Latest / Becker’s Payer Issues Podcast / Smarter, Faster, Fairer? The Next Evolution of Prior Authorization

Transcript

Smarter, Faster, Fairer? The Next Evolution of Prior Authorization · Becker’s Payer Issues Podcast

• 0:00
  Hi, everyone. This is Lucas Voss with Becker's
• 0:02
  Healthcare. Thanks so much for tuning in to
• 0:03
  the Becker's Healthcare podcast series. Today, we're talking
• 0:06
  about the next evolution of prior authorization. And
• 0:10
  joining me for today's discussion, very excited to
• 0:12
  have her, is Elizabeth Crowley. She's the vice
• 0:14
  president for clinical and care management solutions at
• 0:17
  EXL.
• 0:18
  Liz has more than thirty years of industry
• 0:20
  expertise in health care and various leadership and
• 0:22
  executive roles in the payer consulting and delegated
• 0:24
  services market. Market. She's focused on leading health
• 0:26
  care operations and business process optimization in highly
• 0:29
  complex and matrixed organizations
• 0:31
  to bridge the gap between clinical technology and
• 0:34
  business objectives. Liz, thanks so much for being
• 0:37
  here today.
• 0:38
  Lucas, thank you so much for having me
• 0:40
  on your podcast. I'm really excited today
• 0:43
  to dig into
• 0:44
  our topic and,
• 0:46
  have a great conversation.
• 0:47
  Yes. We do have a lot to dig
• 0:50
  into. There's a lot to talk about here,
• 0:53
  specifically with prior authorizations.
• 0:56
  Again, it's been traditionally framed sort of as
• 0:58
  this administrative burden,
• 1:00
  but then we're seeing that shift to increasingly
• 1:02
  being discussed as part of the member experience,
• 1:04
  provider relations,
• 1:06
  and certainly compliance issues, and it it's happening
• 1:09
  all at once. Right?
• 1:11
  From your perspective,
• 1:13
  what's fundamentally broken into today's prior auth model,
• 1:17
  and what's been missing most specifically from these
• 1:20
  conversations
• 1:20
  around how to fix all of this?
• 1:23
  Yeah. You know, fundamentally,
• 1:25
  the prior authorization
• 1:26
  model, it's totally viewed as a gatekeeping, right,
• 1:29
  or limiting access
• 1:31
  to
• 1:32
  care by the provider community and by members.
• 1:34
  However,
• 1:35
  you know, prior auth was instituted
• 1:38
  as part of the larger managed care program
• 1:40
  evolution over the last, you know, twenty, twenty
• 1:43
  five years.
• 1:44
  So over time,
• 1:46
  the the process has been burdened
• 1:49
  administratively
• 1:50
  due to the increasing in regulatory
• 1:52
  pressures,
• 1:53
  compliance and quality requirements,
• 1:55
  and just the scope of services,
• 1:57
  new therapies, new medications
• 2:00
  that have, you know, become part of mainstream
• 2:04
  clinical care delivery,
• 2:06
  has just made prior auth very, very complex.
• 2:10
  There's lots of
• 2:12
  different things the payer community has been trying
• 2:15
  to do to make this prior authorization process
• 2:18
  easier,
• 2:19
  but, traditionally,
• 2:21
  they've been kind of like gold carding programs.
• 2:24
  Right? So giving certain providers certain services a
• 2:27
  free pass, right, for so many services or
• 2:29
  for certain types of members or
• 2:32
  other
• 2:33
  rule based,
• 2:35
  solutions where,
• 2:37
  you know, low,
• 2:39
  cost and,
• 2:41
  low abused services, for a lack of a
• 2:43
  better word, were kinda given an auto approval
• 2:46
  pass. You know? But that didn't really remove
• 2:48
  the burden of prior authorization. Right? The provider
• 2:51
  still had to go through the motions of
• 2:53
  submitting
• 2:54
  prior authorization and things of that nature. So,
• 2:57
  you know, fundamentally,
• 2:58
  that's
• 2:59
  that's the problem. Right? So but over the
• 3:01
  past twelve to twenty four months, there's been
• 3:04
  this heightened attention, right, to the whole end
• 3:06
  to end experience in prior authorization and
• 3:10
  how to make it better.
