Blues executive says clinicians must make medical-necessity denials, not just review AI’s call
Sean Robbins, Mandy Cohen and Warris Bokhari agreed that faster paperwork is an inadequate goal. Their disagreement centered on whose clinical judgment should determine whether a patient gets care.
A licensed clinician should make a denial based on medical necessity, rather than merely review a decision produced by artificial intelligence, Blue Cross Blue Shield Association executive Sean Robbins said. The distinction drew agreement from other participants in a November 17 discussion of healthcare’s financial harms, but it also exposed a deeper dispute: whether an insurer’s clinician should be making that decision at all in some high-stakes cases.
“A licensed clinician should make the call, not review the call,” Robbins said. He supported using AI to assist that person, while rejecting an AI system as the decision-maker for a medical-necessity denial. Claimable CEO Dr. Warris Bokhari agreed about human judgment, then argued that in oncology the qualified person should be the treating physician.
Chartis chief product and technology officer Sara Vaezy moderated Do No Financial Harm: Fixing Claims Denials at the Root on the Ferrari Stage. Joining Robbins and Bokhari was Dr. Mandy Cohen, a Manatt Health national advisor and former CDC director who also led North Carolina’s health department and served as CMS chief operating officer. Their discussion moved beyond whether AI can process requests faster to whether the underlying payment and benefit rules produce care people can actually obtain.
A human in the loop does not settle whose judgment counts
Robbins, BCBSA’s executive vice president and chief operating officer, described insurance as a promise of protection. Financial harm occurs when people cannot afford needed care, avoid it because of the price or learn the full cost only afterward. He said transparency and understandable explanations have to accompany faster decisions.
Bokhari, a physician and former health-plan executive whose company helps patients appeal denials, challenged the distance between the treating team and an insurer’s reviewer. He argued that describing a decision as one about coverage does not change its clinical consequences for a patient. The panel did not resolve how to allocate that authority across every service or circumstance.
That disagreement matters when a technology proposal promises to keep a human in the loop. A workflow can include a person while still leaving open whether that person has enough evidence, time and authority to make the decision. Robbins’ formulation distinguished actual clinical decision-making from a nominal sign-off on software’s conclusion.
Reduce the transactions that do not need to happen
Robbins said his objective was to “eliminate the need for appeals altogether” by reducing unnecessary prior authorization and getting decisions right initially. He described removing requirements where approval rates were already very high and simplifying processes for providers with a strong record of appropriate requests. His argument began upstream of the appeal letter.
He cited 6.5 million fewer prior authorizations as evidence of progress across the industry, while saying it was not enough. BCBSA and AHIP’s April progress report gives the same figure, based on a survey of participating plans following voluntary commitments announced in June 2025. It is an industry-reported reduction in administrative transactions, not a direct measurement of how much faster each patient received care.
Bokhari pressed for access and outcomes as the starting measures. Removing a requirement can reduce work for clinicians and patients, but a smaller transaction count cannot by itself show that an individual received treatment in time or could afford it. The participants were discussing different points in the same process, from the request to the patient’s ability to act on the answer.
An approval can still leave treatment out of reach
“Having insurance doesn't mean that you have access,” Bokhari said. He described what he called ghost approvals: situations in which a treatment is approved but the patient’s cost remains too high to obtain it. Those cases can disappear from a denial-based measure even though the person still goes without the intended care.
He also described the difficulty of preparing an appeal that identifies the relevant clinical evidence. Knowing that a denial can be challenged is different from knowing what to submit, which data apply and how to make the argument before a delay becomes harmful. His business has an interest in making that work easier, a perspective distinct from Robbins’ emphasis on avoiding unnecessary authorization in the first place.
“We put AI on the side of the patients because they are the unrepresented party,” Bokhari said. That was his case for appeal technology as leverage for people already facing a denial. It did not establish that more automated appeals alone would fix benefit design, affordability or the clinical rules that produced the dispute.
Cohen warns against automating the administrative contest
Cohen argued that powerful AI tools could make healthcare more expensive if payers and providers use them to reinforce opposing administrative processes. “We are automating some of the terrible processes that just increase costs,” she said. Her concern was that each side becomes more capable at defending its position without moving the patient closer to appropriate care.
She linked that problem to payment incentives. Fee-for-service arrangements can reward additional volume, while utilization controls attempt to limit spending. Cohen said low-value care and patients failing to receive care can both exist in that system, making it insufficient to assume that only one party’s workflow needs improvement.
“We can't wait for that to be true before we get to work,” Cohen said of the broader transition to value-based care. She supported immediate improvements while arguing that the industry must also rethink what it pays for. That leaves room for better tools today without treating the current authorization process as the permanent design to optimize.
Benefit design and accountability belong in the same discussion
Cohen extended the argument to out-of-pocket costs for primary and preventive care. A benefit can discourage an earlier, useful intervention through the deductible, then pay for a more expensive problem later. Faster authorization cannot fully repair that incentive because the barrier may be the patient’s bill rather than the speed of the answer.
Vaezy asked the panel to connect structural reform with the work that organizations can do now. Robbins emphasized clear next steps when a denial occurs and timely decisions where care is urgent. Cohen called for government to convene the parties and establish guardrails, while warning against regulating individual interactions so narrowly that opportunities to redesign the overall process are lost.
The panel’s clearest area of agreement was that adding AI to an existing workflow does not make that workflow worth preserving. The disagreement about who should decide medical necessity remained consequential, as did Bokhari’s demand to measure whether approved care is actually received. For buyers of authorization and appeal tools, those are more meaningful questions than how many documents a model can generate or how quickly a decision moves through a queue.