Skip to main content

Webinar

Navigating the complexities of healthcare automation

Learn how to assess your organization’s readiness for automation. Get strategies to unlock the full potential of your automation investments.

1

00:00:00,280 --> 00:00:03,720

Good morning, everyone.

I'm Katie LeBlanc.

 

2

00:00:04,360 --> 00:00:08,080

In terms of the agenda

today, we will be

 

3

00:00:08,080 --> 00:00:10,310

going through a couple

of really key areas

 

4

00:00:10,310 --> 00:00:13,300

to share with you.

We'll go through the AI

 

5

00:00:13,300 --> 00:00:16,190

market drivers and

technical advancements,

 

6

00:00:16,420 --> 00:00:18,960

target areas for

automation, framing your

 

7

00:00:18,960 --> 00:00:21,660

roadmap, some key

methods of success,

 

8

00:00:21,660 --> 00:00:24,460

and then move into

some closing remarks.

 

9

00:00:34,730 --> 00:00:35,670

All right.

 

10

00:00:35,850 --> 00:00:38,110

So, you know,

for those of us

 

11

00:00:38,110 --> 00:00:39,310

who are privileged

enough to

 

12

00:00:39,310 --> 00:00:40,890

work in the

healthcare industry,

 

13

00:00:41,210 --> 00:00:43,330

we all know the

landscape here.

 

14

00:00:43,790 --> 00:00:45,590

We've got rising costs,

 

15

00:00:45,590 --> 00:00:47,170

reimbursement

compression,

 

16

00:00:47,350 --> 00:00:48,690

labor shortages,

 

17

00:00:49,290 --> 00:00:51,770

things like value

-based care, which are

 

18

00:00:51,770 --> 00:00:56,030

really challenging us to

focus on high-quality,

 

19

00:00:56,030 --> 00:00:59,230

complete documentation

capture, which is

 

20

00:00:59,230 --> 00:01:01,610

critical. And it's

a really good thing,

 

21

00:01:01,610 --> 00:01:04,070

right, because that is

driving better patient

 

22

00:01:04,070 --> 00:01:07,030

outcomes but it's a

challenge from the

 

23

00:01:07,030 --> 00:01:10,350

perspective that it's

really labor-intensive

 

24

00:01:10,470 --> 00:01:13,050

it adds additional

cost and staffing

 

25

00:01:13,050 --> 00:01:16,410

Challenges. So I

know that many of us

 

26

00:01:16,410 --> 00:01:19,790

as we're learning

more about AI we're

 

27

00:01:20,230 --> 00:01:22,790

really looking at

AI as one of the

 

28

00:01:22,790 --> 00:01:26,470

things that can help

us manage through

 

29

00:01:26,570 --> 00:01:28,410

all of these

challenges that

 

30

00:01:28,410 --> 00:01:30,410

we're facing

as an industry.

 

31

00:01:31,130 --> 00:01:34,850

You'll see here on the

slide talking about

 

32

00:01:34,850 --> 00:01:38,630

98% of leaders

mentioning that you're

 

33

00:01:38,630 --> 00:01:41,390

planning to implement

an AI strategy. I know

 

34

00:01:41,390 --> 00:01:44,370

here at Banner Health,

we're spending a lot

 

35

00:01:44,370 --> 00:01:47,690

of time thinking about

how do we deploy these

 

36

00:01:47,690 --> 00:01:50,090

technologies, these

new technologies in

 

37

00:01:50,090 --> 00:01:54,010

ways that are ethical,

in ways that are

 

38

00:01:54,010 --> 00:01:58,160

really helping us to

imagine what the future

 

39

00:01:58,160 --> 00:02:01,770

of our revenue cycle

can look like and really

 

40

00:02:01,770 --> 00:02:04,060

meet the needs of

addressing some of the

 

41

00:02:04,060 --> 00:02:07,270

major issues that are

plaguing us today.

 

42

00:02:09,730 --> 00:02:11,270

I'm going to

do a check-in.

 

43

00:02:11,270 --> 00:02:12,670

Lori, are you there yet?

 

44

00:02:14,510 --> 00:02:17,350

All right. We'll

move forward.

 

45

00:02:30,220 --> 00:02:32,940

So when I think about

all of the different

 

46

00:02:32,940 --> 00:02:34,900

things that have

been coming into

 

47

00:02:34,900 --> 00:02:40,500

our lens from an AI

perspective, we're

 

48

00:02:40,500 --> 00:02:42,560

all seeing this

flurry of headlines,

 

49

00:02:42,820 --> 00:02:45,020

whether that's

around all of

 

50

00:02:45,020 --> 00:02:46,920

these new product

releases, think

 

51

00:02:46,920 --> 00:02:49,900

of OpenAI,

Anthropic with Claude,

 

52

00:02:50,340 --> 00:02:53,460

Google releasing

Gemini. There's a

 

53

00:02:53,460 --> 00:02:56,500

lot of new words

and frankly, markets

 

54

00:02:56,500 --> 00:02:59,580

that have emerged.

So from a Banner

 

55

00:02:59,580 --> 00:03:01,440

Health perspective,

we're really taking

 

56

00:03:01,440 --> 00:03:03,460

this learn-before

-we-lead approach.

 

57

00:03:04,120 --> 00:03:07,220

I, as a leader

here at Banner, and

 

58

00:03:07,220 --> 00:03:09,340

from a revenue

cycle perspective,

 

59

00:03:09,560 --> 00:03:11,960

one of the things that

I've personally been

 

60

00:03:11,960 --> 00:03:14,420

doing and been

challenging my teams to do

 

61

00:03:14,420 --> 00:03:17,080

is to engage with all

of these platforms,

 

62

00:03:17,080 --> 00:03:20,060

engage with the AI,

whether that's from a

 

63

00:03:20,060 --> 00:03:23,600

professional or personal

perspective, until

 

64

00:03:23,600 --> 00:03:26,610

we understand more about

these technologies,

 

65

00:03:26,610 --> 00:03:29,540

it's really hard

for us to be able to

 

66

00:03:29,540 --> 00:03:33,700

Imagine, in a business

Setting, what we can

 

67

00:03:33,700 --> 00:03:35,980

use, how we can use

these things, and how

 

68

00:03:35,980 --> 00:03:39,560

it's really going to

transform our business.

 

69

00:03:40,720 --> 00:03:42,420

I think, yes.

 

70

00:03:43,320 --> 00:03:44,700

And just to add one thing

 

71

00:03:44,700 --> 00:03:46,000

here, I think the other

 

72

00:03:46,160 --> 00:03:49,260

important thing, and

sorry to the folks on

 

73

00:03:49,260 --> 00:03:51,000

the phone, I had

technical difficulties.

 

74

00:03:51,080 --> 00:03:54,180

Katie jumped right in,

but I think the other

 

75

00:03:54,180 --> 00:03:56,220

thing is that as you're

thinking about your

 

76

00:03:56,220 --> 00:03:59,340

future and selecting

vendors, that you've got

 

77

00:03:59,340 --> 00:04:01,880

to be careful because

technology is advancing

 

78

00:04:01,880 --> 00:04:04,540

so quick. You want

to make sure that

 

79

00:04:04,540 --> 00:04:08,760

vendor has both the

investments and the time

 

80

00:04:08,760 --> 00:04:11,460

to keep pace with

technologies. One of the

 

81

00:04:11,460 --> 00:04:13,340

things we're seeing

is we're bringing new

 

82

00:04:13,340 --> 00:04:17,340

models in, it feels like

at least monthly, if

 

83

00:04:17,340 --> 00:04:20,020

not more frequent, to

understand what they

 

84

00:04:20,020 --> 00:04:23,460

bring to the mix that

may be better than other

 

85

00:04:23,460 --> 00:04:26,220

models or complementary

to other models.

 

86

00:04:28,700 --> 00:04:29,880

Absolutely.

 

87

00:04:30,080 --> 00:04:33,380

So that whole

notion of really

 

88

00:04:33,380 --> 00:04:36,160

learning and

understanding this

 

89

00:04:36,160 --> 00:04:39,260

landscape, and the

challenge is that

 

90

00:04:39,260 --> 00:04:41,420

it is evolving

very, very quickly.

