AI adoption in healthcare is exploding, driven by promises to improve efficiency and lower costs. But that promise flies in the face of a more potent truth about our healthcare system: decades of technological innovation and increased efficiency have only led to an increase in healthcare spending. We now spend more than $15,000 per person per year for healthcare, which consumes 18% of GDP, yet we have lower life expectancy than other developed countries. That reality hangs like an albatross around our economy and our families.

At PHTI, we’ve independently assessed more than 60 digital health technologies for chronic conditions that affect roughly half of all Americans. The pattern is clear: efficiency gains from new health technologies do not automatically translate into lower costs. The reasons are many, but a key driver is how we pay for care.

The real question is not whether AI can reduce the total cost of care, but whether it will. A 2023 study from the National Bureau of Economic Research estimated that AI could deliver $200 to $360 billion in annual healthcare savings. The technology clearly has that capability. But will healthcare leaders, payers, employers, and policymakers make the hard decisions required to turn efficiency gains into actual savings for patients?

Not all actors in healthcare have an incentive to reduce costs, and the way our system is designed makes it very difficult to push back against price increases. Like so many other technologies, AI may get tacked on as an extra service with a high price tag, or health systems may simply absorb efficiencies gained without reducing prices. Without making changes to the current fee-for-service payment model, AI could simply make expensive care more profitable rather than more affordable.

Current fee-for-service models reward volume, not efficiency. When AI makes providers more efficient, they can actually lose revenue. We have already seen that that AI scribes are being used not to reduce costs, but to maximize billing by helping providers increase revenue for each patient encounter.

Beyond the sticker price, the “installation period” for AI is chaotic and expensive. Training workers to use new platforms, managing transitions, and building infrastructure all add significant costs that rarely show up in the pitch deck. A systematic review published in 2025 in npj Digital Medicine found that many AI cost-effectiveness studies “may overestimate benefits” by failing to account for “indirect costs, infrastructure investments” and other hidden implementation expenses. In the short term, AI adoption may actually increase costs—even if long-term savings are possible.

Then there is the question of cost-savings through labor substitution or augmentation. Cultural resistance, workforce concerns, and regulatory barriers make it politically difficult to let AI replace human labor in healthcare, even in settings with severe provider shortages or to manage administrative tasks. Look no further than the pushback to the idea of using AI for prior authorization. We also currently lack the regulatory and payment frameworks that would enable more autonomous AI-driven clinical care.

Not all healthcare actors want to reduce costs. Providers, technology vendors, and other stakeholders often benefit from the current high-spending environment—and even when AI creates real efficiency gains, those savings tend to get absorbed by healthcare organizations rather than passed on to the patients, employers, and taxpayers who pay the bills. Making matters worse, most purchasers don’t have the information they need to evaluate whether AI investments will actually reduce their spending or just improve someone else’s margins. HR departments rely on consultants and brokers who don’t always act in the employer’s best interest. The result: AI improves the bottom line for healthcare organizations while costs keep rising for patients and employers.

So what would it actually take for AI to deliver on its promise of reducing the cost of care paid by health plans and systems, employers, and, ultimately, patients?

Start with how we pay for care. For efficiency gains to translate into actual cost savings, payers would need to reduce what they pay providers to account for AI-driven productivity improvements. That means health plans must be brave enough to renegotiate rates downward when AI reduces the cost of delivering care—not just accept the savings as improved margins. This requires political courage and a willingness to fundamentally restructure how healthcare gets paid. It is one of the most difficult changes to implement in the industry. It is also the most important.

We also need honest accounting of what AI actually costs to deploy—not only the technology price tag, but the full implementation burden. And we need market discipline to ensure that the introduction of technology plays a deflationary role in healthcare. If a tool saves a health system $10 million in labor costs but gets priced at $12 million, that’s a markup disguised as innovation.

To realize real efficiency gains, there also needs to be an honest conversation about letting AI reduce labor costs and alleviate shortages where appropriate. That conversation is politically difficult but avoiding it doesn’t make the problem go away.

Finally, purchasers need better tools to hold the system accountable. Most employers don’t have the information or leverage to evaluate whether AI investments will actually reduce their costs or just improve efficiency for someone else. Performance-based contracts can help change that—tying payment to measurable clinical outcomes rather than taking vendors at their word. At PHTI, we’ve already built playbooks to help employers and health plans do this for virtual health solutions. As AI tools mature, purchasers will need the same accountability framework to protect their investments and ensure those efficiency gains actually reach patients.

The technology exists for AI to reduce healthcare costs. But operational excellence alone isn’t enough. Without fundamental changes to payment models, incentive structures, and the courage to make hard decisions about pricing and reimbursement, AI will simply make our current high-cost system more efficient at being expensive. We need bold leaders in healthcare to push for those structural reforms—before we spend billions more on technology that makes the same broken system run faster.