l
By Sanjay Subramanian, Healthcare, Payer Business Unit Leader, Cognizant
LinkedIn: Sanjay Subramanian
LinkedIn: Cognizant
Some of the most expensive work in healthcare is the work no one counts as care. Physicians and their teams chase authorization status, resubmit records, clarify denial reasons, and explain coverage confusion to patients. Recent research from the American Medical Association (AMA) shows that physicians complete an average of 39 prior authorizations each week, and 93% say prior authorization delays access to necessary care. Much of that burden reflects the administrative labor that payer processes have pushed into the practice, even though none of it helps physicians decide what care a patient needs.
Payer AI has the potential to reduce this burden by identifying documentation gaps earlier and clarifying coverage requirements before care is delayed. But physicians are already approaching it with caution. The AMA also found that 61% of physicians are concerned that health plans’ use of AI is actually increasing prior authorization denials and adding waste. As payers scale AI, the test for success will be whether physicians get time back to focus on patient care. Otherwise, AI risks becoming one more layer on top of a process that already asks too much of the practice. For payers, this starts with three practical changes.
Prevent avoidable rework
The first job for payer AI is to stop avoidable work before it reaches the physician’s staff. Prior authorization is often where the problem starts. A request is submitted, then paused because the payer needs more information. The practice receives a generic note, searches the record, resubmits documentation, and waits again. When a request sits in that documentation loop, each round of back-and-forth can delay care. Cognizant’s Voice of the Member survey points to the same friction on the member side: nearly half of health plan members (49%) report transparency and efficiency issues tied to prior authorizations and referrals, and 25% waited more than two weeks for a primary care or specialist appointment. Those handoffs help explain why even small process failures carry an excessive cost for practices.
AI can and should change that sequence. Payers can use it to check the request against coverage and clinical requirements before it enters review. When documentation is missing, the practice needs to know what is required while the request is still being prepared. When evidence already exists in the record, the system needs to find it. When more detail is needed, the request has to name the clinical evidence required, not ask for “additional information.” Done well, payer AI keeps avoidable work out of the practice instead of processing it after the fact.
Understand where inconsistencies are creating more labor
CMS-0057-F is pushing payers toward more standardized prior authorization and data exchange, including API requirements that generally begin in 2027. Physicians will judge standardization by whether it leads to clearer, more consistent answers from health plans. A prior authorization may look complete, yet the claim can still be denied later. A member service representative may explain coverage one way, while utilization management or appeals use another standard.
This is one of the clearest opportunities I see for payer transformation. To reduce this administrative burden, payers need a single governed source of decision logic across authorization, claims, appeals, provider relations and member services. AI can help by checking each request against the same benefit rules, clinical policies and documentation requirements before a decision is communicated. The answer given during authorization should match the logic used for the claim. The denial reason sent to the provider should match the reason used in an appeal. The work is less about adding AI to a single step and more about connecting decisions across the operating model, so physicians are not forced to absorb the cost of payer misalignment.
Measure the burden
Payers often measure AI by processing time, cost per claim or manual touches avoided. Those numbers help the business, but they do not show whether physicians spend less time chasing answers or if patients reach the next step in their care sooner.
Healthcare AI should be held to a broader scorecard. A 2025 JAMA Network Open study of ambient AI scribes measured burnout, cognitive task load, after-hours documentation, attention on patients and urgent access to care, not only documentation speed.
Payer AI needs the same discipline. A physician-centered scorecard would track how long authorization takes from the practice’s perspective, how many times staff touch the same case, and how often a patient’s next step is delayed because the payer’s answer is unclear. It would also separate administrative denials from clinical denials, so leaders can see whether AI is reducing avoidable friction or only making internal processes faster.
In healthcare, AI is already helping payers move parts of the process faster. For payer leaders, that progress will mean far more if it also takes work out of the physician’s office and gives patients a clearer path to care. Payers that make this the standard will be better partners to both.