 

91

00:04:41,700 --> 00:04:44,980

So it's absolutely

one of the things

 

92

00:04:44,980 --> 00:04:47,020

that we keep top

of mind. And I

 

93

00:04:47,020 --> 00:04:48,680

think as industry

leaders, it's

 

94

00:04:48,680 --> 00:04:50,460

imperative for all

of us to do that.

 

95

00:04:54,640 --> 00:04:57,160

So as we talk about

the different AI

 

96

00:04:57,160 --> 00:05:00,800

capabilities out there,

there are all types of AI

 

97

00:05:00,800 --> 00:05:05,000

at varying levels of

application, especially

 

98

00:05:05,000 --> 00:05:07,620

for healthcare. But

they all seem to have

 

99

00:05:07,620 --> 00:05:10,040

their strength. And

frankly, some have

 

100

00:05:10,040 --> 00:05:12,600

weaknesses. And so

what we've known in the

 

101

00:05:12,600 --> 00:05:15,580

market for a while is

things like Natural

 

102

00:05:15,580 --> 00:05:18,760

Language Processing.

And it's really still a

 

103

00:05:18,760 --> 00:05:21,020

very important part

of the game because it

 

104

00:05:21,020 --> 00:05:24,000

understands and generates

human language with

 

105

00:05:24,000 --> 00:05:26,920

high accuracy. And in a

place like healthcare,

 

106

00:05:26,920 --> 00:05:29,820

where the difference

between the patient

 

107

00:05:29,820 --> 00:05:32,740

having a disease

versus having a history

 

108

00:05:32,740 --> 00:05:35,660

of a disease versus

family history of a

 

109

00:05:35,660 --> 00:05:39,340

disease can matter when

you're trying to apply

 

110

00:05:39,340 --> 00:05:42,720

AI on top of that.

And so we feel like

 

111

00:05:42,720 --> 00:05:45,200

that Natural Language

Processing still very

 

112

00:05:45,200 --> 00:05:47,160

much, while not being

the latest great

 

113

00:05:47,160 --> 00:05:50,820

technology, plays a

very important role in

 

114

00:05:50,820 --> 00:05:55,280

performing automation

tasks related to health

 

115

00:05:55,280 --> 00:05:57,740

care for that very reason.

So an example would

 

116

00:05:57,740 --> 00:06:00,260

be extracting insights

from unstructured

 

117

00:06:00,260 --> 00:06:04,400

EHR notes, automating

clinical documentation,

 

118

00:06:04,400 --> 00:06:07,500

powering virtual

assistants. One we've

 

119

00:06:07,500 --> 00:06:09,840

been hearing a lot

about because of some of

 

120

00:06:09,840 --> 00:06:12,860

the ambient capabilities

our clinicians are

 

121

00:06:12,860 --> 00:06:15,260

using for documentation

is Natural Language

 

122

00:06:15,260 --> 00:06:18,460

Understanding. And

the strength there is

 

123

00:06:18,460 --> 00:06:21,300

understanding conversational

context between

 

124

00:06:21,300 --> 00:06:24,280

clinicians and patients,

literally taking

 

125

00:06:24,280 --> 00:06:28,580

and automating a script

of that conversation

 

126

00:06:28,580 --> 00:06:31,660

and ultimately

translating that script

 

127

00:06:31,660 --> 00:06:35,260

into documentation that

is then sent back to

 

128

00:06:35,260 --> 00:06:37,860

your EMR. So that

example, again, would be

 

129

00:06:37,860 --> 00:06:41,300

things like converting

lay language to

 

130

00:06:41,300 --> 00:06:43,520

structured clinical

notes and some of the

 

131

00:06:43,520 --> 00:06:47,040

vendors here that are

in this space right now.

 

132

00:06:47,180 --> 00:06:50,100

Large Language Models.

So when we think about

 

133

00:06:50,100 --> 00:06:53,120

tools like ChatGPT

and Copilot and other

 

134

00:06:53,120 --> 00:06:55,580

Large Language Models

out there, their

 

135

00:06:55,580 --> 00:06:58,980

strength is it scales

across tasks with minimal

 

136

00:06:58,980 --> 00:07:01,900

fine-tuning and

contextual understanding.

 

137

00:07:02,140 --> 00:07:05,980

What I see is it can

process large amounts

 

138

00:07:05,980 --> 00:07:10,180

of information at a rate

other models haven't

 

139

00:07:10,180 --> 00:07:12,800

been able to do. The

other thing we saw

 

140

00:07:12,800 --> 00:07:17,000

in our own experimentation

was that while it

 

141

00:07:17,000 --> 00:07:20,720

can process information

quickly, that in

 

142

00:07:20,720 --> 00:07:23,340

fact, when we put our

experts, whether that

 

143

00:07:23,340 --> 00:07:26,260

expert be a call center

expert or a coding

 

144

00:07:26,260 --> 00:07:29,440

expert or someone else

in healthcare, that

 

145

00:07:29,440 --> 00:07:31,940

when you put the right

people with the right

 

146

00:07:31,940 --> 00:07:35,200

technology to create

those prompts to ask

 

147

00:07:35,200 --> 00:07:37,720

the right question to

get the right answer,

 

148

00:07:37,940 --> 00:07:40,380

in fact, they

evolve quickly.

 

149

00:07:41,370 --> 00:07:44,460

And it's used more for

summarizing patient

 

150

00:07:44,460 --> 00:07:46,940

records, answering

clinical questions,

 

151

00:07:46,940 --> 00:07:49,400

and drafting things

like discharge notes.

 

152

00:07:49,520 --> 00:07:53,520

Computer Vision is a

high precision in image

 

153

00:07:53,520 --> 00:07:57,480

recognition and pattern

detection. So examples

 

154

00:07:57,480 --> 00:07:59,600

of those would be

detecting tumors from

 

155

00:07:59,600 --> 00:08:03,100

radiology scans, analyzing

pathology slides.

 

156

00:08:03,100 --> 00:08:06,780

And then Generative

AI creates new content

 

157

00:08:06,780 --> 00:08:09,380

from learned patterns,

generating synthetic

 

158

00:08:09,380 --> 00:08:13,020

patient data for

research, drafting patient

 

159

00:08:13,020 --> 00:08:16,840

education materials,

simulating rare disease

 

160

00:08:16,840 --> 00:08:20,580

scenarios. A few more

I want to cover before

 

161

00:08:20,580 --> 00:08:23,700

we move on is Predictive

Analytics identifies

 

162

00:08:23,700 --> 00:08:27,100

trends and forecasts

outcomes from large

 

163

00:08:27,100 --> 00:08:30,100

data sets, predicting

patient deterioration,

 

164

00:08:30,340 --> 00:08:32,700

or readmission

risk, or optimizing

 

165

00:08:32,700 --> 00:08:34,900

hospital resource

allocation.

 

166

00:08:35,200 --> 00:08:37,260

Autonomous Systems,

which we're

 

167

00:08:37,260 --> 00:08:40,360

hearing a lot about,

operate independently

 

168

00:08:40,360 --> 00:08:42,900

with minimal

human input. So

 

169

00:08:42,900 --> 00:08:46,140

robotic-assisted

surgery is an example.

 

170

00:08:46,840 --> 00:08:50,640

Medication dispensing,

smart logistics in

 

171

00:08:50,640 --> 00:08:54,140

hospitals, and then AI

Agents and Assistants.

 

172

00:08:54,140 --> 00:08:56,140

And we definitely

should not

 

173

00:08:56,140 --> 00:08:59,180

underestimate the power

of those models for

 

174

00:08:59,180 --> 00:09:01,240

things like virtual

nurses for follow

 

175

00:09:01,240 --> 00:09:03,940

-up care, AI

triage bots, or co

 

176

00:09:03,940 --> 00:09:06,380

-pilots for clinical

decision support.

 

177

00:09:11,910 --> 00:09:14,570

So as you're evaluating

AI technologies,

 

178

00:09:14,570 --> 00:09:17,530

it's critical to understand

how these translate

 

179

00:09:17,530 --> 00:09:20,010

to cost-benefit for

your organization.

 

180

00:09:20,170 --> 00:09:23,830

Those savings range

from 5% to more than

 

181

00:09:23,830 --> 00:09:27,400

50% in savings that

can be achieved using

 

182

00:09:27,400 --> 00:09:30,630

AI. The excitement

around automation will

 

183

00:09:30,630 --> 00:09:33,250

tempt you to dive

in headfirst, but

 

184

00:09:33,250 --> 00:09:35,970

depending on methodology,

this can be time

 

185

00:09:35,970 --> 00:09:39,350

-consuming and costly.

