Prior Authorization

Before a doctor in the US performs an expensive treatment or prescribes a particular medication, they often must obtain permission from the insurer. This permission is called Prior Authorization. The insurer then checks whether the treatment is considered necessary under its rules. If it agrees, it covers the costs. If it refuses, the patient must either pay out of pocket or file an appeal. Similar systems exist in other countries in comparable form, but it is especially widespread and debated in the US.

Prior Authorization as a cost brake and bottleneck

For insurers, the procedure is a tool for cost control. Not every expensive treatment is medically essential. The prior review is meant to prevent costs from arising for unnecessary or overpriced services. In theory, this also protects the community of insured members, since unnecessary expenses would drive up premiums.

In practice, however, the procedure is often a bottleneck. Doctors spend hours filling out forms and waiting for decisions. Some approvals take days or weeks. In urgent cases, this delay can be medically problematic. Studies by the American Medical Association show that a significant portion of denied requests are ultimately approved after appeal – suggesting that many denials are not medically but bureaucratically motivated.

How AI is changing the approval process

Traditionally, an insurance employee manually reviews the request against a set of rules. This person reads the patient file, compares the requested service against internal guidelines, and makes a decision. This takes time and costs staff resources. Large US insurers are therefore increasingly relying on AI systems that automate this review. The model reads structured data from the file and matches it against the approval rules. A decision can thus be made in seconds.

This sounds efficient, but it carries a concrete risk. If an algorithm systematically denies certain diagnoses or types of treatment, it affects many patients at once – without anyone easily noticing the error. In the US, there have been several lawsuits and investigations in which insurers were accused of using AI systems that denied requests from elderly or severely ill patients at disproportionately high rates. The difference from a human error: an algorithm makes the same mistake millions of times, consistently.

Proponents emphasize that well-calibrated models can reduce human error and arbitrariness. An algorithm treats request A and request B according to the same rules – which is not always the case with human caseworkers. The decisive question, therefore, is not whether AI fundamentally judges worse, but who sets the rules by which it judges, and how errors are controlled.

Prior Authorization in the current AI debate

The term is appearing increasingly often in tech and financial news, because it sits at the intersection of AI, regulation, and the healthcare industry. Companies like UnitedHealth Group made headlines in 2023 and 2024 after reports emerged about automated mass denials in the prior authorization process. The US Congress and the regulatory agency CMS (Centers for Medicare & Medicaid Services) are working on rules meant to mandate transparency and deadlines for decisions.

For startups, the field is at the same time a growth market. Companies are developing software that helps doctors submit requests faster and more completely – which is meant to lower the denial rate. Other tools help insurers review requests more in line with the rules. The term Prior Authorization is thus not just a medical administrative concept, but a test case for what happens when AI systems directly decide over access to medical care.