Health insurance protects against financial risk, but it also shapes how care is used. Artificial intelligence points to a more targeted approach to managing incentives.
By Xinyi (Cindy) Hu
Health insurance rests on a simple, humane promise: When you get sick, your financial life shouldn’t collapse before your body does. But the very system designed to protect people brings its own tension. Pooling resources to spread individual risk means balancing access with cost, making sure that everyone gets the care they need while avoiding waste, inefficiency, or misuse. Economists have a term for this: moral hazard.
Moral hazard in health insurance refers to the tendency for individuals to consume more medical care—or to take fewer preventive steps—because insurance shields them from the full cost. The phenomenon typically arises from psychology: When someone else is footing the bill, it’s easier to say yes to tests, treatments, consultations, or ER visits that feel reassuring even if they’re not strictly necessary. Similarly, when out-of-pocket costs are low, it’s tempting to delay preventive care. The ripple effects of these choices impact health care systems daily.
Economists distinguish between two primary forms. “Ex-ante” moral hazard emerges before illness: People engage less diligently in health maintenance, knowing that insurance will cushion the consequences. This form is subtle—few individuals consciously think, “I’ll skip my blood pressure meds; insurance will fix me later.” Instead, insurance reduces the sense of urgency, lowering the perceived personal cost of neglect.1 “Ex-post” moral hazard occurs after illness has developed, when insurance lowers the out-of-pocket price of care and thereby influences utilization decisions. Once sick, insured individuals may be more likely to use additional or higher-intensity medical services—such as diagnostic imaging, emergency care, or specialist visits—because insurance shields them from the full financial cost.2 Providers can play a role, too, by ordering extra tests or procedures out of caution, habit, reimbursement structures, or fear of litigation.
Structural Inefficiencies
Long before artificial intelligence (AI) entered health care, insurers were already grappling with these inherent inefficiencies. Their early solutions focused on making patients share some of the cost. Deductibles, copayments, and coinsurance are classic examples. By requiring patients to pay the first portion of health care expenses themselves, deductibles discourage unnecessary early-year spending. Copays add friction to each visit or prescription, reminding patients that resources are finite. Coinsurance—covering a percentage of medical costs—keeps patients mindful of price, even for costly services.
These methods reduce not just unnecessary care, but they also discourage necessary care, too. Studies like the RAND Health Insurance Experiment show that when people face higher out-of-pocket costs, they use less health care overall—but the reduction occurs indiscriminately. People skip unnecessary doctor visits and important ones alike. Preventive services fall just as much as redundant services. This creates a paradox: Moral hazard goes down, but so does health, especially among low-income and chronically ill individuals.
Realizing that cost-sharing alone could not solve the system’s deeper structural inefficiencies, insurers in the 1980s and 1990s began to experiment with managed care. They created narrower networks, established primary care physicians as gatekeepers to specialists, and tested new payment models aimed at curbing overuse. Instead of paying providers for every service, some insurers adopted capitation—paying doctors a fixed amount per patient regardless of how many services they delivered. Others used bundled payments, compensating providers for an entire episode of care rather than itemizing each test and procedure. These approaches targeted provider-side moral hazard, recognizing that clinicians wield enormous influence over health spending because they diagnose illness, recommend treatments, and determine which services patients ultimately receive—or forgo.3
Yet, for many people, managed care became synonymous with restriction.4 Utilization reviews and prior authorization processes, meant to prevent unnecessary spending, often delayed critical treatments and buried physicians in administrative tasks.5 Patients felt uncertain about whether care would be approved, and physicians felt that insurers were second-guessing their clinical judgment.6 While these mechanisms did reduce some waste, they also created friction that left patients, providers, and insurers frustrated.
AI’s Potential to Shape Health Care Use
AI now offers the potential to transform how insurers approach this longstanding dilemma. Instead of restricting or imposing financial burdens on patients, AI can reshape the system through precision, prediction, personalization, and prevention,7 impacting nearly every aspect of health care utilization.
