The landscape of American healthcare is undergoing a radical transformation as industry titans pivot toward a future defined by automation and algorithmic precision. Leading this charge is UnitedHealth Group, a company currently navigating a complex path between massive technological investment and a deep-seated public skepticism regarding the role of machines in medical decision-making. Simon Glairy, a distinguished expert in insurance and Insurtech, joins us to unpack the implications of this shift. With his extensive background in risk management and AI-driven assessment, Glairy offers a unique perspective on how the intersection of multi-billion dollar investments and patient care is reshaping the financial and ethical boundaries of the insurance sector.
UnitedHealth has recently committed to an massive $3 billion investment in artificial intelligence over the next two years. From your perspective in risk management, what does a projected 2-to-1 return on investment signal about the future of administrative efficiency in healthcare?
A 2-to-1 return on investment is a powerful signal to Wall Street that the era of manual, paper-heavy processing is finally reaching its breaking point. When a company of this scale identifies a way to turn every dollar of AI spending into two dollars of value, it suggests that the $80 billion currently wasted annually on administrative transactions is ripe for disruption. We are seeing a shift where AI isn’t just a luxury but a fundamental necessity to maintain profitability, especially after the financial volatility the company faced when profits collapsed last year. By automating data-intensive processes like prior authorization, the company is betting that efficiency will provide the cushion needed to stabilize earnings. It is a calculated move to prove that the friction of faxes and phone calls can be replaced by high-speed digital interactions that satisfy both shareholders and the bottom line.
There is a significant push to integrate AI into the daily lives of healthcare workers, from nurses listening to automated chart summaries to bots calling doctor offices. How do these specific applications of “administrative AI” change the fundamental relationship between the insurer and the provider?
The integration of tools like Optum Real, which has already processed about a billion transactions, creates a real-time bridge that has never truly existed between providers and insurers. When you have AI agents calling doctor’s offices to schedule appointments, you are essentially attempting to remove the “grunt work” that usually leads to burnout for medical staff. This level of automation aims to reduce the friction that patients feel when they are caught in the middle of a bureaucratic tug-of-war. However, the emotional weight of this change cannot be ignored; while it might save time for a nurse driving to a patient’s home, it also introduces a layer of machine-driven interaction that can feel cold to those used to human-to-human coordination. The goal is to make the system feel invisible and seamless, but the transition requires a massive cultural shift for both the doctors and the administrative teams involved.
Despite the financial optimism that has seen shares rise by 21% this year, there is a glaring trust deficit, with nearly 70% of the public skeptical of corporate AI. How can a company successfully “earn trust” when its cost-control tactics are often viewed as barriers to necessary care?
Earning trust in this environment is an uphill battle that requires more than just successful pilot programs or slick marketing. The skepticism is rooted in years of frustration with delays and denials, which about half of insured Americans have encountered according to recent polls. To counter this, the company is attempting to be more transparent, even inviting reporters to their Minneapolis headquarters to pull back the curtain on their 117 large language models and 1,000 different AI uses. They are trying to demonstrate that their internal review boards—which include clinicians and medical ethicists—are acting as a safeguard against algorithmic bias. Ultimately, the public will only trust the technology when they stop feeling like the system is designed to say “no,” and that process will take a considerable amount of time to be felt at the patient level.
The controversy surrounding the naviHealth algorithm and the subsequent litigation regarding care denials has become a focal point for critics. In your view, how should a company manage the risk of “algorithmic drift” or unintended consequences when medical outcomes are at stake?
Managing algorithmic risk requires a “boots-on-the-ground” approach to governance where alerts are constantly firing the moment a model begins to behave unexpectedly. The recent federal inspector general report, which found that denials triggered by certain algorithms were almost always overturned on appeal, serves as a stark warning about the dangers of over-relying on automated systems. To mitigate this, the company must maintain a strict boundary where 99% of AI applications remain administrative rather than clinical, avoiding diagnostic AI altogether. It is about creating a feedback loop where human clinicians have the final word, ensuring that the algorithm is a tool for support rather than a replacement for professional judgment. When a company has 20,000 engineers working on these systems, the challenge is ensuring that the “human-in-the-loop” philosophy doesn’t get lost in the drive for $1 billion in annual cost savings.
UnitedHealth has been very clear about its policy of “only approving using AI” and never using it to deny care. What are the operational challenges of maintaining such a strict ethical line while also trying to embed AI into every facet of the business?
The operational challenge lies in the sheer scale of the deployment and the pressure to have employees use AI at least once every single day. When you are tracking daily usage among staff, there is an inherent risk that the speed of the technology might outpace the nuance required for complex medical cases. The company’s leadership has to ensure that the mantra of “never deny using AI” is reflected in the actual code and the training data used for their various models. This requires constant auditing and the courage to pull back a model if its behavior doesn’t align with these ethical guidelines, as they have reportedly done in the past. It’s a delicate balancing act: you want to empower your 20,000 engineers to innovate, but you also have to keep them confined within a framework that prioritizes patient access over the efficiency of a denial.
What is your forecast for the role of AI in the insurance sector over the next five years?
I expect that we will see a “great separation” where companies that can successfully pair AI efficiency with genuine transparency will pull far ahead of those that use technology as a shield for opaque decision-making. Over the next five years, the industry will likely move past the trial phase and into a period where real-time, AI-driven coverage determinations become the standard for all 80 billion administrative transactions. However, this growth will be shadowed by increasingly stringent federal oversight and a public that is much more litigious regarding algorithmic “black boxes.” Success will not be measured just by a 2-to-1 return on investment, but by whether these companies can reduce the percentage of Americans who view their insurer as a barrier to their well-being. The technology is here to stay, but the human oversight will have to become much more sophisticated to keep pace with the $3 billion being poured into these engines.
