AI, Humans, and the Future of Benefits: A Conversation with HealthJoy's CEO

AI is everywhere in healthcare and benefits right now. From your seat as CEO, what are you actually seeing?

What we're seeing in benefits is that a lot of the unglamorous work is where AI is being leveraged first. The first inclination of most leaders who are taking traditional incumbent businesses and trying to modernize them, is to look at cost structure: how do we do the same thing, in a more cost-effective —  and faster —  way with technology?

The hospitals were the first to adopt at scale, and what they've done is turn it into an arms race around revenue cycle management and upcoding. The carriers are adopting their own payment integrity products to combat that. But the people in the middle, HR and the employee, are lagging from a whole-market perspective. We have to figure out how to better arm employees and the HR leaders who are managing much more complex plans than they were two years ago, and allow them to actually attain the goals they're trying to reach with their benefits.

The challenge is that you're doing all of this against a backdrop of massive cost inflation, trying to be innovative and creative while combating a cost structure that is increasing by more than 100% in some cases. That's the reality for a lot of companies right now, especially this renewal cycle especially.

Beyond the back-office work, there are a lot of innovations happening at the member level — interacting through AI, using large language models to process data. Where it gets particularly interesting is the frontier models: the most advanced AI systems available today. A lot of interfaces you thought would never go away are moving over. The idea of going to Dr. Google is now "I'm going to go talk to an AI." That's a real and significant shift.

When you look at the benefits technology landscape today, where do you think AI is being used well and where is it being oversold?

I would actually say the AI capability in benefits is undersold and massively underestimated. What most vendors have shipped is the part that's oversold. But what the models can actually do is a different story entirely.

The frontier models are advancing so quickly that sometimes the things that were made six months ago are already legacy software. And every release cycle of these models is almost replacing an entire segment of work that people said six months ago could not be replaced. So it's mixed emotions, excitement and then "wow, that's a whole category of the market that has to be rethought now."

The biggest way I see this show up is in the interfaces. The carrier website is my classic example. No one's been going to it, no one likes it, and no one's ever had a great experience with it. Carriers were losing to Dr. Google for the last two decades. They're going to lose this next wave too. The idea that you're going to have a legacy piece of software or manage all these separate interfaces will fall by the wayside.

I think the technology itself is massively underestimated, which is incredible, especially since a lot of people are saying it's going to end the world. It's easy and comfortable to say "I see there's change" and then keep doing what you're doing. My view is that if we need to build it, the technology is going to get there a lot faster than we're going to be able to go through our sales cycle. That's the reality we're operating in.

There's a narrative in tech that AI will eventually replace human support entirely. Do you believe that's true in benefits?

My honest assessment is that most of the specific tasks people are doing today, if they keep doing them the same way, AI will most likely be able to do. But that doesn’t mean the work itself disappears. It means the nature of the work changes. That’s especially true for benefits consultants, whose value is increasingly about strategy, relationships, and judgment. AI handles the administrative tasks. The humans move up to higher-stakes work that’s more consequential. 

What's going to happen is the acuity level of human work is going to rise, focused on increasingly complex cases. And then there's going to be a point where a lot of those jobs are handled by AI. But responsibility and accountability aren't the same thing. AI can be responsible for an outcome. But being accountable for it, owning it when something goes wrong, that still has to be a human. 

When AI fails, a human is there — someone who wrote the prompt or steered it to where it needed to go. I don't believe the frontier models are going to sign a BAA and go at risk for a 2,000-life group for an incorrect diagnosis. Not at least in the next five years, maybe ten. In the current risk profile, the way our firms are contracting, accountability is going to have to be a human. And accountability means the judgment layer is also going to be predominantly human.

On the concern that AI is going to take jobs: the back-office and administrative work in healthcare is where AI is having the most immediate impact, and that creates an opportunity. It frees humans up to focus on the work that actually requires them like complex care, higher acuity cases, and the conditions that need a more thoughtful, data-integrated approach. There’s going to be a creative element to what human work looks like in benefits. It means we can do something better and more impactful than what was possible before.

What does a human bring to a benefits interaction that AI genuinely can't replicate?

It's judgment and accountability, when it comes down to it.

The conventional wisdom has been that human empathy is irreplaceable, the primary driver of behavior change in healthcare. But there's emerging research suggesting that the empathy argument is more complicated than we thought. Studies are showing that AI can be highly effective at driving behavior change in certain contexts — building rapport and coming across as unbiased in ways that actually move people toward better decisions. Behavior change is the core challenge in healthcare. How do we get someone to actually follow through on the right clinical protocols? It's going to be interesting to see how that evolves.

What I'm more confident about are the judgment and accountability layers. Hospital systems aren't going to be able to blame the machine. The machine manufacturer isn't going to take that responsibility and accountability, not for at least five or six years. On the judgment layer, making sure the overall system works, humans are going to have to be responsible for the steering. And right now, in the near term, there's clearly an important role for human empathy in driving patients toward the right outcomes. That's something we've deliberately invested in.

At HealthJoy, our team is able to work on harder and harder problems because AI is handling more of the routine work better than ever before. And there's always more to do.  There's never a moment where we run out of ways to drive better outcomes for our customers and members.

How do you think about the relationship between JOY and HealthJoy's care team?

We definitely have a human-led agentic workforce, meaning AI takes action on behalf of members, but humans set the boundaries for how it operates. We use AI to get the answer, but our human concierge team sets the judgment boundaries. They own the call. If something goes wrong for a member, the buck stops with the concierge team. We want to make sure we're ultimately getting members to the most affordable, high-quality option within the network. As the liability framework matures, maybe that boundary moves over time. But today, that's how it sits, with people. We do not believe that AI judgment is safer yet, by any stretch of the imagination.

