1
00:00:00,000 --> 00:00:02,240
Hi, everyone. This is Lucas Voss with Becker's

2
00:00:02,240 --> 00:00:03,839
Healthcare. Thanks so much for tuning in to

3
00:00:03,839 --> 00:00:06,879
the Becker's Healthcare podcast series. Today, we're talking

4
00:00:06,879 --> 00:00:10,160
about the next evolution of prior authorization. And

5
00:00:10,160 --> 00:00:12,244
joining me for today's discussion, very excited to

6
00:00:12,244 --> 00:00:14,644
have her, is Elizabeth Crowley. She's the vice

7
00:00:14,644 --> 00:00:17,285
president for clinical and care management solutions at

8
00:00:17,285 --> 00:00:17,785
EXL.

9
00:00:18,244 --> 00:00:20,004
Liz has more than thirty years of industry

10
00:00:20,004 --> 00:00:22,005
expertise in health care and various leadership and

11
00:00:22,005 --> 00:00:24,164
executive roles in the payer consulting and delegated

12
00:00:24,164 --> 00:00:26,789
services market. Market. She's focused on leading health

13
00:00:26,789 --> 00:00:29,589
care operations and business process optimization in highly

14
00:00:29,589 --> 00:00:31,449
complex and matrixed organizations

15
00:00:31,990 --> 00:00:34,630
to bridge the gap between clinical technology and

16
00:00:34,630 --> 00:00:37,510
business objectives. Liz, thanks so much for being

17
00:00:37,510 --> 00:00:38,170
here today.

18
00:00:38,604 --> 00:00:40,284
Lucas, thank you so much for having me

19
00:00:40,284 --> 00:00:42,545
on your podcast. I'm really excited today

20
00:00:43,085 --> 00:00:43,984
to dig into

21
00:00:44,524 --> 00:00:45,725
our topic and,

22
00:00:46,204 --> 00:00:47,184
have a great conversation.

23
00:00:47,965 --> 00:00:50,284
Yes. We do have a lot to dig

24
00:00:50,284 --> 00:00:53,100
into. There's a lot to talk about here,

25
00:00:53,739 --> 00:00:55,359
specifically with prior authorizations.

26
00:00:56,140 --> 00:00:58,059
Again, it's been traditionally framed sort of as

27
00:00:58,059 --> 00:00:59,839
this administrative burden,

28
00:01:00,299 --> 00:01:02,699
but then we're seeing that shift to increasingly

29
00:01:02,699 --> 00:01:04,799
being discussed as part of the member experience,

30
00:01:04,859 --> 00:01:05,840
provider relations,

31
00:01:06,265 --> 00:01:09,225
and certainly compliance issues, and it it's happening

32
00:01:09,225 --> 00:01:10,685
all at once. Right?

33
00:01:11,305 --> 00:01:12,365
From your perspective,

34
00:01:13,064 --> 00:01:17,005
what's fundamentally broken into today's prior auth model,

35
00:01:17,225 --> 00:01:20,140
and what's been missing most specifically from these

36
00:01:20,140 --> 00:01:20,640
conversations

37
00:01:20,939 --> 00:01:22,799
around how to fix all of this?

38
00:01:23,340 --> 00:01:24,560
Yeah. You know, fundamentally,

39
00:01:25,340 --> 00:01:26,400
the prior authorization

40
00:01:26,700 --> 00:01:29,740
model, it's totally viewed as a gatekeeping, right,

41
00:01:29,740 --> 00:01:30,844
or limiting access

42
00:01:31,265 --> 00:01:31,765
to

43
00:01:32,185 --> 00:01:34,905
care by the provider community and by members.

44
00:01:34,905 --> 00:01:35,405
However,

45
00:01:35,864 --> 00:01:37,965
you know, prior auth was instituted

46
00:01:38,344 --> 00:01:40,765
as part of the larger managed care program

47
00:01:40,984 --> 00:01:43,545
evolution over the last, you know, twenty, twenty

48
00:01:43,545 --> 00:01:44,204
five years.

49
00:01:44,549 --> 00:01:45,609
So over time,

50
00:01:46,790 --> 00:01:48,810
the the process has been burdened

51
00:01:49,109 --> 00:01:49,609
administratively

52
00:01:50,229 --> 00:01:52,009
due to the increasing in regulatory

53
00:01:52,310 --> 00:01:52,810
pressures,

54
00:01:53,109 --> 00:01:55,049
compliance and quality requirements,

55
00:01:55,750 --> 00:01:57,674
and just the scope of services,

56
00:01:57,975 --> 00:01:59,835
new therapies, new medications

57
00:02:00,615 --> 00:02:03,435
that have, you know, become part of mainstream

58
00:02:04,215 --> 00:02:05,594
clinical care delivery,

59
00:02:06,215 --> 00:02:08,955
has just made prior auth very, very complex.

