The New AI Arms Race in Healthcare: How Payers Are Setting the Pace
Plus a Brief Technical Primer for Healthcare Leader
In my last article for RCR|HUB, I explored how AI is beginning to transform the work of healthcare teams. I focused on practical areas RCM leaders can address today—and I’ll continue that thread here by unpacking a few core trends in AI technology and highlighting how payers are already putting this tech to work at scale.
This time, we’ll go one layer deeper: offering a brief technical primer, and connecting it to what’s already shifting in provider-payer relationships. Because while it’s tempting to focus on chatbots and co-pilots, the real story is what’s happening behind the scenes.
Riding the Wave
We’re in an AI arms race—and believe it or not, that’s a good thing for you.
Healthcare vendors building their own models might make headlines, but the smarter play is to build flexibly and ride the wave of what’s now widely available. Here's why:
Lower Costs, Bigger Upside
Model training costs have dropped 70–80% in the last two years.
Operational costs per query are down 100x—from $0.01 to ~$0.0001.
92% of enterprises have adopted a "multi-model" strategy to avoid lock-in and tap into newer, cheaper small language models (SLMs).
Expanded Capabilities
AI isn’t just about chatting anymore. Models can now function as agents—tools that can reason, plan, and take action within workflows.
A few quick definitions:
Agent: An AI model that can take action—click buttons, submit forms, or route claims—without human prompting.
Tool Use: Agents can call APIs or interact with software like an assistant might. Thanks to protocols like MCP, they can do this securely—just like early internet protocols opened the door for web apps and eventually mobile to flourish.
Reasoning: The model can break down a task, consider context, and plan next steps—more than just autocomplete.
Autonomy: Some agents can complete entire tasks independently, others need light supervision. You can tune this.
The takeaway: These tools are getting cheaper, faster, and more powerful—and you don’t need to build them yourself. You just need the right foundation to take advantage of them.
The Payer Has Already Upgraded—And They're Setting the Pace
While providers are still evaluating AI pilots, payers are already live. Insurers like UnitedHealthcare, Cigna, and Aetna are using AI agents to automate decisions in claims and prior authorization workflows—at scale.
These aren't back-office experiments. They’re fully operational systems designed to reduce processing time and cut costs.
Denials Are Increasing—And AI Is a Major Factor
The data is hard to ignore:
UnitedHealthcare denies up to 33% of claims
Cigna and Aetna deny between 17–30%
The American Hospital Association reports a 20% rise in denials industry-wide over five years
AI models flag and reject claims rapidly—but without clinical nuance. That means uncommon or complex cases, even when justified, are more likely to be denied.
The Human Cost of Algorithmic Denials
This shows up everywhere:
61% of physicians say AI tools are denying necessary care
82% report prior auth delays lead to treatment abandonment
AI-driven denials now occur 16x more often than human-reviewed ones
This isn’t just annoying. It’s harmful. It slows teams down, burdens staff, and frustrates patients.
What This Means for RCM Leaders
If payers are optimizing with AI agents, and your team is still reliant on inboxes and manual workflows, the imbalance only grows.
The strategy here isn’t just about catching up—it’s about positioning for what’s next.
Key takeaways:
Automate proactively: Use AI to support denial triage, appeals, and pre-submission review.
Evaluate defensively: Ask vendors how their models work, how often they update them, and whether you can swap them out.
Stay engaged: Advocacy matters. AMA and AHA are pushing for transparency—and your voice helps drive momentum.
Don’t Fall Behind—But Step Forward Intentionally
Here’s the good news: these tools are being commoditized for you just as they are for the payers. What once required a team of machine learning engineers can now be configured, embedded, or licensed—and used strategically.
Yes, payers are doing things they’ve long wanted to do: cut labor costs, reduce payouts, scale decision-making. But these trends are still young—just 1–2 years old.
The real differentiator now is nimbleness.
If you can design your workflows and infrastructure to swap tools easily, iterate fast, and apply pressure where it matters—you won’t just survive this shift, you’ll come out ahead.
And no, you don’t need to chase every shiny AI tool. You just need systems that are adaptable, interoperable, and resilient. Because the next wave is already here—and the payer is riding it.