Smarter Prior Authorization: Using AI to Improve Care Access and Cut Costs

In this interview, Brian Covino, MD, and Siva Namasivayam of Cohere Health discuss how AI is revolutionizing the prior authorization process by dramatically improving speed, reducing administrative burden, and enhancing care accuracy—while keeping clinicians in the loop and addressing provider concerns.

Please introduce yourself by stating your name, title, and any relevant experience you’d like to share.

Brian Covino, MD: I'm Brian Covino. I'm an orthopedic surgeon. I practiced orthopedic surgery for 28 years. I've been with Cohere Health as chief medical officer for 5 years.

Siva Namasivayam: I am Siva Namasivayam CEO and co-founder of Cohere Health. I've been with the company since the beginning.

How are health plans currently leveraging AI in the prior authorization process?

Dr Covino: Health plans are actually slow to utilize AI in the prior authorization process. That could be for multiple reasons. They turn to other companies, like us, to help them with that initiative and use it in a responsible way.

Namasivayam: AI can be used for approving prior authorization requests. One of the issues today is that there is a lot of paperwork involved and a lot of back and forth that happens any time a prior authorization request is submitted. The decisions do take some time. At Cohere, we ask the provider to submit the medical record, other information about the patient, and the service that's required. This is unstructured data. This is a lot of written notes and things like that. That's where AI comes in. AI is able to interpret the unstructured data and get the information out of it.

If the request can be approved, based on the AI algorithms, it can be immediately approved within 10 or 15 seconds. However, at Cohere, if the requests require the judgment of a clinician, it goes to our nurses and clinicians in the same specialty. Clinicians are the only ones who can make a determination about whether something needs to be denied or changed. The clinician is always in the loop, and they are the ones that make those decisions. The other approvals can be done very fast using AI.

Dr Covino: Just to level set—so you understand where the industry has been and where the industry still is, in a large place—other than the process that Siva just talked about, a lot of times it's a purely manual review. That means a nurse or physician has to actually read through all the documents. That is a slow process, which results in delays. It's a significant administrative expense for the health plan or the delegate that the health plan uses. There's probably about a 10% error rate, if you look at the literature on that.

The second method that some plans and vendors use is to ask a series of questions. They ask the doctor's staff to answer a series of questions. Answering questions is what takes the staff the longest in completing the prior authorization request. Probably about 30% of the time, the answers to the questions don't match what's in the medical records. So, there can be a high degree of inaccuracy.

By using clinician-trained machine learning models, it's much faster to approve appropriate care, like Siva said, in real time. The cost of doing it is much less if you take the people out of it, other than the clinicians who are training the models. The error rate is much less, probably in the low single digits.

In what ways have you seen AI improve or complicate the efficiency and accuracy of prior authorization decisions?

Dr Covino: If you look at efficiency, I was just talking about speeding up the time to appropriate care, where 80 to 85% of appropriate care can be approved in real time. What we've seen is that when we survey the providers and their staff who use our AI systems, they tell us that up to 80% of the time, they can schedule their patients faster for that appropriate care. Again, reducing the delays.

I talked about increased accuracy, that's been proven as well. For the end user who is submitting the requests, the time to submit the request is much less. We've done timestamp studies to see, and we've seen AI cut down their administrative expense by about 40% to 45% in terms of the submission they request. Across the board, accuracy, efficiency, and decreased administrative expense and time for the requester have all improved.

How can payers proactively address physician concerns around increased denials and patient harm?

Namasivayam: First and foremost, there is quite a bit of transparency that's involved. One of the areas where things get complicated is the policy. Today, many of the policies, such as proprietary policies, have quite a bit of ambiguity. That's where misinterpretation or judgment issues can arise.

I don't think that the physicians or the payers have a problem in terms of diagnosing what is wrong with the patient. It's more of, what is this policy saying? The policies are not very clear, and there is quite a bit of ambiguity there. That is one of the primary areas of concern. If things can be really clear, then both sides are more likely to agree on it.

Dr Covino: On the policy front, when the American Medical Association (AMA) surveyed physicians on prior authorization, over one-third of them said that they thought that the policy was not evidence-based.

As Siva was saying, what we do is align policies with what the medical specialty societies put out as their appropriate use criteria or their clinical recommendations, so that it's hard for somebody to say it's not evidence-based if the society is what is backing it. That's one way to manage that.

We've seen decreased denials using our system. The reason is because, in the workflow, we are guiding the user to try to get to the appropriate "yes." It can be as simple as once they submit medical records, if they've missed something—say, they've missed an x-ray report—machine learning can read it and say, "By the way, we don't see the x-ray report. We need that to make a decision." A lot of these administrative denials that are because of a lack of information can be guided by machine learning to the right information.

Physicians and the people they work for are not exactly sure what to order sometimes. Those suggestions can be helpful for getting appropriate care. All of that reduces the physician concerns and lessens the number of denials.

Again, the whole goal in the end is to not do patient harm, but to increase the quality of care, increase the access to appropriate care, and make it much faster.

How can the industry promote better experiences and fairness for providers and payers in the absence of federal regulation?

Dr Covino: There are some regulations coming in 2026 and 2027 around interoperability and transparency. The goal is to make this a better experience. At the end of the day, the majority of physicians do a great job taking care of their patients. What we felt we needed to do was develop a system where we could make it easy for them to get approvals, get out of their way, and let them take care of their patients, and then just focus on the smaller percentage that need guidance in terms of getting to the right answer.

Namasivayam: The costs in this country are very high. It continues to go up every year. In order to bring down the cost for the whole population, this is one of the processes that can be employed. But I think that there has to be a little bit more education on it.

Having said that, if the process can be really fast, transparent, and everyone involved feels like they have a clear understanding of why a decision was made, then there will be a much better dialogue.

Dr Covino: We won't need federal regulations if everybody does understand that there's a method to do this in an approved fashion. But all physicians have to understand that, unfortunately, unnecessary care is still being delivered delivered in this country. It amounts to hundreds of billiions of dollars, and some of that could be controlled.

If we can improve this process and continue to control for unwanted variations and costs, there is a path forward that does not require more federal regulation.