60
00:02:10,215 --> 00:02:11,754
There's lots of

61
00:02:12,110 --> 00:02:15,310
different things the payer community has been trying

62
00:02:15,310 --> 00:02:18,209
to do to make this prior authorization process

63
00:02:18,829 --> 00:02:19,329
easier,

64
00:02:19,709 --> 00:02:20,930
but, traditionally,

65
00:02:21,310 --> 00:02:24,289
they've been kind of like gold carding programs.

66
00:02:24,349 --> 00:02:27,069
Right? So giving certain providers certain services a

67
00:02:27,069 --> 00:02:29,495
free pass, right, for so many services or

68
00:02:29,495 --> 00:02:31,275
for certain types of members or

69
00:02:32,375 --> 00:02:32,875
other

70
00:02:33,335 --> 00:02:34,155
rule based,

71
00:02:35,254 --> 00:02:36,555
solutions where,

72
00:02:37,014 --> 00:02:38,395
you know, low,

73
00:02:39,990 --> 00:02:41,050
cost and,

74
00:02:41,669 --> 00:02:43,830
low abused services, for a lack of a

75
00:02:43,830 --> 00:02:46,330
better word, were kinda given an auto approval

76
00:02:46,389 --> 00:02:48,789
pass. You know? But that didn't really remove

77
00:02:48,789 --> 00:02:51,449
the burden of prior authorization. Right? The provider

78
00:02:51,590 --> 00:02:53,430
still had to go through the motions of

79
00:02:53,430 --> 00:02:53,930
submitting

80
00:02:54,525 --> 00:02:57,245
prior authorization and things of that nature. So,

81
00:02:57,245 --> 00:02:58,305
you know, fundamentally,

82
00:02:58,605 --> 00:02:59,105
that's

83
00:02:59,485 --> 00:03:01,645
that's the problem. Right? So but over the

84
00:03:01,645 --> 00:03:03,905
past twelve to twenty four months, there's been

85
00:03:04,125 --> 00:03:06,365
this heightened attention, right, to the whole end

86
00:03:06,365 --> 00:03:09,105
to end experience in prior authorization and

87
00:03:10,409 --> 00:03:12,830
how to make it better.

88
00:03:13,770 --> 00:03:16,409
Uncovering the root causes needs to be the

89
00:03:16,409 --> 00:03:19,689
top priority, so not just making it easier

90
00:03:19,689 --> 00:03:20,830
to follow the process,

91
00:03:21,129 --> 00:03:24,569
but is the process itself broken? So things

92
00:03:24,569 --> 00:03:26,465
like the FHIR interoperability

93
00:03:27,004 --> 00:03:28,305
standards of electronic

94
00:03:28,605 --> 00:03:31,485
data sharing, that has kinda spring boarded us

95
00:03:31,485 --> 00:03:31,985
into

96
00:03:32,685 --> 00:03:33,985
more robust automation,

97
00:03:34,605 --> 00:03:37,004
AI forward automation, you know, to now that

98
00:03:37,004 --> 00:03:38,925
we have the data connection, how do we

99
00:03:38,925 --> 00:03:41,719
leverage that and really make the experience more

100
00:03:41,719 --> 00:03:42,219
beneficial?

101
00:03:43,560 --> 00:03:45,959
You mentioned the end to end aspect of

102
00:03:45,959 --> 00:03:47,799
this, which is really crucial as you just

103
00:03:47,799 --> 00:03:48,299
highlighted.

104
00:03:48,680 --> 00:03:51,419
And there is a very growing shift toward

105
00:03:51,879 --> 00:03:54,375
AI first work flow, specifically this year. There

106
00:03:54,375 --> 00:03:56,955
was some last year, but certainly in 2026.

107
00:03:57,575 --> 00:03:59,435
How do you distinguish between

108
00:04:00,055 --> 00:04:02,694
basic automation and these AI agents that are

109
00:04:02,694 --> 00:04:05,015
coming up that change how clinical and care

110
00:04:05,015 --> 00:04:06,235
management teams work

111
00:04:06,615 --> 00:04:07,515
day to day?

112
00:04:07,879 --> 00:04:08,699
Yeah. So,

113
00:04:09,159 --> 00:04:10,300
you know, automation

114
00:04:10,680 --> 00:04:14,599
has typically been characterized as, like, RPA. Right?