One of AI’s most accessible and immediate contributions lies in triage.8 Many unnecessary health care encounters stem from uncertainty: People aren’t always sure whether their symptoms are serious, whether they need a doctor, or whether a late-night ER visit is warranted. AI-powered symptom checkers and triage tools are beginning to fill this gap, offering real-time guidance that directs individuals to the appropriate level of care. These tools can recognize patterns across millions of examples, providing a level of statistical sophistication that no human nurse hotline could match.9 AI can reduce the ex-post moral hazard by directing people away from high-cost settings when unnecessary—and toward appropriate care when needed.
Predictive analytics takes this even further.10 AI can analyze vast datasets to identify individuals likely to overuse certain services or underuse others. For example, it might detect patterns suggesting someone is prone to unnecessary ER visits or at risk of missing essential preventive screenings. Insurers can use these insights to intervene early, offering personalized reminders, targeted benefits, or care management support. AI can help insurers to anticipate poor choices and gently guide patients toward better decisions.
AI’s Potential to Reduce Waste
Fraud, waste, and abuse have long challenged health care systems.11 Traditional rule-based detection systems tend to catch only the most obvious abuses—patterns defined in advance by analysts and compliance teams—while missing more subtle or evolving schemes.12 AI, however, is exceptionally good at identifying anomalies in large datasets. It can detect unusual billing patterns, phantom claims, suspicious prescriptions, and clusters of behavior indicative of coordinated exploitation. By targeting the financial incentives that enable abusive patterns, AI indirectly reduces a significant portion of systemic moral hazard.
On the clinical front, AI-powered decision support systems help clinicians avoid unnecessary care by providing evidence-based guidance at the point of service.13 These tools, embedded within electronic health records, can flag duplicative imaging orders, highlight more cost-effective medication options, or identify deviations from clinical guidelines. Unlike traditional prior authorization systems—which often require lengthy paperwork and insurer approval—AI decision support for clinicians is immediate, integrated, and advisory.
Personalization, Prevention, and the Future of Incentives
Perhaps AI’s most transformative promise is its ability to personalize incentives. Traditional cost-sharing structures assume that all patients respond similarly to financial incentives, but real-world behavior is far more complex. AI can analyze individual health behaviors, risk profiles, and usage patterns to design cost-sharing frameworks tailored to each patient. For example, a patient with a chronic condition might receive lower cost-sharing for preventive visits, medication refills, or monitoring devices that reduce the likelihood of complications. Meanwhile, another individual who tends to overuse low-value services might see more targeted cost-sharing for those services, rather than uniformly higher out-of-pocket fees. This degree of personalization shifts incentives from punitive to supportive and rational.
Wearables and remote monitoring technologies, combined with AI, introduce an entirely new approach to address ex-ante moral hazard.14 Devices that track heart rate, sleep patterns, blood pressure, glucose, and physical activity generate continuous streams of data. AI can interpret these trends, detect early warnings, and trigger interventions before costly complications arise. Insurers can encourage participation through incentives, rewards, or reduced premiums. Patients gain greater self-awareness, and insurers benefit from reduced long-term costs.
AI can also bring greater transparency to health care. In a system where costs are opaque and quality can be hard to assess,15 AI-driven tools can help patients understand the financial implications of their choices. Algorithms can compare hospitals or clinics based on cost and outcomes, predict out-of-pocket expenses in real time, and suggest alternatives when a chosen treatment is clinically equivalent to a cheaper option. With clearer information, patients make more cost-conscious decisions—not because they are forced to, but because they finally understand how.
Care coordination represents a persistent challenge in health care delivery—one that aligns closely with AI’s strengths.16 Fragmented records, disconnected providers, and redundant testing often result not from intentional overuse but from poor communication across the system. AI can bridge these gaps by integrating data across providers, flagging inconsistencies, and ensuring clinicians share accurate, up-to-date information. This reduces waste at a structural level—without the friction that traditional utilization controls introduce.
Ethical Dilemmas
Yet the rise of AI also introduces ethical complexities. Privacy concerns are key when intimate health data is analyzed continuously. Algorithmic bias could reinforce or worsen inequities if left unchecked.17 Some worry that AI “nudges” might become paternalistic or manipulative, raising questions of patient autonomy.18 Clinicians, already managing digital overload, may view AI tools as yet another layer of intrusion unless they are thoughtfully designed.19
Despite these challenges, AI has a potential to improve how we manage moral hazard. Instead of asking patients to pay more, it helps them understand better. Instead of questioning doctors after the fact, it can support them in the moment. Instead of treating everyone the same, it adapts to individual needs.