• 3:13
  Uncovering the root causes needs to be the
• 3:16
  top priority, so not just making it easier
• 3:19
  to follow the process,
• 3:21
  but is the process itself broken? So things
• 3:24
  like the FHIR interoperability
• 3:27
  standards of electronic
• 3:28
  data sharing, that has kinda spring boarded us
• 3:31
  into
• 3:32
  more robust automation,
• 3:34
  AI forward automation, you know, to now that
• 3:37
  we have the data connection, how do we
• 3:38
  leverage that and really make the experience more
• 3:41
  beneficial?
• 3:43
  You mentioned the end to end aspect of
• 3:45
  this, which is really crucial as you just
• 3:47
  highlighted.
• 3:48
  And there is a very growing shift toward
• 3:51
  AI first work flow, specifically this year. There
• 3:54
  was some last year, but certainly in 2026.
• 3:57
  How do you distinguish between
• 4:00
  basic automation and these AI agents that are
• 4:02
  coming up that change how clinical and care
• 4:05
  management teams work
• 4:06
  day to day?
• 4:07
  Yeah. So,
• 4:09
  you know, automation
• 4:10
  has typically been characterized as, like, RPA. Right?
• 4:14
  Robotic process automation and workflow business rules.
• 4:18
  These two approaches, they're heavily defined but with
• 4:22
  scripts and logic. Right? And that logic and
• 4:24
  those scripts have to constantly be updated
• 4:27
  with change in program execution,
• 4:30
  change in criteria sets, or change in requirements,
• 4:34
  regulatory requirements.
• 4:36
  With agentic
• 4:38
  forward
• 4:39
  automation,
• 4:41
  the agentic AI agents actually
• 4:43
  interpret requirements
• 4:45
  and interpret guidelines
• 4:46
  as a human would.
• 4:48
  So instead of having to,
• 4:51
  constantly maintain
• 4:52
  scripts and algorithms,
• 4:55
  the agentic agent reads source
• 4:57
  materials and source guidance each and every time
• 5:01
  the agent is engaged
• 5:03
  to support a a prior auth process.
• 5:06
  So so this is a couple of things.
• 5:08
  Right? It
• 5:09
  significantly reduces
• 5:12
  the administrative burden to the payer of having
• 5:14
  to maintain scripts and logic. Right? So they
• 5:17
  can then turn their attention in looking at
• 5:19
  the outcomes of the AI, looking at the
• 5:22
  data that is being
• 5:24
  harnessed through the end to end workflow.
• 5:27
  It creates the opportunity
• 5:28
  to continuously
• 5:29
  capture and audit and evaluate
• 5:32
  the accuracy of decisions
• 5:34
  and the time required to process each step
• 5:37
  of the workflow.
• 5:38
  Right? So this creates more information for leaders
• 5:41
  to manage and process
• 5:43
  in real time.
• 5:44
  Now you've touched on the advantages just now
• 5:46
  too. Is there one specific piece of this,
• 5:49
  one specific workflow
• 5:51
  that comes to mind for you where this
• 5:53
  just is a no brainer, where this is
• 5:56
  illustrated perfectly?
• 5:57
  Yeah. Yeah. So when you think about
• 6:00
  how clinical decision support or medical necessity reviews
• 6:04
  are done,
• 6:06
  rule based automation,
• 6:08
  for that step of the prior authorization process
• 6:11
  requires that there needs to be a documented
• 6:14
  definition
• 6:16
  every single step of that criteria statement. What
• 6:19
  does it mean to meet that criteria? What
• 6:21
  does it mean not to meet that criteria?
• 6:24
  Aside from that, then there's the complexity of
• 6:27
  all of the following must be demonstrated
• 6:30
  or any
• 6:31
  two or more of the following need to
• 6:34
  be demonstrated.
• 6:35
  And then even more complicated is if there
• 6:39
  are time sequence parameters
• 6:41
  throughout the medical evidence: six months' worth of
• 6:44
  therapy,
• 6:45
  two failed courses of treatment of a medication,
• 6:49
  makes it very difficult to quantify that in
• 6:52
  logic and definition.