So evaluating what

 

186

00:09:39,350 --> 00:09:42,190

success looks like

prior to your selection

 

187

00:09:42,190 --> 00:09:45,130

is critical to assuring

you are achieving

 

188

00:09:45,130 --> 00:09:47,990

your goals. It's not

just about picking

 

189

00:09:47,990 --> 00:09:50,410

the latest great

technology. It's about

 

190

00:09:50,410 --> 00:09:53,610

picking the right

technology for the use

 

191

00:09:53,610 --> 00:09:55,950

case that you believe

will drive the right

 

192

00:09:55,950 --> 00:09:58,100

benefits for you and

your organization.

 

193

00:09:58,710 --> 00:10:01,150

Katie, can you share

more on how you

 

194

00:10:01,150 --> 00:10:05,290

see ROI on the path

to AI maturity?

 

195

00:10:05,870 --> 00:10:07,730

Sure, Lorri. Thank you.

 

196

00:10:08,270 --> 00:10:11,660

You know, this continuum

is exciting, right?

 

197

00:10:11,660 --> 00:10:15,090

You know, we all see

these numbers, and we

 

198

00:10:15,090 --> 00:10:17,470

want to get to the

far right-hand side of

 

199

00:10:17,470 --> 00:10:20,530

this continuum as

quickly as possible. But

 

200

00:10:20,530 --> 00:10:23,290

really, we have to step

back and ask ourselves,

 

201

00:10:23,290 --> 00:10:25,950

what has to be true

for us to get there?

 

202

00:10:26,170 --> 00:10:28,890

Just chasing that

ROI without making

 

203

00:10:28,890 --> 00:10:30,810

sure that you have some

of the fundamentals

 

204

00:10:30,810 --> 00:10:34,290

down would be a

very costly mistake.

 

205

00:10:34,410 --> 00:10:37,130

So we think about

things like data

 

206

00:10:37,130 --> 00:10:40,550

structure. Do we

have the right data

 

207

00:10:40,550 --> 00:10:42,730

structure in place

to be able to truly

 

208

00:10:42,730 --> 00:10:46,250

leverage and be able to

trust these technologies

 

209

00:10:46,250 --> 00:10:49,890

that, as we go,

again, more towards

 

210

00:10:49,890 --> 00:10:54,650

this autonomous,

deep learning AI,

 

211

00:10:54,870 --> 00:10:57,390

do we have the right

data structure and the

 

212

00:10:57,390 --> 00:10:59,970

right data integrity

to really be able to

 

213

00:10:59,970 --> 00:11:02,670

move into that space?

Do we have the right

 

214

00:11:02,670 --> 00:11:06,450

governance that is

there? Do our vendors

 

215

00:11:06,450 --> 00:11:08,790

that we're relying

on have that right

 

216

00:11:08,790 --> 00:11:11,110

governance, which we'll

talk about a little

 

217

00:11:11,110 --> 00:11:15,350

bit later. But certainly,

as we're evaluating

 

218

00:11:15,350 --> 00:11:18,790

different vendor

solutions, and frankly,

 

219

00:11:18,790 --> 00:11:21,410

different ways that

we can leverage AI

 

220

00:11:21,410 --> 00:11:24,710

internal to Banner,

these are some of the

 

221

00:11:24,710 --> 00:11:27,830

things that we think

about. And wanting to

 

222

00:11:27,830 --> 00:11:30,910

make sure that wherever

we choose to make

 

223

00:11:30,910 --> 00:11:33,850

those investments,

we're very targeted and

 

224

00:11:33,850 --> 00:11:36,450

we understand what

success is going to look

 

225

00:11:36,450 --> 00:11:38,730

like and making sure

that we've got those

 

226

00:11:38,730 --> 00:11:41,930

critical elements that

I just talked about

 

227

00:11:41,930 --> 00:11:44,590

to ensure that the

outcome is going to be

 

228

00:11:44,590 --> 00:11:47,170

there. It's not

just a promise. It's

 

229

00:11:47,170 --> 00:11:49,350

something that we know

that we can deliver on.

 

230

00:11:50,150 --> 00:11:51,690

Thank you, Katie.

 

231

00:11:53,590 --> 00:11:56,450

So cost efficiency

for health systems

 

232

00:11:56,450 --> 00:11:58,290

is one of the most

common drivers we

 

233

00:11:58,290 --> 00:12:01,270

hear for automation.

You can see in the

 

234

00:12:01,270 --> 00:12:03,670

slide some of the

business use cases that

 

235

00:12:03,670 --> 00:12:06,050

are top of mind.

Autonomous coding,

 

236

00:12:06,070 --> 00:12:09,050

digital intake and

triage, chatbot

 

237

00:12:09,050 --> 00:12:11,990

support, ambient

listening are just a few.

 

238

00:12:12,170 --> 00:12:15,130

On the right, you

can see the impact of

 

239

00:12:15,130 --> 00:12:17,950

AI by level of

adoption, right? We all

 

240

00:12:17,950 --> 00:12:20,790

know we are in what

Gartner would tell us

 

241

00:12:20,790 --> 00:12:24,290

is the hype curve of

AI, but ultimately

 

242

00:12:24,290 --> 00:12:26,650

adoption is happening

at a very different

 

243

00:12:26,650 --> 00:12:30,330

scale than the

interest in the hype.

 

244

00:12:30,630 --> 00:12:34,110

Key here is this is a

journey. An organization

 

245

00:12:34,110 --> 00:12:37,490

should have a 5

to 10 year plan as AI

 

246

00:12:37,490 --> 00:12:39,850

matures. It doesn't

mean you wait 5 to

 

247

00:12:39,850 --> 00:12:42,770

10 years to implement,

but you recognize to

 

248

00:12:42,770 --> 00:12:44,970

reach the full cost

benefit you're looking

 

249

00:12:44,970 --> 00:12:48,190

for will evolve over

years. it'll evolve over

 

250

00:12:48,190 --> 00:12:50,990

years because the

technologies will get better

 

251

00:12:50,990 --> 00:12:53,630

it will evolve over

years as you learn

 

252

00:12:53,630 --> 00:12:56,570

to see the barriers of

automation within your

 

253

00:12:56,570 --> 00:13:00,430

own organization and

as adoption increases.

 

254

00:13:00,510 --> 00:13:03,130

And so as we think

about this 5 to

 

255

00:13:03,130 --> 00:13:06,730

10 year plan as AI

matures and organizations

 

256

00:13:06,730 --> 00:13:08,750

become more

comfortable with

 

257

00:13:08,750 --> 00:13:11,690

automation, you will

start to achieve the

 

258

00:13:11,690 --> 00:13:14,490

full benefits and then

by then, we'll have

 

259

00:13:14,490 --> 00:13:16,710

even new capabilities

in the market.

 

260

00:13:20,410 --> 00:13:24,170

So as we look

at all of the

 

261

00:13:24,170 --> 00:13:27,250

different activities

that have really

 

262

00:13:27,250 --> 00:13:28,970

been positioned

for automation,

 

263

00:13:29,300 --> 00:13:33,250

I think back to, you

know, think back five,

 

264

00:13:33,250 --> 00:13:36,490

even ten years when

before we were talking

 

265

00:13:36,490 --> 00:13:40,430

about AI and using that

vernacular, we were

 

266

00:13:40,430 --> 00:13:43,350

talking about things

like NLP. We were talking

 

267

00:13:43,350 --> 00:13:46,010

about machine learning

we were leaning into

 

268

00:13:46,010 --> 00:13:49,210

technologies like

computer assisted coding.

 

269

00:13:49,210 --> 00:13:56,530

We moved into broader

use of RPA, remote process

 

270

00:13:56,530 --> 00:13:59,750

Automation, and so

that's when chat bots

 

271

00:13:59,750 --> 00:14:05,130

and other bots

really came into play. We

 

272

00:14:05,130 --> 00:14:08,790

from a banner health

perspective, we leverage

 

273

00:14:08,790 --> 00:14:12,730

bots extensively. We're

leveraging them in

 

274

00:14:12,730 --> 00:14:17,330

some of the usual areas

like claim statusing.