115
00:04:14,599 --> 00:04:17,980
Robotic process automation and workflow business rules.

116
00:04:18,519 --> 00:04:22,120
These two approaches, they're heavily defined but with

117
00:04:22,120 --> 00:04:24,555
scripts and logic. Right? And that logic and

118
00:04:24,555 --> 00:04:27,055
those scripts have to constantly be updated

119
00:04:27,675 --> 00:04:29,855
with change in program execution,

120
00:04:30,475 --> 00:04:34,495
change in criteria sets, or change in requirements,

121
00:04:34,555 --> 00:04:35,775
regulatory requirements.

122
00:04:36,634 --> 00:04:37,935
With agentic

123
00:04:38,235 --> 00:04:38,735
forward

124
00:04:39,569 --> 00:04:40,069
automation,

125
00:04:41,089 --> 00:04:43,189
the agentic AI agents actually

126
00:04:43,569 --> 00:04:44,629
interpret requirements

127
00:04:45,089 --> 00:04:46,629
and interpret guidelines

128
00:04:46,930 --> 00:04:48,469
as a human would.

129
00:04:48,769 --> 00:04:50,629
So instead of having to,

130
00:04:51,329 --> 00:04:52,629
constantly maintain

131
00:04:52,930 --> 00:04:54,069
scripts and algorithms,

132
00:04:55,194 --> 00:04:57,455
the agentic agent reads source

133
00:04:57,915 --> 00:05:01,035
materials and source guidance each and every time

134
00:05:01,035 --> 00:05:02,814
the agent is engaged

135
00:05:03,514 --> 00:05:06,334
to support a a prior auth process.

136
00:05:06,875 --> 00:05:08,634
So so this is a couple of things.

137
00:05:08,634 --> 00:05:09,269
Right? It

138
00:05:09,829 --> 00:05:11,050
significantly reduces

139
00:05:12,149 --> 00:05:14,629
the administrative burden to the payer of having

140
00:05:14,629 --> 00:05:17,509
to maintain scripts and logic. Right? So they

141
00:05:17,509 --> 00:05:19,829
can then turn their attention in looking at

142
00:05:19,829 --> 00:05:22,229
the outcomes of the AI, looking at the

143
00:05:22,229 --> 00:05:23,529
data that is being

144
00:05:24,224 --> 00:05:26,164
harnessed through the end to end workflow.

145
00:05:27,104 --> 00:05:28,404
It creates the opportunity

146
00:05:28,865 --> 00:05:29,604
to continuously

147
00:05:29,985 --> 00:05:31,925
capture and audit and evaluate

148
00:05:32,865 --> 00:05:34,404
the accuracy of decisions

149
00:05:34,865 --> 00:05:37,699
and the time required to process each step

150
00:05:37,699 --> 00:05:38,439
of the workflow.

151
00:05:38,819 --> 00:05:41,220
Right? So this creates more information for leaders

152
00:05:41,220 --> 00:05:42,600
to manage and process

153
00:05:43,139 --> 00:05:44,120
in real time.

154
00:05:44,899 --> 00:05:46,740
Now you've touched on the advantages just now

155
00:05:46,740 --> 00:05:49,220
too. Is there one specific piece of this,

156
00:05:49,220 --> 00:05:50,439
one specific workflow

157
00:05:51,295 --> 00:05:53,375
that comes to mind for you where this

158
00:05:53,375 --> 00:05:56,095
just is a no brainer, where this is

159
00:05:56,095 --> 00:05:57,235
illustrated perfectly?

160
00:05:57,855 --> 00:05:59,714
Yeah. Yeah. So when you think about

161
00:06:00,574 --> 00:06:04,115
how clinical decision support or medical necessity reviews

162
00:06:04,254 --> 00:06:04,915
are done,

163
00:06:06,040 --> 00:06:07,339
rule based automation,

164
00:06:08,040 --> 00:06:11,259
for that step of the prior authorization process

165
00:06:11,639 --> 00:06:14,139
requires that there needs to be a documented

166
00:06:14,600 --> 00:06:15,100
definition

167
00:06:16,120 --> 00:06:19,064
every single step of that criteria statement. What

168
00:06:19,064 --> 00:06:21,465
does it mean to meet that criteria? What

169
00:06:21,465 --> 00:06:23,805
does it mean not to meet that criteria?