• 6:55
  The AI agent, as I I shared before,
• 6:58
  reads the medical policy guidance or the CMS
• 7:01
  criteria bulletin, whatever is being used as that
• 7:04
  source of
• 7:05
  medical necessity
• 7:06
  criteria, and then reads the evidence submitted by
• 7:09
  the provider and then makes those matches
• 7:13
  just as a human would.
• 7:15
  But the AI agent does a little bit
• 7:17
  more than what a human typically does. A
• 7:19
  human will find what is necessary to move
• 7:23
  down the algorithm.
• 7:24
  Right? One of the following. As soon as
• 7:26
  one is found, typically,
• 7:28
  a human reviewer will go to the next
• 7:30
  step, and and, Agentic AI Forward workflow will
• 7:34
  find every piece of evidence. So it makes
• 7:37
  the review more robust. It makes the review
• 7:39
  definitely more member centric and more appropriate to
• 7:43
  that member in that situation.
• 7:46
  And then because the agent reads from the
• 7:48
  source,
• 7:50
  if there's a change to a policy, if
• 7:52
  CMS updates a a policy bulletin,
• 7:55
  the agent is reading that real time. So
• 7:57
  the decision tree that was in place at
• 7:59
  the last iteration of that workflow
• 8:02
  is validated by that agent reading that source
• 8:05
  document again and updating it if necessary if
• 8:08
  there's a change to criteria.
• 8:10
  Now
• 8:12
  with the technology itself, we're having a lot
• 8:15
  of conversations about governance,
• 8:18
  about checks and balances. Right? You just talked
• 8:20
  about the human in the loop, which is
• 8:22
  certainly really, really crucial. Right? And certainly one
• 8:25
  concern for payer leaders specifically is how to
• 8:28
  balance that, right, with speed that you just
• 8:30
  mentioned really, really quickly, really, really accurately
• 8:33
  with clinical rigor when AI is involved in
• 8:36
  that decision support. Right?
• 8:37
  What have you seen work well in maintaining
• 8:40
  that clinician trust, that that oversight, and certainly
• 8:43
  also
• 8:44
  accountability
• 8:46
  while still, again, improving the efficiencies that you've
• 8:49
  just talked about?
• 8:51
  Yeah. I think whenever an AI forward workflow
• 8:54
  is
• 8:55
  being implemented,
• 8:56
  it's really important to
• 8:58
  be very focused and purposeful
• 9:01
  about where you want that,
• 9:03
  agentic solution to start. So it's really a
• 9:07
  a a small segment of the overall prior
• 9:10
  offline. So maybe it's a single service type
• 9:14
  or
• 9:15
  pie you know, putting it in place with
• 9:16
  one provider group and all of their requests.
• 9:19
  This allows clinicians and leaders to evaluate the
• 9:22
  accuracy and the effectiveness
• 9:24
  of the AI workflow outcome
• 9:27
  in a more timely manner. Right? So you
• 9:29
  could imagine a payer that gets thousands of
• 9:31
  transaction
• 9:34
  possibly
• 9:35
  quality
• 9:36
  review the AI output of every single one
• 9:38
  of those transactions
• 9:40
  in early implementation
• 9:41
  to gain that confidence and trust, right, that
• 9:44
  the AI agents are delivering
• 9:46
  clinically solid, consistent
• 9:48
  recommendations.
• 9:50
  So starting small allows there to be almost
• 9:52
  a side by side or one to one
• 9:54
  and watching that AI evidence.
• 9:57
  Then
• 9:58
  it's able to be scaled. Once the clinicians
• 10:01
  have trust, the AI,
• 10:03
  agent is accurate, consistent, and
• 10:07
  is saving them time,
• 10:09
  then we scale. However,
• 10:11
  most importantly, though, is the oversight doesn't stop.