JOY's job is to give the best possible answer, as fast as possible, from the data available. When there's a gap in that data, our concierge team steps in. The humans get to work on the higher-acuity, more specialized, and complicated cases.

The challenge with healthcare is that there has to be a compliance boundary. The level of security around protected health information means that a lot of the technology other industries get to use, we don't get to use for a while. We’re in all 50 states, employees from rural Montana to downtown LA, across every possible combination. The data is not perfect everywhere, and won't be for a very long time. That's why we pair the team.

Ultimately, the human-led concierge team sets the boundaries, restrictions, and judgment, and we let the AI do its work. Then we track the impact. The concierge team is the human in the loop, seeing where the AI does well and where the user didn't like the answer, and helping us understand whether they didn't like it because it wasn't what they wanted to hear or because it was actually wrong. That's what helps us train the system long term. The concierge team is a critical part of making our model great. It wouldn't be great without the millions of conversations, interactions, and episodes of care we've helped members with over the years.

There's a trust dimension here too. Do you think employees trust AI with their health decisions? What does it take to earn that trust?

I think the trust problem is probably wrong answers, not AI.

Trust is inherent: if you believe something is smart and working in your interest, you trust you're going to get the right answers. I don't believe a lot of people trust that when they call the carrier they're going to get the right answer, and those are humans on the other end of the line. Just from carriers' NPS scores, clearly that's not the case. People also distrust the answers because they suspect the carrier's interests aren't aligned with their own. 

There was a study published in a leading scientific journal, across 77,000 adults, where AI was more persuasive than a human debater at measurably shifting beliefs. What strikes me about that finding is that the AI built rapport and came across as unbiased, and that's what drove people toward better decisions. That's trust. I think we're all going to trust it a lot more as it gives us better answers.

Think about Google. There was a super high amount of trust in Google. It took a while, but once it got there, it was a verb. You would Google anything and think, "I'm going to get the right answer." To me, that's trust because you got to the right answer the fastest possible way. I believe we're approaching that moment with AI in benefits. And the better the answers get, the faster that trust will follow.

Where do you think the benefits industry gets AI wrong most often?

The industry is vastly underestimating how much change is about to happen.

The next five years are going to feel like fifty: ten times more change than we've seen in the last five. If the last five were challenging, the next five are going to demand a fundamentally different posture, one of urgency and active adoption rather than observation.

The players that have already adopted are going to accelerate faster. Hospital systems are going to get even better at upcoding and revenue optimization. Carriers will have their own sets of tactics. Employees and HR are lagging. Long term, there should be significant relief across the whole ecosystem as unit economics come down. But in the near term, those already in motion are going to pull further ahead.

Benefits is not the primary business of most companies. The average company in the United States didn't choose to be in the benefits business, but they are. And that's going to create a real challenge as the pace of change accelerates.

Andy Grove wrote a book called Only the Paranoid Survive, and I think about it constantly right now. This is the time to be paranoid, to make sure we’re not caught flat-footed by one of the sharpest inflection points we’ll see in our lifetimes. That’s a hard ask for teams that are already stretched thin. But the tools to help have finally arrived. When I think of benefits leaders, especially on the brokerage side: if you’re not aggressively figuring this out right now, a competitor in your territory already is. It’s already come, and it’s only going to accelerate. 

What would you say to an HR leader who is skeptical that AI can actually improve their employees' benefits experience?

I would ask them: how many of your employees still phone in to order pizza?

I can tell you it's a very small minority of people who still want to pick up the phone and call in their takeout or delivery order. People want answers faster, and they trust the tools once they see the right answers coming from them.

I'd also say: ask any vendor what happened after their AI answered the question. If a vendor can't tell you that, and if you've been skeptical because you didn't have a good experience, they were probably just building an FAQ for you. And that's not very useful. Good answers aren't just ingesting a benefits guide and dropping it into a generic AI model as there's far too much context that approach will never capture. 

Were you skeptical of using Google when it came out? Being skeptical of AI in benefits right now is like being skeptical of fundamental technology that’s already all around you, and it’s only going to become more pervasive. 

And finally, where do you think the future of AI in benefits is headed?

In the near term, all of the back-office and support aspects of delivering benefits are going to get a lot easier for everyone to administer. At the same time, I don't see prices magically plateauing. We're still going to have this backdrop of hyperinflationary healthcare costs, but layered on top of that, ideally, is an easier system to administer and significantly better data. The data is going to continue to get a lot better, especially now that you can point AI at larger datasets to make sense of them.

In the midterm, five to ten years, I think we'll start seeing unit economics go down. It's hard for me to think about the future of benefits without thinking about the future of healthcare in the United States. You have to look at one to understand how the other is going to interact.

What I think of as the ideal state is attaining the HealthJoy mission: how do we get most people the most affordable, high-quality option on their health plan? It's the first time I've really felt that's truly attainable at mass scale. We were several years ahead of where the technology stack was actually able to deliver on a promise that big, and now we're able to deliver on it. 

Our belief is that employees have to have skin in the game. They have to be part of the solution. This can't just be on the consultant and employer anymore. Those days are gone. The tools are now good enough that employees can shop for their care in a way that simply wasn't possible before. And with that comes a new kind of responsibility.

We've talked about empowering employees for a long time. But empowerment without accountability only goes so far. What we're moving toward is a world where employees have real tools, real information, and a real role in managing their own healthcare decisions. Those who engage with that opportunity will be better off for it. The opportunity is there. The question is whether they’ll take it. 

What comes through in this conversation is a consistent throughline: at HealthJoy, AI and humans aren't in competition. They're in collaboration, each doing what they do best, so the member gets the outcome they need. That's the platform we've built. And if Justin's read on the next five years is right, the time to get serious about it is now.

Read more on what the AI + humans model looks like in practice