And perhaps the most fitting anecdote comes from a veteran actuary who once quipped, “Trying to manage moral hazard without good data is like trying to find your glasses without your glasses.” AI, in many ways, is that missing pair—a way to finally see clearly the patterns, behaviors, and opportunities that have long shaped the uneasy dance among risk, responsibility, and care.
XINYI (CINDY) HU, MAAA, ASA, is an actuarial associate at Mutual of Omaha specializing in Medicare supplement and group insurance. She can be reached at xinyihu68@gmail.com.
Endnotes
- Kenneth Arrow (1963), “Uncertainty and the Welfare Economics of Medical Care,” American Economic Review.
- RAND Health Insurance Experiment (Newhouse et al., 1980s; summarized in Newhouse 1993).
- Robert Evans (1974), “Supplier-Induced Demand: Some Empirical Evidence and Implications,” The Economics of Health and Medical Care.
- KFF reports from 1997 on public attitudes toward managed care. KFF surveys found that large majorities of Americans believed managed care plans made it harder to see specialists and reduced the time doctors spend with patients.
- American Medical Association surveys (e.g., the 2022 and 2024 AMA Prior Authorization Physician Surveys) consistently find that prior authorization requirements delay patient care, impose substantial administrative burdens on physicians and their staff, and that a notable share of physicians report clinically adverse outcomes associated with these delays.
- David Mechanic (2004), “The rise and fall of managed care,” Journal of Health and Social Behavior.
- National Academy of Medicine (2019), “Artificial Intelligence in Health Care: The Hope, the Hype, the Promise, the Peril”.
- Tyler, S., Olis, M., Aust, N., Patel, L., Simon, L., Triantafyllidis, C., Patel, V., Lee, D. W., Ginsberg, B., Ahmad, H., & Jacobs, R. J. (2024). “Use of Artificial Intelligence in Triage in Hospital Emergency Departments: A Scoping Review”.
- Beam & Kohane (2018), “Big data and machine learning in health care,” JAMA.
- National Academy of Medicine (2019), “Artificial Intelligence in Health Care: The Hope, the Hype, the Promise, the Peril”.
- Centers for Medicare & Medicaid Services estimates and reports on fraud, waste, and abuse (FWA) in Medicare and Medicaid. CMS and the HHS Office of Inspector General describe FWA as persistent, large-scale, difficult to eliminate with traditional controls; Government Accountability Office reports labeling health care fraud as a “high-risk” area for decades.
- Md Kamrul Hasan Chy (2024), “Proactive fraud defense: Machine learning’s evolving role in against online fraud,” World Journal of Advanced Research and Reviews.
- National Academy of Medicine (2019), “Artificial Intelligence in Health Care: The Hope, the Hype, the Promise, the Peril”. supporting clinician judgment, and reducing unnecessary or low-value services.
- Eric J. Topol (2019), “High-performance medicine: the convergence of human and artificial intelligence,” Nature Medicine.
- Institute of Medicine (2001), “Crossing the Quality Chasm,” and Government Accountability Office reports on health care price transparency consistently find that patients rarely know prices in advance, have difficulty assessing quality, and lack usable information to compare options, conditions that undermine informed consumer choice in health care markets.
- National Academy of Medicine (2019), “Artificial Intelligence in Health Care: The Hope, the Hype, the Promise, the Peril”.
- Ziad Obermeyer et al. (2019), Science, “Dissecting racial bias in an algorithm used to manage the health of populations.”
- Richard Thaler and Cass Sunstein (2008), “Nudge”, and Hastings Center bioethics commentary on digital nudging in health care highlight ethical concerns surrounding choice architecture, including the tension between guidance and manipulation, the importance of transparency and informed consent, and the need to respect patient autonomy—concerns that are heightened when nudges are personalized, opaque, or embedded in digital and clinical systems.
- National Academies of Sciences, Engineering, and Medicine (2019). “Taking Action Against Clinician Burnout: A Systems Approach to Professional Well-Being”; and American Medical Association survey (2025).