 

275

00:14:17,330 --> 00:14:22,670

We use chatbots in

certain locations. But I

 

276

00:14:22,670 --> 00:14:24,650

think one of the things

that's exciting that

 

277

00:14:24,650 --> 00:14:27,970

we're exploring and have

some active projects

 

278

00:14:27,970 --> 00:14:31,230

around today, is really

learning from other

 

279

00:14:31,230 --> 00:14:33,730

industries around call

center automation.

 

280

00:14:33,970 --> 00:14:36,630

This is an area where,

you know, moving

 

281

00:14:36,630 --> 00:14:40,530

from simply chatbots

into more voice

 

282

00:14:40,530 --> 00:14:45,130

-enabled bots has a

lot of promise, excuse

 

283

00:14:45,130 --> 00:14:47,480

me. And it's not

just to drive cost

 

284

00:14:47,480 --> 00:14:49,630

reduction, but the

way we see it is

 

285

00:14:49,630 --> 00:14:51,670

really improving the

patient experience.

 

286

00:14:51,970 --> 00:14:55,070

So thinking about

those agents for self

 

287

00:14:55,070 --> 00:14:58,630

-service that instead of

waiting for a call back

 

288

00:14:58,630 --> 00:15:02,670

or, are being on hold

to talk to an agent,

 

289

00:15:02,670 --> 00:15:04,670

really moving into

this space where

 

290

00:15:04,670 --> 00:15:06,570

we can use some

of those spots

 

291

00:15:06,930 --> 00:15:09,370

to really meet our

patients where they're

 

292

00:15:09,370 --> 00:15:12,710

at. We're all seeking

a more convenient,

 

293

00:15:14,630 --> 00:15:17,170

timely interaction

to get some of

 

294

00:15:17,170 --> 00:15:19,330

those basic

questions answered.

 

295

00:15:19,470 --> 00:15:21,170

And frankly,

in the future,

 

296

00:15:21,170 --> 00:15:23,090

more complex

things like being

 

297

00:15:23,090 --> 00:15:24,990

able to schedule

an appointment

 

298

00:15:25,510 --> 00:15:28,870

through a chat or a

voice bot interaction.

 

299

00:15:28,870 --> 00:15:31,010

So these are some

of the things that,

 

300

00:15:31,010 --> 00:15:32,840

again, as we look

at that continuum

 

301

00:15:32,840 --> 00:15:35,230

that Lorri was

talking about, we're

 

302

00:15:35,230 --> 00:15:38,130

getting very

excited to watch not

 

303

00:15:38,130 --> 00:15:40,290

only the evolution

of the technology,

 

304

00:15:40,440 --> 00:15:42,890

but really

imagine how we can

 

305

00:15:42,890 --> 00:15:44,690

use that technology

in ways to

 

306

00:15:44,690 --> 00:15:47,600

make our patients'

lives easier

 

307

00:15:48,390 --> 00:15:51,450

and help our

agents across

 

308

00:15:51,450 --> 00:15:53,650

Banner Health

out as well. Some

 

309

00:15:58,630 --> 00:16:00,970

of the key

considerations in framing

 

310

00:16:00,970 --> 00:16:03,610

future ambition, how

does your automation

 

311

00:16:03,610 --> 00:16:06,160

vision fit into your

overall strategic

 

312

00:16:06,160 --> 00:16:08,640

plan? How are you

going to define,

 

313

00:16:08,640 --> 00:16:10,730

measure, and achieve

success for your

 

314

00:16:10,730 --> 00:16:13,030

revenue cycle or

clinical automation?

 

315

00:16:13,250 --> 00:16:16,390

What are the key

capabilities, governance,

 

316

00:16:16,390 --> 00:16:18,550

and infrastructure

that your organization

 

317

00:16:18,550 --> 00:16:21,450

needs to improve revenue

cycle and clinical

 

318

00:16:21,450 --> 00:16:24,610

operations moving

forward? And what's the

 

319

00:16:24,610 --> 00:16:26,790

most critical for your

health organization

 

320

00:16:26,790 --> 00:16:29,610

to achieve in order

to drive long-term

 

321

00:16:29,610 --> 00:16:33,410

sustained access within

your provider operations?

 

322

00:16:33,750 --> 00:16:35,390

How are you going

to address staff

 

323

00:16:35,390 --> 00:16:38,450

workforce shortages? And

where would you redeploy

 

324

00:16:38,450 --> 00:16:40,950

your talent? Do you

have the internal

 

325

00:16:40,950 --> 00:16:43,770

resources and expertise

to achieve cost and

 

326

00:16:43,770 --> 00:16:46,070

revenue targets? And

how will you determine

 

327

00:16:46,070 --> 00:16:47,970

what use cases

provide the greatest

 

328

00:16:47,970 --> 00:16:52,090

opportunity to improve

yield via automation or

 

329

00:16:52,090 --> 00:16:55,010

vendor partnerships.

And what revenue cycle

 

330

00:16:55,010 --> 00:16:57,630

improvement opportunities

will you prioritize

 

331

00:16:57,630 --> 00:16:59,810

across your organization

and how will

 

332

00:16:59,810 --> 00:17:03,170

you assess variables to

consider when

 

333

00:17:03,170 --> 00:17:05,690

evaluating improvement

tactics. I think

 

334

00:17:05,690 --> 00:17:08,860

we all know that

in this journey comes

 

335

00:17:08,860 --> 00:17:11,310

both benefits and

risk and i think it's

 

336

00:17:11,310 --> 00:17:13,610

really important to

understand and acknowledge

 

337

00:17:13,610 --> 00:17:16,450

and document those

and to be able to

 

338

00:17:16,450 --> 00:17:19,890

strategize around each

of those points. Katie,

 

339

00:17:19,890 --> 00:17:21,810

anything else you

would want to add here?

 

340

00:17:21,810 --> 00:17:24,470

All really, really

great considerations.

 

341

00:17:24,670 --> 00:17:27,190

The one thing

I would add is,

 

342

00:17:27,350 --> 00:17:31,390

look, we know

that we have this

 

343

00:17:31,390 --> 00:17:34,210

opportunity to

automate process,

 

344

00:17:34,350 --> 00:17:37,450

but I would challenge

us to go beyond that.

 

345

00:17:37,450 --> 00:17:39,990

It's really about reimagining the process

 

346

00:17:39,990 --> 00:17:42,770

versus layering any of

these AI tools on top

 

347

00:17:42,770 --> 00:17:45,470

of our traditional

ways of working. And I

 

348

00:17:45,470 --> 00:17:47,310

think one of the greatest

features of these

 

349

00:17:47,310 --> 00:17:50,150

new technologies is the

ability to pull large

 

350

00:17:50,150 --> 00:17:52,830

data sets and use that

knowledge to drive

 

351

00:17:52,830 --> 00:17:55,250

more effective outcomes.

So whether we're

 

352

00:17:55,250 --> 00:17:58,210

talking about clean

claims, reducing denials,

 

353

00:17:58,210 --> 00:18:00,590

or really

ultimately driving

 

354

00:18:00,590 --> 00:18:02,290

better clinical outcomes,

 

355

00:18:02,650 --> 00:18:06,350

this is our opportunity

to really harness all

 

356

00:18:06,350 --> 00:18:09,250

of our creative powers

and that art of the

 

357

00:18:09,250 --> 00:18:11,210

possible And it's not

just about allowing

 

358

00:18:11,210 --> 00:18:16,300

ourselves to work in

our old ways and making

 

359

00:18:16,300 --> 00:18:18,310

sure that we're not

constrained by those

 

360

00:18:18,310 --> 00:18:22,410

limitations of past

practice. So in addition to

 

361

00:18:22,660 --> 00:18:25,210

all of the things that

you mentioned, Lorri,

 

362

00:18:25,210 --> 00:18:29,390

I would just like to

really underscore and

 

363

00:18:29,390 --> 00:18:31,930

really challenge all

of us as an industry,

 

364

00:18:32,110 --> 00:18:34,890

let's reimagine

the way we work.

 

365

00:18:34,890 --> 00:18:38,570

We really have

this opportunity. I

 

366

00:18:38,570 --> 00:18:41,550

think this is our

moment and, you know,

 

367

00:18:41,550 --> 00:18:43,830

we shouldn't squander

the opportunity.