• 10:14
  The program monitoring and the analytics to monitor
• 10:17
  the consistency and accuracy has to stay in
• 10:19
  place,
• 10:20
  and there's reason for that. Right? Brand new
• 10:23
  policies like anything, whether it's a a group
• 10:26
  of clinicians
• 10:27
  reviewing against a new policy
• 10:29
  for the first time,
• 10:31
  or an AI agent, a Genentech agent reviewing
• 10:34
  against a policy for the first time. It
• 10:36
  needs deep oversight.
• 10:39
  So
• 10:39
  that, component needs to be consistent. And I
• 10:42
  I think sometimes that is a misstep,
• 10:45
  that as soon as we see a process
• 10:46
  or a technology driven solution work,
• 10:50
  there's sometimes too much trust in the technology,
• 10:53
  so the oversight is critical.
• 10:55
  Also,
• 10:57
  this is all evidence based output. Right? So
• 11:00
  the decisions and the recommendations
• 11:02
  are being captured by that agentic workflow,
• 11:05
  and this really creates opportunity
• 11:08
  for
• 11:09
  evidence based rigor and decision making
• 11:12
  in the workflow and can really help reduce
• 11:14
  and educate human bias.
• 11:17
  So to be able to enable that then,
• 11:20
  right,
• 11:21
  looking ahead,
• 11:23
  what are those things from an investment perspective
• 11:25
  and from a focus perspective
• 11:28
  that payers should really ensure is actually there
• 11:32
  and is met for the future? What are
• 11:35
  those things that need to be in place
• 11:36
  right now? Yeah. So I'm really not even
• 11:39
  gonna talk about the technology
• 11:40
  components, right, because
• 11:42
  those are defined. I think, one of the
• 11:45
  biggest things that
• 11:46
  I found to be a miss,
• 11:50
  and that causes delay and a a lot
• 11:52
  of uptime in trying to get these, agentic
• 11:55
  workflows in place is,
• 11:57
  a defined change management and oversight process. Right?
• 12:00
  So when we're bringing AI into the workflow,
• 12:04
  that is materially
• 12:05
  changing
• 12:06
  the day in the life
• 12:07
  of all the humans in the loop, whether
• 12:09
  they are support staff, nurses,
• 12:12
  physicians. It it doesn't matter. So
• 12:15
  the the change management aspect is critical,
• 12:19
  to ensure that there is adoption,
• 12:22
  process adherence, and that being able to evaluate
• 12:26
  process effectiveness.
• 12:28
  So so to to me, that is
• 12:31
  quintessentially
• 12:32
  important.
• 12:33
  It also allows
• 12:35
  for the human in the loop to more
• 12:37
  effectively
• 12:38
  gain trust when they understand
• 12:41
  what their new day in the life is
• 12:43
  gonna be like, why it is important to
• 12:45
  embrace it, and demonstrating for them the value
• 12:48
  that bringing technology and agentic support into their
• 12:52
  workflow will give them.
• 12:54
  And as with anything else, the the end
• 12:57
  users can then start to rely on the
• 13:00
  data that is being presented to them
• 13:03
  and kind of change the way they
• 13:06
  execute their responsibilities
• 13:08
  instead of having to build the case example
• 13:11
  and then making the recommendation.
• 13:14
  They can re case example and then validate
• 13:17
  a recommended
• 13:18
  result and modify that as necessary.
• 13:22
  Now the end goal, obviously,
• 13:24
  if and when this is all happening is
• 13:26
  the fact that this needs to be scalable
• 13:28
  across an enterprise. This needs to be scalable
• 13:31
  for a large number of people, for multiple
• 13:33
  business units, etcetera, whoever is working in those
• 13:36
  processes. Right? However,
• 13:38
  I like to say we've been living in
• 13:39
  the year of point solutions in 2024 and
• 13:42
  2025.
• 13:43
  Yep. Many organizations
• 13:45
  struggle to move beyond those point solutions in
• 13:49
  prior auth, specifically,
• 13:50
  especially when there is legacy platforms evolved, etcetera.
• 13:54
  Right?
• 13:55
  What does it take to really scale those
• 13:57
  AI driven decision supports
• 13:59
  across an enterprise?
• 14:01
  Yeah. So I think it's really
• 14:03
  it all lies in
• 14:05
  the fundamental development architecture
• 14:07
  of that AI forward workflow.
