed on the platform is accurate.

From there, we go through user testing. An AI tool is only as useful as people are willing to adopt it and know how to use it, so we're very thoughtful about how we guide people through the subcomponents of Arbital AI.

I'll walk through those for a moment. Arbital AI ultimately has three distinct components that are incredibly helpful to our partners. One is a standard chatbot that effectively enables [users] to talk to their data and understand the minutiae of performance, as in my prior example: Where am I losing the most money, and why? That's obviously critically important.

The next is what we call Arbital Workspaces, which essentially enables us to better memorialize those conversations in collaborative workspaces, with dynamic charts and graphs that can be shared across team members within a given company. So it's not just one person receiving an insight. You're actually enabling that insight to be institutionalized throughout the organization.

And last are agentic workflows, which ultimately require less of the user to develop the series of questions and insights and walk through what's required. You simply provide the problem to Arbital AI for it to understand what the multiple steps are and what relevant databases need to be pulled from to ultimately solve that problem. That could be something as complex as, "Show me where my MLR is above benchmark based on patient population and ICD-10 codes, and what the suggested next best actions are." All of that can kick off a series of workflows that Arbital AI will develop on behalf of that particular user to provide insights.

I share all that to say those three components can be incredibly helpful, but they do require some user training. So we ensure everything on the platform is accurate by going through that reconciliation phase, and then, in turn, we ensure users actually know how to navigate these really complex but easy-to-use tools in a way that's going to provide real value.

For provider group leaders weighing their first value-based contract or renegotiating the ones they have, how does Arbital help them keep up with regulatory changes and new cost trends?

Taking a step back, what makes Arbital unique in the marketplace is really an amalgam of expertise. Technically, we have a series of AI engineers who have built a variety of AI tools, have those insights and are at the cutting edge of understanding the art of the possible. And then, equally important, our other pillar of intellect and subject matter expertise is actuaries who are regularly in the market, tracking trends from a CMMI [Center for Medicare and Medicaid Innovation] standpoint or broader CMS standpoint, and informing clients and customers about where there are real opportunities.

We recognize that in this ecosystem of value-based care, or payer and provider contracts, having sophisticated technology and software is fantastic, but it doesn't eliminate the actual people component of what makes this business work. Fortunately, we have a team of just north of 25 VBC-trained actuaries who are regularly in the market, understanding regulatory changes, and can provide those insights directly to our payer and provider partners.

What I think is most interesting as well, when I think about the opportunities with AI and, in particular, actuarial AI and our Arbital AI tool, is that we're actually able to codify some of those insights. Wherever we can pull in relevant regulatory information as a requisite data source for Arbital AI, what we call our sciences team, our subject matter expertise team, is regularly ensuring it's available to Arbital AI to provide those insights.

So there's a very human component to this, and there's also a platform component that, in aggregate, provides a lot of value in understanding which programs I should be in, whether it's ACO LEAD, one of the MA [Medicare Advantage] contracts or any of the complex programs that are out there.

Is there anything we glossed over or didn't get to today that you think leaders should keep in mind?

While we've spoken a lot about value-based care contracts, which is critically important, I think it's worth highlighting that we at Arbital fundamentally believe actuarial AI applies broadly to any payer and provider contract. Value-based care oftentimes is put on a pedestal, appropriately so, for the impact it can have and the associated complexity. That doesn't mean fee-for-service contracts aren't also incredibly challenging for a variety of reasons.

What we're realizing is that, by the very nature of creating Arbital AI, we're not limited to value-based care. That's the beauty of these tools. Through democratizing access to actuarial expertise, it can be applied to fee-for-service contracts and to broader health care finance contracts that are incredibly complex.

I say all that to say that, as your readers are evaluating all of this, we're at a very exciting point in time, where AI is really pacing with the opportunity to have an impact on the broader health care ecosystem, and we at Arbital are very excited to play what we believe is some role in that in this era of innovation.