 

368

00:18:44,490 --> 00:18:47,250

We really do have

a significant

 

369

00:18:47,250 --> 00:18:48,590

one on the horizon here.

 

370

00:18:48,920 --> 00:18:51,270

I love that feedback,

Katie, and I'll

 

371

00:18:51,270 --> 00:18:52,990

share from a

personal experience.

 

372

00:18:52,990 --> 00:18:57,590

As we were in the

market with the upcoming

 

373

00:18:57,590 --> 00:19:00,370

ICD-10 looming and

putting computer

 

374

00:19:00,370 --> 00:19:03,430

-assisted coding out

to the market, One of

 

375

00:19:03,430 --> 00:19:06,290

the things that we

see in hindsight is at

 

376

00:19:06,290 --> 00:19:08,750

the time, it was all

about organizations

 

377

00:19:08,750 --> 00:19:12,190

wanting to survive

the impact of ICD-10.

 

378

00:19:12,230 --> 00:19:16,470

And after ICD-10 was over

 

379

00:19:16,470 --> 00:19:19,310

and change had evolved,

 

380

00:19:20,020 --> 00:19:23,210

they looked back and

thought, wait, this was

 

381

00:19:23,210 --> 00:19:27,110

so much bigger than

helping survive ICD-10.

 

382

00:19:27,110 --> 00:19:30,510

And so when you think

about AI or Natural

 

383

00:19:30,510 --> 00:19:33,230

Language Processing

at the point of care,

 

384

00:19:33,230 --> 00:19:36,790

processing information,

and understanding a

 

385

00:19:36,790 --> 00:19:39,890

potential view into a

claim long before a patient

 

386

00:19:39,890 --> 00:19:42,090

leaves the hospital

and the record is final

 

387

00:19:42,090 --> 00:19:45,230

coded, you can really

think your operations.

 

388

00:19:45,230 --> 00:19:47,370

And I talk a lot about

the middle revenue

 

389

00:19:47,370 --> 00:19:51,410

cycle and how you can

shift from that reactive,

 

390

00:19:51,410 --> 00:19:54,930

why was this coded on

a case, to proactive

 

391

00:19:54,930 --> 00:19:57,090

and orchestrator of

health information

 

392

00:19:57,090 --> 00:20:00,350

every step of the way,

confirming a potential

 

393

00:20:00,350 --> 00:20:04,370

quality event, closing

a documentation gap,

 

394

00:20:04,610 --> 00:20:08,010

getting the one

piece of information

 

395

00:20:08,010 --> 00:20:10,670

needed to accurately

code that PCS code.

 

396

00:20:10,670 --> 00:20:14,050

Those things were

really overlooked in

 

397

00:20:14,050 --> 00:20:19,090

our rapid race to

survive ICD-10. But

 

398

00:20:19,090 --> 00:20:23,110

looking back, it

tells us what a lesson

 

399

00:20:23,110 --> 00:20:25,670

learned for the future,

which is how do I

 

400

00:20:25,670 --> 00:20:28,270

rethink, as Katie

said, my operations

 

401

00:20:28,270 --> 00:20:30,750

besides just thinking

automation is a

 

402

00:20:30,750 --> 00:20:33,010

great way to do it

the way we always did.

 

403

00:20:33,030 --> 00:20:33,590

Yeah.

 

404

00:20:35,350 --> 00:20:38,460

So let's talk about

some key methods for

 

405

00:20:38,460 --> 00:20:41,890

success. And so you

want to commit to a

 

406

00:20:41,890 --> 00:20:44,710

standardized implementation

process to ensure the

 

407

00:20:44,710 --> 00:20:47,630

responsible use and

value realization that

 

408

00:20:47,630 --> 00:20:50,070

you're looking for. So

you want to assess the

 

409

00:20:50,070 --> 00:20:52,910

organization's overall

preparedness for AI

 

410

00:20:52,910 --> 00:20:55,530

adoption, focusing

on infrastructure,

 

411

00:20:55,530 --> 00:20:59,770

processes, and the

culture, to ensure a smooth

 

412

00:20:59,770 --> 00:21:03,490

integration process with

existing systems. To

 

413

00:21:03,490 --> 00:21:06,770

think you can put

automation in and all of

 

414

00:21:06,770 --> 00:21:10,090

these things just fall

in line is no different

 

415

00:21:10,090 --> 00:21:12,930

than the concept of

junk in, junk out from

 

416

00:21:12,930 --> 00:21:15,590

reporting, right? We

have to be prepared

 

417

00:21:15,590 --> 00:21:18,930

to understand what are

our people's bias to

 

418

00:21:18,930 --> 00:21:21,660

automation? How do we

make them comfortable

 

419

00:21:21,660 --> 00:21:24,370

when they have been the

ones scrutinized for

 

420

00:21:24,370 --> 00:21:26,210

a year or so and the

difference between right

 

421

00:21:26,210 --> 00:21:28,630

and wrong, right?

How do we think about

 

422

00:21:28,630 --> 00:21:31,010

our documentation

differently, and what are

 

423

00:21:31,010 --> 00:21:33,270

those things that we

have in our documentation

 

424

00:21:33,270 --> 00:21:35,190

templates that are

going to make it more

 

425

00:21:35,190 --> 00:21:38,510

difficult for AI to be

accurate? So assessing

 

426

00:21:38,510 --> 00:21:41,470

the enterprise readiness

across not just

 

427

00:21:41,470 --> 00:21:43,970

the people and the

process, but the people,

 

428

00:21:43,970 --> 00:21:46,610

process, and technologies

that surround it.

 

429

00:21:46,630 --> 00:21:49,810

Your AI legal, compliance,

regulatory, and

 

430

00:21:49,810 --> 00:21:52,850

governance readiness.

We know we're early in

 

431

00:21:52,850 --> 00:21:56,010

this journey and that

regulations are going

 

432

00:21:56,010 --> 00:21:58,550

to evolve. We all

know in healthcare, it

 

433

00:21:58,550 --> 00:22:01,850

takes an incident and

suddenly new regulatory

 

434

00:22:02,070 --> 00:22:07,810

items become evident.

And so you want to have

 

435

00:22:07,810 --> 00:22:10,550

your own ability to

adhere to what you

 

436

00:22:10,550 --> 00:22:13,090

believe is the responsible

and ethical use of

 

437

00:22:13,090 --> 00:22:15,670

AI, but you also want

to make sure that you're

 

438

00:22:15,670 --> 00:22:18,430

assessing your vendor

for responsible and

 

439

00:22:18,430 --> 00:22:21,610

ethical use of AI.

Testing against that AI

 

440

00:22:21,610 --> 00:22:25,410

model's bias, being able

to detect hallucinations

 

441

00:22:25,410 --> 00:22:28,650

that may create poor

outcomes, understanding

 

442

00:22:28,650 --> 00:22:32,350

if that vendor is

producing the full

 

443

00:22:32,350 --> 00:22:35,570

human capacity when it

was being touched by a

 

444

00:22:35,570 --> 00:22:38,410

human or just parts of

that human's capacity.

 

445

00:22:38,410 --> 00:22:41,130

And so understanding

your legal, compliance,

 

446

00:22:41,130 --> 00:22:43,130

regulatory, and

governance readiness is

 

447

00:22:43,130 --> 00:22:45,770

ideal. And you should

have a way that you're

 

448

00:22:45,770 --> 00:22:49,260

assessing your vendors,

but also a way you're

 

449

00:22:49,260 --> 00:22:52,030

assessing your

organization's use of AI.

 

450

00:22:52,030 --> 00:22:55,250

AI operational

readiness, organization's

 

451

00:22:55,250 --> 00:22:58,490

capacity to manage these

solutions effectively,

 

452

00:22:58,490 --> 00:23:00,690

including monitoring,

maintenance,

 

453

00:23:00,810 --> 00:23:04,170

adoption, and optimization

to ensure continued

 

454

00:23:04,170 --> 00:23:07,090

performance and

effectiveness. One of

 

455

00:23:07,090 --> 00:23:09,370

the things everybody

hates to talk about when

 

456

00:23:09,370 --> 00:23:13,910

it comes to AI is

while we want to bring

 

457

00:23:13,910 --> 00:23:16,590

down our costs, it

becomes a much harder

 

458

00:23:16,590 --> 00:23:19,790

conversation when we

talk about losing people.