170
00:06:24,745 --> 00:06:27,564
Aside from that, then there's the complexity of

171
00:06:27,944 --> 00:06:30,205
all of the following must be demonstrated

172
00:06:30,585 --> 00:06:31,324
or any

173
00:06:31,970 --> 00:06:34,129
two or more of the following need to

174
00:06:34,129 --> 00:06:34,790
be demonstrated.

175
00:06:35,889 --> 00:06:39,009
And then even more complicated is if there

176
00:06:39,009 --> 00:06:40,629
are time sequence parameters

177
00:06:41,569 --> 00:06:44,610
throughout the medical evidence: six months' worth of

178
00:06:44,610 --> 00:06:45,110
therapy,

179
00:06:45,845 --> 00:06:48,745
two failed courses of treatment of a medication,

180
00:06:49,524 --> 00:06:52,564
makes it very difficult to quantify that in

181
00:06:52,564 --> 00:06:53,865
logic and definition.

182
00:06:55,204 --> 00:06:57,865
The AI agent, as I I shared before,

183
00:06:58,004 --> 00:07:01,129
reads the medical policy guidance or the CMS

184
00:07:01,269 --> 00:07:04,149
criteria bulletin, whatever is being used as that

185
00:07:04,149 --> 00:07:04,889
source of

186
00:07:05,269 --> 00:07:06,169
medical necessity

187
00:07:06,949 --> 00:07:09,750
criteria, and then reads the evidence submitted by

188
00:07:09,750 --> 00:07:12,889
the provider and then makes those matches

189
00:07:13,189 --> 00:07:14,625
just as a human would.

190
00:07:15,345 --> 00:07:17,425
But the AI agent does a little bit

191
00:07:17,425 --> 00:07:19,904
more than what a human typically does. A

192
00:07:19,904 --> 00:07:23,024
human will find what is necessary to move

193
00:07:23,024 --> 00:07:23,925
down the algorithm.

194
00:07:24,384 --> 00:07:26,384
Right? One of the following. As soon as

195
00:07:26,384 --> 00:07:27,605
one is found, typically,

196
00:07:28,064 --> 00:07:30,064
a human reviewer will go to the next

197
00:07:30,064 --> 00:07:34,360
step, and and, Agentic AI Forward workflow will

198
00:07:34,360 --> 00:07:37,319
find every piece of evidence. So it makes

199
00:07:37,319 --> 00:07:39,879
the review more robust. It makes the review

200
00:07:39,879 --> 00:07:43,160
definitely more member centric and more appropriate to

201
00:07:43,160 --> 00:07:44,620
that member in that situation.

202
00:07:46,045 --> 00:07:48,525
And then because the agent reads from the

203
00:07:48,525 --> 00:07:49,025
source,

204
00:07:50,045 --> 00:07:52,285
if there's a change to a policy, if

205
00:07:52,285 --> 00:07:54,705
CMS updates a a policy bulletin,

206
00:07:55,324 --> 00:07:57,564
the agent is reading that real time. So

207
00:07:57,564 --> 00:07:59,939
the decision tree that was in place at

208
00:07:59,939 --> 00:08:02,120
the last iteration of that workflow

209
00:08:02,740 --> 00:08:05,460
is validated by that agent reading that source

210
00:08:05,460 --> 00:08:08,259
document again and updating it if necessary if

211
00:08:08,259 --> 00:08:09,560
there's a change to criteria.

212
00:08:10,900 --> 00:08:11,400
Now

213
00:08:12,115 --> 00:08:15,394
with the technology itself, we're having a lot

214
00:08:15,394 --> 00:08:17,334
of conversations about governance,

215
00:08:18,115 --> 00:08:20,514
about checks and balances. Right? You just talked

216
00:08:20,514 --> 00:08:22,194
about the human in the loop, which is

217
00:08:22,194 --> 00:08:25,479
certainly really, really crucial. Right? And certainly one

218
00:08:25,479 --> 00:08:28,439
concern for payer leaders specifically is how to

219
00:08:28,439 --> 00:08:30,360
balance that, right, with speed that you just

220
00:08:30,360 --> 00:08:32,700
mentioned really, really quickly, really, really accurately

221
00:08:33,320 --> 00:08:36,039
with clinical rigor when AI is involved in

222
00:08:36,039 --> 00:08:37,579
that decision support. Right?

223
00:08:37,995 --> 00:08:40,654
What have you seen work well in maintaining

224
00:08:40,794 --> 00:08:43,674
that clinician trust, that that oversight, and certainly

225
00:08:43,674 --> 00:08:44,174
also

226
00:08:44,634 --> 00:08:45,134
accountability

227
00:08:46,475 --> 00:08:49,674
while still, again, improving the efficiencies that you've

228
00:08:49,674 --> 00:08:50,575
just talked about?

