Latest / Fintech Conversations with Fexingo: Banking, Payments, and Financial Software Companies / How Fintechs Use AI for Personalized Health Savings
Transcript
- Lucas: So Luna, here's a number that stopped me this week: there are now over 35 million health savings accounts in the United States, holding north of $130 billion in assets. And the vast majority of those accounts are sitting in cash, earning next to nothing. Luna: I've got an HSA myself, and honestly, I'm probably guilty of that. You set up the account, you put some money in, and then it just sits there. Lucas: Exactly. And that's the classic fintech opportunity, right? You've got a huge pool of assets that are under-optimized, and the incumbents haven't done much to change the behavior. But a few startups are now using AI to tackle this problem head-on — not by building a robo-advisor for HSAs, but by personalizing the savings and investment recommendations based on each user's actual health and financial data. Luna: So they're not just saying 'invest in a target-date fund.' They're actually looking at your specific situation? Lucas: Right. Take a company like Carrot, which started as a fertility benefits platform and now does broader health savings and investment guidance. Their AI looks at claims history, pharmacy data, even the user's tax bracket. If the model sees you're in a high deductible health plan and you've had consistent prescription costs, it might recommend keeping more cash in the account for near-term expenses. But if you're young, healthy, and in a high tax bracket, it might suggest maxing out the contribution and putting the surplus into a moderate-growth ETF. Luna: That's actually useful. The standard advice is always 'max out your HSA,' but that doesn't make sense if you know you're going to need the money for a surgery next quarter. Lucas: Exactly. And the AI can get more granular. One feature I saw from a fintech called Lively — they're not as big as HealthEquity, but they have a slick digital platform — they use machine learning to predict your out-of-pocket medical costs for the next twelve months based on your past claims and any known upcoming procedures. Then they recommend a specific contribution amount, down to the dollar, that covers those expenses while maximizing the tax benefit. Luna: So it's almost like a budget for your health spending. How accurate are those predictions? Lucas: According to their own case studies, within about 15 percent of actual spending for the median user. Which is pretty good, considering how variable medical costs can be. And the system gets better over time — it learns from the whole user base, so if a lot of people with similar profiles end up spending more on something like physical therapy, the model adjusts. Luna: Right, because you've got this feedback loop. The more people use it, the smarter the recommendations get. That's the classic AI flywheel. Lucas: Exactly. And the bigger play here isn't just contribution amounts — it's investment allocation. Most HSA custodians offer a handful of mutual funds, but users don't know which to pick. So HealthEquity, which is the largest HSA provider by assets, recently rolled out an ai driven 'Smart Portfolio' that asks a few questions about your risk tolerance and time horizon, then allocates your balance across a mix of ETFs. But the more interesting part is that it rebalances automatically based on changes in your health spending patterns. If you suddenly start incurring more medical expenses, it shifts some assets back to cash so you don't have to sell investments at a loss. Luna: That's a smart feature. It's like a dynamic glide path, but for health rather than retirement. Lucas: Precisely. And this is where the fintech mindset really changes things. The traditional HSA industry was essentially a custodial business — hold the cash, charge a fee, maybe offer a few fund options. But now you've got these startups using AI to turn the HSA into an active financial planning tool. And the numbers suggest it's working. HealthEquity reported that users who opted into the Smart Portfolio increased their average contribution by about 23 percent year-over-year, compared to 8 percent for users who didn't. Luna: Twenty-three percent is huge. That's not just a nudge; that's a real behavior change. And it makes sense — if someone tells you exactly how much to save and where to put it, you're more likely to do it. Lucas: And that's the thing we keep coming back to on this show. If these conversations are useful for what you're building or running, we want to keep them coming without interruption. We deliberately don't run ads on these episodes — it's just the two of us talking through the angles that matter to you. If you want to support that choice, the link is buy me a coffee dot com slash fexingo. It's a small way to keep the conversation ad-free and focused. Luna: Yeah, it's a simple way to help. And honestly, it makes a difference. Now, back to the AI side — I'm curious about the data privacy angle. These AI models are ingesting claims data, pharmacy records, maybe even lab results. That's incredibly sensitive information. Lucas: It is, and it's the biggest barrier to adoption for some of these features. The companies have to be hipaa compliant, obviously, and they typically anonymize the data before it goes into any training model. But the most interesting approach I've seen is from a startup called Picwell — they use a technique called differential privacy to add statistical noise to the data, so the model can learn patterns without ever seeing an individual's actual health information. That's the kind of technical trust-building that'll be necessary if these tools are going to go mainstream. Luna: So the AI never actually sees your diagnosis codes; it just sees a noisy version. That seems like a smart way to handle it. Lucas: It is. And there's another layer: some of these platforms are starting to integrate with employer benefits administration APIs. So when you log into your benefits portal at work, the HSA recommendation engine can see your salary, your tax filing status, and even your expected bonus — all with permission, of course — and then calculate the optimal contribution to minimize your FICA taxes. Luna: Wait — FICA taxes? HSA contributions are pre fica? I thought they were just pre-income tax. Lucas: They are if you contribute through payroll deduction. That's the triple tax advantage people talk about: contributions avoid federal income tax, state income tax in most states, and FICA — that's Social Security and Medicare taxes. So if you're a high earner, the FICA savings alone can be substantial. But most people don't realize it, because the payroll deduction just shows up as a line item. The AI can surface that: 'If you increase your HSA contribution by $500 per pay period, you'll save about $38 in FICA taxes each paycheck.' That's a concrete, personalized nudge. Luna: Okay, that's powerful. And it's exactly the kind of thing a general financial advisor might not think to mention, because it's so specific to the HSA mechanism. Lucas: Right. And the fintechs are building for that specificity. Let me give you another example: a company called Further, which is a newer entrant, uses natural language processing to scan the user's explanation of benefits documents — those EOBs that nobody reads — and extract key information like remaining deductible, out-of-pocket maximum, and which providers are in-network. Then it predicts how much you're likely to spend over the rest of the year and recommends a contribution strategy. They claim their users save an average of $1,200 per year in taxes and out-of-pocket costs. Luna: That's a lot of money for something that's essentially automated. And it ties back to the broader trend of embedded finance — you're not going to a separate app to manage your HSA; it's integrated into your benefits experience. Lucas: Exactly. And that integration is the key distribution channel. The incumbents like HealthEquity and Optum Bank are also investing in these AI features, but they're playing catch-up to the startups in terms of user experience. The real question is whether the large HSA custodians will acquire these startups or build their own. We've already seen some movement — HealthEquity acquired a small AI firm called Punch in 2024 to bolster their recommendation engine. Luna: Makes sense. And as HSAs continue to grow — especially with the 2026 contribution limit increase to $4,300 for individuals and $8,600 for families — the market is only going to get bigger. Lucas: Right, those limits are indexed to inflation, and with healthcare costs rising, the accounts are becoming a bigger part of household balance sheets. So the fintech opportunity here isn't just about optimizing an existing product — it's about redefining what an HSA can be. If these AI tools work, the HSA could evolve from a passive savings account into an active health spending and investment command center. Luna: And that's a future I'd sign up for — especially if it saves me from having to read those EOBs myself. Lucas: Exactly. No one wants to do that.