 

459

00:23:19,890 --> 00:23:23,130

And Katie was very,

very distinct in how she

 

460

00:23:23,130 --> 00:23:24,930

was thinking about

this. So when you think

 

461

00:23:24,930 --> 00:23:28,090

about your AI operations

readiness, it's

 

462

00:23:28,090 --> 00:23:29,790

equally as important to

know how you're going

 

463

00:23:29,790 --> 00:23:33,090

to redeploy talent and

especially redeploy

 

464

00:23:33,090 --> 00:23:35,610

top talent. Maybe it's

top talent to help

 

465

00:23:35,610 --> 00:23:38,690

make sure that the

AI is performing in a

 

466

00:23:38,690 --> 00:23:42,990

compliant way, but

maybe it's also helping

 

467

00:23:42,990 --> 00:23:45,130

to reach goals that you

haven't been able to

 

468

00:23:45,130 --> 00:23:46,810

reach in your

organization because they

 

469

00:23:46,810 --> 00:23:49,930

were so busy with the

task that they couldn't

 

470

00:23:49,930 --> 00:23:51,910

focus on the outcome.

And I think that's

 

471

00:23:51,910 --> 00:23:54,310

just a really important

piece of readiness.

 

472

00:23:54,350 --> 00:23:57,010

And I think it's

important for the human,

 

473

00:23:57,010 --> 00:23:59,650

too, to know what this

means to their role.

 

474

00:24:00,050 --> 00:24:02,890

AI solutions, identify

the most suitable

 

475

00:24:02,890 --> 00:24:05,750

AI tools and

technology for your

 

476

00:24:05,750 --> 00:24:07,870

unique requirements

and challenges.

 

477

00:24:07,870 --> 00:24:11,230

I have seen healthcare

organizations

 

478

00:24:11,790 --> 00:24:14,430

lock into a deal

with a vendor,

 

479

00:24:14,550 --> 00:24:17,710

spend months, if not

years, training a

 

480

00:24:17,710 --> 00:24:20,870

model only to figure

out at the end there

 

481

00:24:20,870 --> 00:24:22,870

was not the cost-benefit they wanted

 

482

00:24:22,870 --> 00:24:26,130

to achieve. And so those

early on assessment

 

483

00:24:26,270 --> 00:24:29,450

of the right AI tools

are so critical.

 

484

00:24:29,450 --> 00:24:31,630

And then assuring

seamless deployment

 

485

00:24:31,630 --> 00:24:34,390

with existing systems

and processes.

 

486

00:24:34,410 --> 00:24:36,610

So you want to

empower with better

 

487

00:24:36,610 --> 00:24:39,610

insights, deliver

the answers faster,

 

488

00:24:39,630 --> 00:24:42,490

create seamless

experiences, and maximize

 

489

00:24:42,490 --> 00:24:45,170

your people's potential

are key takeaways.

 

490

00:24:50,230 --> 00:24:53,570

So, Lorri, you know,

as I think about

 

491

00:24:53,570 --> 00:24:56,390

these common

barriers to success,

 

492

00:24:56,430 --> 00:24:59,390

one of the things

that I think many of

 

493

00:24:59,390 --> 00:25:01,470

us are challenged

with is this notion

 

494

00:25:01,470 --> 00:25:03,830

of, you know, who

do we partner with?

 

495

00:25:03,830 --> 00:25:06,310

This is an incredibly

crowded field,

 

496

00:25:06,310 --> 00:25:08,490

and it gets more

crowded by the day.

 

497

00:25:08,770 --> 00:25:14,430

Everyone is racing to

determine how AI can

 

498

00:25:14,430 --> 00:25:18,630

improve their

operations, how they can,

 

499

00:25:18,630 --> 00:25:20,750

from a vendor

perspective, what are the

 

500

00:25:20,750 --> 00:25:26,230

new tools that can be

created, a new product

 

501

00:25:26,230 --> 00:25:28,670

that kind of spools

out of this. So super

 

502

00:25:28,670 --> 00:25:31,310

crowded field. So,

you know, as we kind

 

503

00:25:31,310 --> 00:25:33,210

of step back and

think about who are

 

504

00:25:33,210 --> 00:25:35,730

those partners that

we would like to work

 

505

00:25:35,730 --> 00:25:39,290

with, from my perspective,

I think about that

 

506

00:25:39,290 --> 00:25:41,410

adage that innovation

moves at the speed

 

507

00:25:41,410 --> 00:25:44,330

of trust. And so, who

are those partners

 

508

00:25:44,330 --> 00:25:47,190

that, for us, share

a common value set?

 

509

00:25:47,410 --> 00:25:51,230

We really focus on that

ethical use of data,

 

510

00:25:51,230 --> 00:25:54,610

the commitment to

improve and protect the

 

511

00:25:54,610 --> 00:25:57,850

patient experience.

Who are these vendors

 

512

00:25:57,850 --> 00:26:01,390

that are really focused

on data integrity?

 

513

00:26:01,390 --> 00:26:04,530

All of these things

are what is going to

 

514

00:26:04,530 --> 00:26:07,650

engender trust, not

just, you know, within

 

515

00:26:07,650 --> 00:26:11,990

our IT organizations or

operations, but within

 

516

00:26:11,990 --> 00:26:15,810

our talent pool, our,

you know, employee

 

517

00:26:15,810 --> 00:26:19,670

base, from our

executive teams, and our

 

518

00:26:19,670 --> 00:26:22,410

patients. You know, I

think that there's a

 

519

00:26:22,410 --> 00:26:26,230

lot of anxiety out

there about the use of

 

520

00:26:26,230 --> 00:26:28,630

these technologies,

and so it's really

 

521

00:26:28,630 --> 00:26:32,050

incumbent upon us to get

to that place of trust

 

522

00:26:32,050 --> 00:26:35,110

and make sure that we're

identifying partners

 

523

00:26:35,110 --> 00:26:40,270

who really share

that vision and who

 

524

00:26:40,270 --> 00:26:42,910

we can trust to go at

this journey with us.

 

525

00:26:46,460 --> 00:26:49,090

so we talked a little

bit about the concept

 

526

00:26:49,090 --> 00:26:51,750

of junk in and junk

out and the truth

 

527

00:26:51,750 --> 00:26:55,410

is none of these AI

models are perfect on

 

528

00:26:55,410 --> 00:26:59,650

their own. Clinical

documentation is complex

 

529

00:26:59,650 --> 00:27:02,930

and health care

regulations are complex

 

530

00:27:02,930 --> 00:27:07,510

coding language is

complex. This article

 

531

00:27:07,510 --> 00:27:10,250

is a great example

about how subtle shifts

 

532

00:27:10,250 --> 00:27:12,830

can create grossly

inaccurate outputs and

 

533

00:27:12,830 --> 00:27:15,510

i thought it was just

a fun article to share

 

534

00:27:15,510 --> 00:27:18,970

with you so this is

actually a scientific

 

535

00:27:18,970 --> 00:27:23,690

paper that was

published on the internet

 

536

00:27:23,690 --> 00:27:27,210

and recently this

article was published

 

537

00:27:27,210 --> 00:27:33,870

that talked about how

AI created its own

 

538

00:27:33,870 --> 00:27:38,990

language because of

how this presented in

 

539

00:27:38,990 --> 00:27:41,910

its internet source.

And so as you see, this

 

540

00:27:41,910 --> 00:27:44,430

article was meant to

be read in columns.

 

541

00:27:44,430 --> 00:27:47,130

But what happened

is because it

 

542

00:27:47,130 --> 00:27:49,290

was side-by-side

in the way that

 

543

00:27:49,290 --> 00:27:53,190

it was presented

on the internet,

 

544

00:27:53,510 --> 00:27:57,670

it actually read across

the columns combining

 

545

00:27:57,670 --> 00:28:02,670

words to create a

term that didn't exist

 

546

00:28:02,670 --> 00:28:05,990

in this industry.