229
00:08:51,039 --> 00:08:54,100
Yeah. I think whenever an AI forward workflow

230
00:08:54,399 --> 00:08:54,899
is

231
00:08:55,360 --> 00:08:56,340
being implemented,

232
00:08:56,799 --> 00:08:58,500
it's really important to

233
00:08:58,879 --> 00:09:00,820
be very focused and purposeful

234
00:09:01,279 --> 00:09:03,379
about where you want that,

235
00:09:03,894 --> 00:09:07,514
agentic solution to start. So it's really a

236
00:09:07,575 --> 00:09:10,855
a a small segment of the overall prior

237
00:09:10,855 --> 00:09:13,914
offline. So maybe it's a single service type

238
00:09:14,214 --> 00:09:14,714
or

239
00:09:15,014 --> 00:09:16,855
pie you know, putting it in place with

240
00:09:16,855 --> 00:09:19,259
one provider group and all of their requests.

241
00:09:19,820 --> 00:09:22,779
This allows clinicians and leaders to evaluate the

242
00:09:22,779 --> 00:09:24,240
accuracy and the effectiveness

243
00:09:24,779 --> 00:09:26,879
of the AI workflow outcome

244
00:09:27,340 --> 00:09:29,100
in a more timely manner. Right? So you

245
00:09:29,100 --> 00:09:31,179
could imagine a payer that gets thousands of

246
00:09:31,179 --> 00:09:31,605
transaction

247
00:09:34,084 --> 00:09:34,584
possibly

248
00:09:35,684 --> 00:09:36,184
quality

249
00:09:36,884 --> 00:09:38,884
review the AI output of every single one

250
00:09:38,884 --> 00:09:39,784
of those transactions

251
00:09:40,164 --> 00:09:41,304
in early implementation

252
00:09:41,684 --> 00:09:44,084
to gain that confidence and trust, right, that

253
00:09:44,084 --> 00:09:45,464
the AI agents are delivering

254
00:09:46,379 --> 00:09:47,920
clinically solid, consistent

255
00:09:48,700 --> 00:09:49,200
recommendations.

256
00:09:50,139 --> 00:09:52,779
So starting small allows there to be almost

257
00:09:52,779 --> 00:09:54,700
a side by side or one to one

258
00:09:54,700 --> 00:09:56,879
and watching that AI evidence.

259
00:09:57,580 --> 00:09:58,080
Then

260
00:09:58,504 --> 00:10:01,004
it's able to be scaled. Once the clinicians

261
00:10:01,544 --> 00:10:03,325
have trust, the AI,

262
00:10:03,945 --> 00:10:06,845
agent is accurate, consistent, and

263
00:10:07,304 --> 00:10:08,605
is saving them time,

264
00:10:09,304 --> 00:10:10,764
then we scale. However,

265
00:10:11,144 --> 00:10:13,769
most importantly, though, is the oversight doesn't stop.

266
00:10:14,250 --> 00:10:17,370
The program monitoring and the analytics to monitor

267
00:10:17,370 --> 00:10:19,769
the consistency and accuracy has to stay in

268
00:10:19,769 --> 00:10:20,269
place,

269
00:10:20,889 --> 00:10:23,370
and there's reason for that. Right? Brand new

270
00:10:23,370 --> 00:10:26,090
policies like anything, whether it's a a group

271
00:10:26,090 --> 00:10:26,750
of clinicians

272
00:10:27,050 --> 00:10:29,070
reviewing against a new policy

273
00:10:29,605 --> 00:10:30,585
for the first time,

274
00:10:31,205 --> 00:10:34,105
or an AI agent, a Genentech agent reviewing

275
00:10:34,245 --> 00:10:36,485
against a policy for the first time. It

276
00:10:36,485 --> 00:10:38,184
needs deep oversight.

277
00:10:39,044 --> 00:10:39,544
So

278
00:10:39,845 --> 00:10:42,884
that, component needs to be consistent. And I

279
00:10:42,884 --> 00:10:44,745
I think sometimes that is a misstep,

280
00:10:45,149 --> 00:10:46,990
that as soon as we see a process

281
00:10:46,990 --> 00:10:49,970
or a technology driven solution work,

282
00:10:50,350 --> 00:10:53,169
there's sometimes too much trust in the technology,

283
00:10:53,470 --> 00:10:54,929
so the oversight is critical.