• 14:11
  So when we think about
• 14:12
  how to bring AI capabilities across multiple utilization
• 14:17
  management workflows.
• 14:19
  It's really the agentic agent approach is the
• 14:22
  most nimble and the most flexible. So, for
• 14:25
  example,
• 14:26
  you can create a single
• 14:28
  AI agent
• 14:30
  that handles
• 14:31
  a prior auth workflow end to end that
• 14:33
  can be integrated in and around any
• 14:37
  product or platform workflow. Right? But it's still
• 14:40
  a single agent.
• 14:42
  So that agent only knows that workflow. It's
• 14:44
  trained on that work flow. But if you
• 14:46
  think about deconstructing
• 14:48
  the end to end of prior auth, let's
• 14:50
  talk about things like a medical record agent
• 14:53
  or a clinical policy agent and a decision
• 14:56
  support agent. So when we think about those
• 14:58
  three key steps,
• 15:00
  the agent is trained to read medical policies
• 15:03
  with extreme accuracy.
• 15:05
  Another agent is trained to extract
• 15:08
  the key relevant clinical data in submitted
• 15:13
  clinical records
• 15:14
  to align
• 15:15
  to that medical policy, and that's what that
• 15:17
  decision support agent does is orchestrating that decision
• 15:20
  tree.
• 15:21
  Because they're independent agents finely tuned on a
• 15:25
  specific skill or step,
• 15:28
  anywhere in the organization that that same function,
• 15:32
  occurs, that agent can be deployed. So when
• 15:35
  you think of utilization management,
• 15:37
  it can pry off we've been talking about.
• 15:40
  Appeals and grievances,
• 15:42
  same set of agents, but just doing a
• 15:44
  different type
• 15:46
  of medical necessity review. But the steps of
• 15:48
  reviewing a record, looking at criteria,
• 15:51
  and supporting that decision
• 15:53
  are there.
• 15:54
  Medical claim review. Right? When claims pen for
• 15:57
  medical necessity review and clinical is needed, same
• 16:01
  set of steps.
• 16:02
  So by investing
• 16:04
  in foundational
• 16:06
  agents
• 16:07
  that are,
• 16:09
  tied to a single executable task.
• 16:11
  That agent can be deployed across multiple systems
• 16:14
  and multiple workflows within the organization.
• 16:18
  Maybe we finally reached the year of the
• 16:20
  agent at this point in time. That might
• 16:22
  be it. Liz, it's so great to have
• 16:23
  you on. So many great nuggets in this
• 16:25
  conversation. Anything else that you'd like to touch
• 16:27
  on that that we haven't mentioned for our
• 16:29
  audience?
• 16:30
  You know, I think the only other thing
• 16:31
  I'd like to talk about is we've talked
• 16:33
  a lot about work flows and AI in
• 16:35
  the workflow,
• 16:36
  but we really have been spent time on
• 16:38
  what is truly required to allow
• 16:41
  any AI forward workflow to execute accurately, and
• 16:45
  that's data. Right? So analyzing the accessibility
• 16:49
  of the data
• 16:50
  and an organization
• 16:52
  to having a data, what we call AI
• 16:54
  ready,
• 16:55
  is a critical first step, right, to any
• 16:58
  AI forward
• 16:59
  workflow effort. So
• 17:01
  being able to convert structured and unstructured data
• 17:04
  means faster retrieval for analytics and operational integration.
• 17:09
  So really the first step. You know, we
• 17:12
  technology can do anything, but that that foundational
• 17:16
  information, which is data. It won't execute as
• 17:19
  effectively as reliably
• 17:20
  and produce the outcomes necessary to be worthwhile.
• 17:25
  You heard it here first. Next podcast with
• 17:27
  Liz on data. You'll have to make sure
• 17:30
  to tune in. Liz, thank you so much
• 17:32
  for being here. It's great to have you.
• 17:34
  Thanks so much, Lucas.
• 17:36
  We also want to thank our podcast sponsor,
• 17:38
  eXol. You can tune in to more podcasts
• 17:40
  from Becker's Health Care by visiting our podcast
• 17:42
  page at beckershospitalreview.com.