And the reason behind

 

547

00:28:05,990 --> 00:28:10,190

this, right, is it couldn't

tell the difference

 

548

00:28:10,190 --> 00:28:14,250

between these two

columns read across

 

549

00:28:14,250 --> 00:28:17,990

the page. And ultimately,

this term has now

 

550

00:28:17,990 --> 00:28:21,190

been cited, I think it

was over 10 times in

 

551

00:28:21,190 --> 00:28:26,870

actual journal articles

because of this one

 

552

00:28:26,870 --> 00:28:30,250

misreading an

application by AI. So I

 

553

00:28:30,250 --> 00:28:32,390

thought that was an

interesting story. So when

 

554

00:28:32,390 --> 00:28:35,950

we start to step back

and look at our clinical

 

555

00:28:35,950 --> 00:28:38,230

documentation, right,

and we're so quick

 

556

00:28:38,230 --> 00:28:41,190

to insert things into

our templates in a

 

557

00:28:41,190 --> 00:28:43,310

rush because suddenly

we're part of a new

 

558

00:28:43,310 --> 00:28:46,610

quality program or

this physician wanted

 

559

00:28:46,610 --> 00:28:49,330

this box in this place

versus this place. And

 

560

00:28:49,330 --> 00:28:51,910

as we think about our

AI journey, part of

 

561

00:28:51,910 --> 00:28:55,230

that is understanding

how do we create the

 

562

00:28:55,230 --> 00:28:59,750

best possible content

to lay AI on top of,

 

563

00:28:59,750 --> 00:29:01,750

whether that is AI

on top of clinical

 

564

00:29:01,750 --> 00:29:06,740

information or AI

to translate

clinical information,

 

565

00:29:06,740 --> 00:29:09,970

it's important that

we build the best

 

566

00:29:09,970 --> 00:29:12,510

documentation structure

to help create the

 

567

00:29:12,510 --> 00:29:14,750

outcome we're looking

for with automation.

 

568

00:29:18,930 --> 00:29:22,110

So some key focus

areas for success I

 

569

00:29:22,110 --> 00:29:24,250

just wanted to

reiterate these, we've

 

570

00:29:24,250 --> 00:29:27,230

touched on them

already, but I think

 

571

00:29:27,230 --> 00:29:30,450

they're worth revisiting.

The people. Don't

 

572

00:29:30,450 --> 00:29:32,790

assume if you build

it they will come.

 

573

00:29:32,790 --> 00:29:35,110

Their perception

of AI is important

 

574

00:29:35,110 --> 00:29:39,810

the threat of AI on

jobs is top of mind

 

575

00:29:40,110 --> 00:29:43,490

to include my own of

course adoption of AI

 

576

00:29:43,490 --> 00:29:47,690

doesn't happen just

because we go live. How

 

577

00:29:47,690 --> 00:29:50,030

we're redeploying the

talent, as I mentioned

 

578

00:29:50,030 --> 00:29:52,550

before, how do we

educate? And what's

 

579

00:29:52,550 --> 00:29:54,830

the executive level

support to help them

 

580

00:29:54,830 --> 00:29:57,430

understand the vision

behind your automation

 

581

00:29:57,430 --> 00:30:00,630

strategy? We've

talked a lot about the

 

582

00:30:00,630 --> 00:30:04,550

importance of the experts

in their fields to

 

583

00:30:04,550 --> 00:30:07,090

help create the right

automation outcomes

 

584

00:30:07,090 --> 00:30:10,590

that we're looking

for. And how do we help

 

585

00:30:10,590 --> 00:30:14,630

to educate them

earlier on, help them

 

586

00:30:14,630 --> 00:30:17,430

understand their role

with the technology in

 

587

00:30:17,430 --> 00:30:20,480

place and the vision

at an executive level

 

588

00:30:20,480 --> 00:30:23,570

to help create the

buy-in. And so I love

 

589

00:30:23,570 --> 00:30:26,270

something Katie said

earlier about whether

 

590

00:30:26,270 --> 00:30:29,790

you do it for business

or personal reasons,

 

591

00:30:30,050 --> 00:30:33,370

actually engaging with

these Large Language

 

592

00:30:33,370 --> 00:30:37,810

Models will help them

see the vision of their

 

593

00:30:37,810 --> 00:30:41,910

role in this new

automated world. And I'll

 

594

00:30:41,910 --> 00:30:44,130

give you an example,

a simple example, where

 

595

00:30:44,130 --> 00:30:48,810

i can ask chap gpt to

tell me the code for

 

596

00:30:48,810 --> 00:30:52,770

a decubitus ulcer but

that doesn't and it

 

597

00:30:52,770 --> 00:30:55,150

will be correct and it

will be fast but that

 

598

00:30:55,150 --> 00:30:58,350

doesn't mean that it's

accurate because to

 

599

00:30:58,350 --> 00:31:01,310

appropriately code a

decubitus ulcer i need to

 

600

00:31:01,310 --> 00:31:04,390

know where it was

located on the body, what

 

601

00:31:04,390 --> 00:31:07,270

side of the body it

was located on, and the

 

602

00:31:07,270 --> 00:31:09,850

staging of that wound.

So when they start

 

603

00:31:09,850 --> 00:31:12,270

to see the importance

of asking the right

 

604

00:31:12,270 --> 00:31:16,450

question to get the

desired answer at a level

 

605

00:31:16,450 --> 00:31:19,210

of accuracy that works

for health care, it

 

606

00:31:19,210 --> 00:31:21,890

helps them envision

their role in the future.

 

607

00:31:21,890 --> 00:31:24,930

But it also helps you

plan for what it's

 

608

00:31:24,930 --> 00:31:28,110

good at and the things

that it's not good at.

 

609

00:31:28,230 --> 00:31:33,090

I recently used it to

try to help me write

 

610

00:31:33,090 --> 00:31:35,810

a video presentation.

And I quickly realized

 

611

00:31:35,810 --> 00:31:39,050

I right now am better

than AI at creating

 

612

00:31:39,050 --> 00:31:41,210

my own video

presentation so i think

 

613

00:31:41,210 --> 00:31:43,950

it's really interesting

to engage with it

 

614

00:31:43,950 --> 00:31:47,590

Yourself. The documentation

and data right

 

615

00:31:47,590 --> 00:31:51,490

there are still check

boxes that exist within

 

616

00:31:51,490 --> 00:31:53,690

many templates there's

still handwriting

 

617

00:31:53,690 --> 00:31:57,510

that exists. Fishbone

diagrams. Physicians

 

618

00:31:57,510 --> 00:32:00,070

love to document their

lab findings in the

 

619

00:32:00,070 --> 00:32:03,490

fishbone diagram. AI

has a really tough time

 

620

00:32:03,490 --> 00:32:06,150

with fish bones. Or

what about templates

 

621

00:32:06,150 --> 00:32:08,590

like the one on the

right where they say bold

 

622

00:32:08,590 --> 00:32:12,970

indicates a

positive and all else

 

623

00:32:12,970 --> 00:32:15,330

should be considered

negative. That's not going

 

624

00:32:15,330 --> 00:32:17,890

to work here. Clear

headers. Understanding

 

625

00:32:17,890 --> 00:32:20,490

where different

parts of your

 

626

00:32:20,490 --> 00:32:23,410

documentation start and

where

 

627

00:32:23,410 --> 00:32:25,790

it should end so we

can appropriately point

 

628

00:32:25,790 --> 00:32:29,250

AI to where in this

large amount of

 

629

00:32:29,250 --> 00:32:33,170

clinical data it's

appropriate to analyze

 

630

00:32:33,170 --> 00:32:36,170

and draw answers from

for your questions.

 

631

00:32:36,530 --> 00:32:39,990

Clear headers,

punctuation matters. Are

 

632

00:32:39,990 --> 00:32:44,450

we using a colon,

semicolon, the way we

 

633

00:32:44,450 --> 00:32:46,590

should so that it

knows it's at the end

 

634

00:32:46,590 --> 00:32:49,210

of a statement and

we don't create new

 

635

00:32:49,210 --> 00:32:51,730

terms, as we saw in

the previous article.

 

636

00:32:51,730 --> 00:32:54,290

And then insertions

of prompt

 

637

00:32:54,290 --> 00:32:56,850

content. So adding

to all that, how

 

638

00:32:56,850 --> 00:32:59,890

do we lay prompting

content on top

 

639

00:32:59,890 --> 00:33:02,230

of these Large

Language Models?

 

640

00:33:05,010 --> 00:33:07,850

Choosing the right

tool for your use case.