284
00:10:55,470 --> 00:10:55,970
Also,

285
00:10:57,884 --> 00:11:00,524
this is all evidence based output. Right? So

286
00:11:00,524 --> 00:11:02,065
the decisions and the recommendations

287
00:11:02,605 --> 00:11:05,184
are being captured by that agentic workflow,

288
00:11:05,485 --> 00:11:07,345
and this really creates opportunity

289
00:11:08,205 --> 00:11:08,705
for

290
00:11:09,085 --> 00:11:11,665
evidence based rigor and decision making

291
00:11:12,159 --> 00:11:14,799
in the workflow and can really help reduce

292
00:11:14,799 --> 00:11:16,419
and educate human bias.

293
00:11:17,759 --> 00:11:20,179
So to be able to enable that then,

294
00:11:20,320 --> 00:11:20,820
right,

295
00:11:21,519 --> 00:11:22,340
looking ahead,

296
00:11:23,039 --> 00:11:25,379
what are those things from an investment perspective

297
00:11:25,679 --> 00:11:27,379
and from a focus perspective

298
00:11:28,294 --> 00:11:32,235
that payers should really ensure is actually there

299
00:11:32,534 --> 00:11:35,095
and is met for the future? What are

300
00:11:35,095 --> 00:11:36,694
those things that need to be in place

301
00:11:36,694 --> 00:11:39,095
right now? Yeah. So I'm really not even

302
00:11:39,095 --> 00:11:40,475
gonna talk about the technology

303
00:11:40,855 --> 00:11:42,315
components, right, because

304
00:11:42,769 --> 00:11:45,169
those are defined. I think, one of the

305
00:11:45,169 --> 00:11:46,470
biggest things that

306
00:11:46,850 --> 00:11:48,789
I found to be a miss,

307
00:11:50,049 --> 00:11:52,529
and that causes delay and a a lot

308
00:11:52,529 --> 00:11:55,190
of uptime in trying to get these, agentic

309
00:11:55,409 --> 00:11:57,110
workflows in place is,

310
00:11:57,715 --> 00:12:00,835
a defined change management and oversight process. Right?

311
00:12:00,835 --> 00:12:03,575
So when we're bringing AI into the workflow,

312
00:12:04,034 --> 00:12:05,014
that is materially

313
00:12:05,475 --> 00:12:05,975
changing

314
00:12:06,355 --> 00:12:07,495
the day in the life

315
00:12:07,955 --> 00:12:09,634
of all the humans in the loop, whether

316
00:12:09,634 --> 00:12:11,495
they are support staff, nurses,

317
00:12:12,100 --> 00:12:14,600
physicians. It it doesn't matter. So

318
00:12:15,220 --> 00:12:18,200
the the change management aspect is critical,

319
00:12:19,539 --> 00:12:21,879
to ensure that there is adoption,

320
00:12:22,659 --> 00:12:26,205
process adherence, and that being able to evaluate

321
00:12:26,205 --> 00:12:27,424
process effectiveness.

322
00:12:28,205 --> 00:12:30,545
So so to to me, that is

323
00:12:31,565 --> 00:12:32,065
quintessentially

324
00:12:32,524 --> 00:12:33,024
important.

325
00:12:33,485 --> 00:12:34,705
It also allows

326
00:12:35,085 --> 00:12:37,804
for the human in the loop to more

327
00:12:37,804 --> 00:12:38,304
effectively

328
00:12:38,605 --> 00:12:40,570
gain trust when they understand

329
00:12:41,350 --> 00:12:43,029
what their new day in the life is

330
00:12:43,029 --> 00:12:45,590
gonna be like, why it is important to

331
00:12:45,590 --> 00:12:48,490
embrace it, and demonstrating for them the value

332
00:12:48,870 --> 00:12:52,149
that bringing technology and agentic support into their

333
00:12:52,149 --> 00:12:53,850
workflow will give them.

334
00:12:54,345 --> 00:12:57,225
And as with anything else, the the end

335
00:12:57,225 --> 00:13:00,424
users can then start to rely on the

336
00:13:00,424 --> 00:13:02,764
data that is being presented to them

337
00:13:03,304 --> 00:13:05,644
and kind of change the way they

338
00:13:06,529 --> 00:13:07,990
execute their responsibilities

339
00:13:08,529 --> 00:13:11,350
instead of having to build the case example

340
00:13:11,570 --> 00:13:13,190
and then making the recommendation.

341
00:13:14,050 --> 00:13:17,269
They can re case example and then validate

342
00:13:17,490 --> 00:13:18,149
a recommended

343
00:13:18,610 --> 00:13:21,029
result and modify that as necessary.

