A recently approved medication has cleared the FDA, prompting a health plan to make crucial decisions about its coverage. This choice sets off an intricate process involving six or seven specialists, often unfolding over two to three months, with final costs hovering around $100,000. Meanwhile, patients are left in limbo.
For some medications, this lag may be simply inconvenient, but when it comes to treatments for schizophrenia, the stakes are significantly higher. Delayed treatment can result in hospitalizations, each costing health plans between $8,000 and $15,000. Aggregate this across a sizable health plan, and you could see expenses balloon to between four and seven million dollars due to one slow-moving drug decision.
Having spent eleven years at Dell, followed by time in startups, I have witnessed how organizations procure technology designed to resolve these types of inefficiencies, yet seldom succeed.
The strains of the COVID pandemic prompted reflection, leading me to reassess my impact in the world. I began asking myself what legacy I would leave for my grandchildren: Would I tell them I helped people navigate AI for shopping and searching, or would I be able to say I made a difference in healthcare? This introspection became even more poignant after losing a friend to breast cancer and hearing their struggles in sifting through extensive medical records to find experimental treatments.
The healthcare sector is not short on expertise; rather, it grapples with the challenge of ensuring that the right expertise reaches the right individuals precisely when it’s needed—a persistent issue that has plagued enterprises for years.
Currently, the discourse surrounding AI centers around deploying agents—discussing how many can be allocated, their level of autonomy, and their speed. However, there is a notable lack of focus on the systems that manage these agents. This is the crucial aspect that genuinely matters, yet it remains largely overlooked.
Analysts have already made predictions concerning the trajectory of AI projects. Gartner forecasts that over 40% of projects involving agentic AI will be abandoned by the end of 2027, attributing this to insufficient risk management, alongside high costs and ambiguous value propositions.
This is not a new scenario. When cloud computing emerged, the promise was transformational change; instead, many organizations simply transferred their existing workflows into the cloud without re-evaluating their efficiency. What resulted was a similar process with a larger price tag, and AI appears to be heading toward a similar fate. Companies are fragmenting automation efforts—optimizing one task at a time—without addressing fundamental issues. In such cases, expediting an inefficient process will only yield quicker mistakes.
Healthcare exemplifies why AI faces unique challenges while simultaneously being an area ripe for exploration. Stringent regulations, significant decision-making implications, fragmented data systems, and an ongoing shortage of qualified clinicians all complicate matters. The Association of American Medical Colleges (AAMC) forecasts a shortage of up to 86,000 physicians by 2036, with nursing staff stretched equally thin.
One healthcare organization we collaborated with has 600 nurses dedicated to prior authorization and payment accuracy. These professionals, trained to care for patients, spend their days buried in paperwork instead. Although technology was expected to alleviate this burden, most merely digitized existing filing systems. According to the American Medical Association’s (AMA) 2024 physician survey, the process of prior authorization can take an average of 13 hours of physicians’ and staff members’ time each week, with 93% of doctors noting that it delays patient care.
In essence, the healthcare sector highlights the issues faced by various industries. Financial services, insurance, government, and energy are similarly caught in a web of trapped expertise, strict regulations, and a need for transparent decision-making. The pressing question is not whether AI can perform certain tasks, but if organizations can sufficiently understand, govern, and put trust in the decisions that result from AI participation.
Our analysis reveals that for a pharmacy benefit manager or health plan, the approval of a new drug can require a lengthy process, involving six to seven specialists—ranging from pharmacists to compliance specialists—over a period of 60 to 90 days, with costs reaching around $100,000 per approval. A major pharmacy benefits manager conducts anywhere from 200 to 300 of these evaluations annually, often leaving patients in a state of uncertainty.
In contrast, a coordinated AI process can streamline that same assessment, reducing the time to just four to eight hours. In this scenario, a clinical pharmacist oversees the results instead of generating them, effectively eliminating coverage delays and reducing labor costs by 97%. Moreover, every action performed by the agents is properly documented, facilitating accountability during future compliance inquiries—eliminating the need to sift through emails or rely on memory.
The overarching aim should not be mere automation but rather redefining possibilities. Too many AI initiatives fail to create visible change, simply speeding up steps in processes that remain fundamentally unchanged. Deploying numerous uncoordinated agents is reminiscent of the scattered point solutions prevalent in enterprises a decade prior, achieving little in terms of improvement.
Healthcare didn’t choose to be at the forefront of this issue; it has been compelled to confront these challenges sooner due to administrative overload, resulting in diminishing patient care and faltering coverage processes. This system is indeed broken, necessitating urgent intervention. The industry is now faced with a new reality—not only about deploying AI but also about how to govern, audit, and ensure its efficacy for the benefit of those it serves.
Eventually, other regulated sectors will face similar pressures. The statistics from MIT and Gartner indicate that many are already experiencing this. The critical question remains: will they learn from the challenges faced by others, or will they hesitate until they reach their breaking point? My experience demonstrates that the priority should shift from merely adding more AI agents to establishing a robust management system for them, alongside the clarity to redefine employee roles effectively.
