AI doesn't deny care, but shapes review, says expert

The debate over AI's role in insurance denials has centered on whether algorithms make final decisions, but the reality is more subtle. UnitedHealth Group's CFO and others maintain AI doesn't approve or deny, which is technically accurate: the technology doesn't press the deny button. However, as Sutter explains, AI determines which cases are auto-approved as low-risk and which are routed for human review, shaping the outcome before a clinician ever sees the claim. This routing function is where the power lies, and it's what providers are struggling to navigate.

The information asymmetry is the core tension. Payers position their AI knowledge as proprietary, leaving providers to guess what triggers scrutiny. This has led to an 'AI arms race,' where providers invest in tools like ambient documentation and AI scribes to improve clinical detail upfront, trying to avoid denials or delays. Yet this effort may be misdirected without transparency. Sutter's call for payers to disclose approval rates, denial rates, and outcome comparisons between routed and non-routed requests is aimed at leveling the playing field.

The Centers for Medicare and Medicaid Services is even testing AI for Medicare prior authorizations in six states, though it has faced criticism for low approval rates. This context underscores the stakes: the question isn't whether AI belongs in healthcare, but whether it's making care more efficient or denials more efficient. Without transparency, that distinction remains obscured, and providers are left chasing ghosts.