344
00:13:22,245 --> 00:13:23,705
Now the end goal, obviously,

345
00:13:24,245 --> 00:13:26,965
if and when this is all happening is

346
00:13:26,965 --> 00:13:28,904
the fact that this needs to be scalable

347
00:13:28,965 --> 00:13:31,285
across an enterprise. This needs to be scalable

348
00:13:31,285 --> 00:13:33,705
for a large number of people, for multiple

349
00:13:33,845 --> 00:13:36,470
business units, etcetera, whoever is working in those

350
00:13:36,470 --> 00:13:37,769
processes. Right? However,

351
00:13:38,230 --> 00:13:39,909
I like to say we've been living in

352
00:13:39,909 --> 00:13:42,230
the year of point solutions in 2024 and

353
00:13:42,230 --> 00:13:43,209
2025.

354
00:13:43,750 --> 00:13:45,209
Yep. Many organizations

355
00:13:45,509 --> 00:13:49,144
struggle to move beyond those point solutions in

356
00:13:49,245 --> 00:13:50,065
prior auth, specifically,

357
00:13:50,605 --> 00:13:54,065
especially when there is legacy platforms evolved, etcetera.

358
00:13:54,284 --> 00:13:54,784
Right?

359
00:13:55,164 --> 00:13:57,884
What does it take to really scale those

360
00:13:57,884 --> 00:13:59,264
AI driven decision supports

361
00:13:59,884 --> 00:14:01,184
across an enterprise?

362
00:14:01,959 --> 00:14:03,480
Yeah. So I think it's really

363
00:14:03,959 --> 00:14:05,339
it all lies in

364
00:14:05,720 --> 00:14:07,500
the fundamental development architecture

365
00:14:07,959 --> 00:14:10,539
of that AI forward workflow.

366
00:14:11,159 --> 00:14:12,459
So when we think about

367
00:14:12,919 --> 00:14:17,485
how to bring AI capabilities across multiple utilization

368
00:14:17,625 --> 00:14:18,684
management workflows.

369
00:14:19,384 --> 00:14:22,424
It's really the agentic agent approach is the

370
00:14:22,424 --> 00:14:25,304
most nimble and the most flexible. So, for

371
00:14:25,304 --> 00:14:25,804
example,

372
00:14:26,264 --> 00:14:28,125
you can create a single

373
00:14:28,504 --> 00:14:29,725
AI agent

374
00:14:30,184 --> 00:14:30,845
that handles

375
00:14:31,169 --> 00:14:33,730
a prior auth workflow end to end that

376
00:14:33,730 --> 00:14:37,029
can be integrated in and around any

377
00:14:37,730 --> 00:14:40,450
product or platform workflow. Right? But it's still

378
00:14:40,450 --> 00:14:41,509
a single agent.

379
00:14:42,289 --> 00:14:44,850
So that agent only knows that workflow. It's

380
00:14:44,850 --> 00:14:46,785
trained on that work flow. But if you

381
00:14:46,785 --> 00:14:48,165
think about deconstructing

382
00:14:48,625 --> 00:14:50,865
the end to end of prior auth, let's

383
00:14:50,865 --> 00:14:53,345
talk about things like a medical record agent

384
00:14:53,345 --> 00:14:56,304
or a clinical policy agent and a decision

385
00:14:56,304 --> 00:14:58,785
support agent. So when we think about those

386
00:14:58,785 --> 00:14:59,924
three key steps,

387
00:15:00,529 --> 00:15:03,350
the agent is trained to read medical policies

388
00:15:03,410 --> 00:15:04,870
with extreme accuracy.

389
00:15:05,570 --> 00:15:07,910
Another agent is trained to extract

390
00:15:08,610 --> 00:15:12,309
the key relevant clinical data in submitted

391
00:15:13,044 --> 00:15:13,865
clinical records

392
00:15:14,404 --> 00:15:15,144
to align

393
00:15:15,684 --> 00:15:17,764
to that medical policy, and that's what that

394
00:15:17,764 --> 00:15:20,884
decision support agent does is orchestrating that decision

395
00:15:20,884 --> 00:15:21,384
tree.

396
00:15:21,764 --> 00:15:25,205
Because they're independent agents finely tuned on a

397
00:15:25,205 --> 00:15:27,304
specific skill or step,

398
00:15:28,110 --> 00:15:31,409
anywhere in the organization that that same function,

399
00:15:32,830 --> 00:15:35,470
occurs, that agent can be deployed. So when

400
00:15:35,470 --> 00:15:37,009
you think of utilization management,

401
00:15:37,870 --> 00:15:40,110
it can pry off we've been talking about.

402
00:15:40,110 --> 00:15:41,169
Appeals and grievances,

403
00:15:42,095 --> 00:15:44,654
same set of agents, but just doing a

404
00:15:44,654 --> 00:15:45,555
different type

405
00:15:46,014 --> 00:15:48,735
of medical necessity review. But the steps of

406
00:15:48,735 --> 00:15:51,235
reviewing a record, looking at criteria,

407
00:15:51,695 --> 00:15:53,235
and supporting that decision

408
00:15:53,855 --> 00:15:54,514
are there.

409
00:15:54,959 --> 00:15:57,839
Medical claim review. Right? When claims pen for

410
00:15:57,839 --> 00:16:01,279
medical necessity review and clinical is needed, same

411
00:16:01,279 --> 00:16:02,419
set of steps.

412
00:16:02,799 --> 00:16:04,179
So by investing

413
00:16:04,639 --> 00:16:05,939
in foundational

414
00:16:06,319 --> 00:16:06,819
agents

415
00:16:07,199 --> 00:16:08,100
that are,

416
00:16:09,054 --> 00:16:11,634
tied to a single executable task.

417
00:16:11,934 --> 00:16:14,815
That agent can be deployed across multiple systems

418
00:16:14,815 --> 00:16:16,914
and multiple workflows within the organization.

419
00:16:18,334 --> 00:16:20,174
Maybe we finally reached the year of the

420
00:16:20,174 --> 00:16:22,059
agent at this point in time. That might

421
00:16:22,059 --> 00:16:23,820
be it. Liz, it's so great to have

422
00:16:23,820 --> 00:16:25,820
you on. So many great nuggets in this

423
00:16:25,820 --> 00:16:27,740
conversation. Anything else that you'd like to touch

424
00:16:27,740 --> 00:16:29,420
on that that we haven't mentioned for our

425
00:16:29,420 --> 00:16:29,920
audience?

426
00:16:30,540 --> 00:16:31,980
You know, I think the only other thing

427
00:16:31,980 --> 00:16:33,420
I'd like to talk about is we've talked

428
00:16:33,420 --> 00:16:35,475
a lot about work flows and AI in

429
00:16:35,475 --> 00:16:36,134
the workflow,

430
00:16:36,675 --> 00:16:38,595
but we really have been spent time on

431
00:16:38,595 --> 00:16:41,495
what is truly required to allow

432
00:16:41,795 --> 00:16:45,394
any AI forward workflow to execute accurately, and

433
00:16:45,394 --> 00:16:48,789
that's data. Right? So analyzing the accessibility

434
00:16:49,250 --> 00:16:50,070
of the data

435
00:16:50,529 --> 00:16:51,990
and an organization

436
00:16:52,449 --> 00:16:54,529
to having a data, what we call AI

437
00:16:54,529 --> 00:16:55,029
ready,

438
00:16:55,730 --> 00:16:58,049
is a critical first step, right, to any

439
00:16:58,049 --> 00:16:58,870
AI forward

440
00:16:59,250 --> 00:17:00,870
workflow effort. So

441
00:17:01,174 --> 00:17:04,075
being able to convert structured and unstructured data

442
00:17:04,214 --> 00:17:08,554
means faster retrieval for analytics and operational integration.

443
00:17:09,414 --> 00:17:12,054
So really the first step. You know, we

444
00:17:12,775 --> 00:17:15,769
technology can do anything, but that that foundational

445
00:17:16,070 --> 00:17:19,029
information, which is data. It won't execute as

446
00:17:19,029 --> 00:17:20,330
effectively as reliably

447
00:17:20,950 --> 00:17:23,930
and produce the outcomes necessary to be worthwhile.

448
00:17:25,430 --> 00:17:27,670
You heard it here first. Next podcast with

449
00:17:27,670 --> 00:17:30,309
Liz on data. You'll have to make sure

450
00:17:30,309 --> 00:17:32,444
to tune in. Liz, thank you so much

451
00:17:32,444 --> 00:17:34,464
for being here. It's great to have you.

452
00:17:34,684 --> 00:17:35,904
Thanks so much, Lucas.

453
00:17:36,525 --> 00:17:38,285
We also want to thank our podcast sponsor,

454
00:17:38,285 --> 00:17:40,144
eXol. You can tune in to more podcasts

455
00:17:40,285 --> 00:17:42,365
from Becker's Health Care by visiting our podcast

456
00:17:42,365 --> 00:17:45,585
page at beckershospitalreview.com.