 

641

00:33:07,850 --> 00:33:10,810

you know for my

personal journey as a

 

642

00:33:10,810 --> 00:33:13,690

product management leader,

we were moving from

 

643

00:33:13,690 --> 00:33:16,310

a computer-assisted

coding world to this

 

644

00:33:16,310 --> 00:33:19,730

new autonomous coding

world and we ultimately

 

645

00:33:19,730 --> 00:33:22,090

started off with

a bake-off what do

 

646

00:33:22,090 --> 00:33:25,610

these models do better

than what we do today

 

647

00:33:25,610 --> 00:33:28,690

and so when we started

to compare those, like

 

648

00:33:28,690 --> 00:33:30,650

we talked about earlier,

there's

 

649

00:33:30,650 --> 00:33:32,950

a knowledge of the

internet scale that

 

650

00:33:32,950 --> 00:33:36,110

these Large Language

Models not only access,

 

651

00:33:36,380 --> 00:33:40,370

but they can rapidly

draw conclusions from

 

652

00:33:40,370 --> 00:33:42,970

that large amount

of data. The language

 

653

00:33:42,970 --> 00:33:45,810

understanding fuzzy

context-based understanding

 

654

00:33:45,810 --> 00:33:48,770

of text and natural

language prompts and the

 

655

00:33:48,770 --> 00:33:51,870

reasoning is an

ability to form limited

 

656

00:33:51,870 --> 00:33:55,430

reasoning on simple

multi-step problems. On

 

657

00:33:55,430 --> 00:33:59,910

the right-hand side

we believe that from a

 

658

00:33:59,910 --> 00:34:04,370

symbolic AI perspective,

that it's great at

 

659

00:34:04,370 --> 00:34:07,710

explainable and unverifiable

clinical detail,

 

660

00:34:07,810 --> 00:34:10,650

that the technologies

are supported because

 

661

00:34:10,650 --> 00:34:13,350

we have years of

cumulative experience

 

662

00:34:13,350 --> 00:34:17,110

reading and applying

context or translations

 

663

00:34:17,110 --> 00:34:20,510

to clinical

documentation. We use SME

 

664

00:34:20,510 --> 00:34:23,250

curated and clinical

knowledge resources

 

665

00:34:23,250 --> 00:34:25,870

versus large scale

of internet to draw

 

666

00:34:25,870 --> 00:34:29,130

conclusions from and then

that experience extensive

 

667

00:34:29,130 --> 00:34:32,870

breadth of domain

coverage. But what i

 

668

00:34:32,870 --> 00:34:35,870

wanted to tell you

here is that it wasn't

 

669

00:34:35,870 --> 00:34:39,330

about choosing one

model over the other in

 

670

00:34:39,330 --> 00:34:42,630

the end. It was about

a combination of

 

671

00:34:42,630 --> 00:34:47,650

tools that created

monumental differences in

 

672

00:34:47,650 --> 00:34:50,270

our results. The Large

Language Model on

 

673

00:34:50,270 --> 00:34:53,970

its own did not

outperform. But symbolic AI

 

674

00:34:53,970 --> 00:34:57,370

reached a peak, but

didn't scale over time

 

675

00:34:57,370 --> 00:35:00,290

as quickly as it should.

And so this hybrid

 

676

00:35:00,290 --> 00:35:02,590

AI approach means

more than one thing.

 

677

00:35:02,590 --> 00:35:04,110

And I think this is

a really interesting

 

678

00:35:04,110 --> 00:35:06,670

conversation as you're

evaluating vendors.

 

679

00:35:06,910 --> 00:35:10,650

A) as we talked about

earlier, technology is

 

680

00:35:10,650 --> 00:35:12,570

moving at a rapid

pace. So you don't want

 

681

00:35:12,570 --> 00:35:17,530

to be locked in to one

AI approach that can't

 

682

00:35:17,530 --> 00:35:20,250

evolve on pace with

this rapidly moving. So

 

683

00:35:20,250 --> 00:35:21,930

what if something better

comes out tomorrow

 

684

00:35:21,930 --> 00:35:24,250

Are you locked into 10

years with that vendor

 

685

00:35:24,250 --> 00:35:27,010

do they have the

investments to be able to

 

686

00:35:27,010 --> 00:35:30,570

take in and retire a

model or combine a model

 

687

00:35:30,570 --> 00:35:33,470

as we're talking about?

And then ultimately

 

688

00:35:33,470 --> 00:35:37,970

not just bringing one

weapon to our five but

 

689

00:35:37,970 --> 00:35:40,450

bringing multiple

things because the truth

 

690

00:35:40,450 --> 00:35:44,530

is each of these

different models have the

 

691

00:35:44,530 --> 00:35:47,810

ability to bring a

benefit to solving the

 

692

00:35:47,810 --> 00:35:50,330

larger picture of

healthcare use cases we saw

 

693

00:35:50,330 --> 00:35:52,990

on that other page.

One of the things we

 

694

00:35:52,990 --> 00:35:55,990

discovered in the power

of combination of these

 

695

00:35:55,990 --> 00:35:59,750

models is that if we

can create, take this

 

696

00:35:59,750 --> 00:36:03,410

large clinical record

and create a more

 

697

00:36:03,410 --> 00:36:07,130

condensed record of the

things that are pertinent

 

698

00:36:07,130 --> 00:36:10,430

for a use case, that

we can understand and

 

699

00:36:10,430 --> 00:36:14,050

interpret terms that

we can link terms to

 

700

00:36:14,050 --> 00:36:16,330

get to the right answer,

like I was talking

 

701

00:36:16,330 --> 00:36:19,150

about with that one

decubitus ulcer, that

 

702

00:36:19,150 --> 00:36:21,730

there is an ability to

create an output that

 

703

00:36:21,730 --> 00:36:24,410

not one of these models

can do on their own.

 

704

00:36:24,620 --> 00:36:26,850

And so I'll give you

an example. This is

 

705

00:36:26,850 --> 00:36:28,730

an example of what

we call our knowledge

 

706

00:36:28,730 --> 00:36:32,690

graph. So this is

where CLI, our Natural

 

707

00:36:32,690 --> 00:36:35,110

Language Processing

and knowledge graph

 

708

00:36:35,730 --> 00:36:39,810

created, gave the LLM

a better chance to

 

709

00:36:39,810 --> 00:36:42,430

derive the right

answer. So instead of

 

710

00:36:42,430 --> 00:36:45,170

presenting this large

amount of chart text,

 

711

00:36:45,170 --> 00:36:48,710

we processed that

chart text first, we

 

712

00:36:48,710 --> 00:36:51,230

extracted the important

elements for the

 

713

00:36:51,230 --> 00:36:54,590

use case, created the

linkages needed to

 

714

00:36:54,590 --> 00:36:57,330

get to the right

answer, and pre-auth is

 

715

00:36:57,330 --> 00:36:59,770

a great example of

where this works well,

 

716

00:36:59,770 --> 00:37:03,250

so that then the Large

Language Model can

 

717

00:37:03,250 --> 00:37:06,670

come in with its

strengths and help drive

 

718

00:37:06,670 --> 00:37:08,890

the outcome that

you're looking for. So

 

719

00:37:15,250 --> 00:37:18,910

this is a great example

um of sorry excuse

 

720

00:37:18,910 --> 00:37:23,150

me this is an

example of automation

 

721

00:37:23,150 --> 00:37:25,310

and a process to

help you achieve

 

722

00:37:25,310 --> 00:37:27,270

Success. You want to

have your strategy

 

723

00:37:27,270 --> 00:37:29,850

enablement know your

strategy when you go in.

 

724

00:37:30,370 --> 00:37:33,190

Do the data analysis

are you ready

 

725

00:37:33,190 --> 00:37:37,310

for applying AI to

your data how is

 

726

00:37:37,310 --> 00:37:39,590

this vendor going

to use your data how

 

727

00:37:39,590 --> 00:37:41,790

are you making sure

you protect your

 

728

00:37:41,790 --> 00:37:44,610

Data, what's the

value of your data?

 

729

00:37:45,290 --> 00:37:48,330

Automation development,

ROI forecasting

 

730

00:37:48,330 --> 00:37:51,210

and recognition,

workflow standardization,

 

731

00:37:51,210 --> 00:37:53,230

process

improvements, leading

 

732

00:37:53,230 --> 00:37:56,430

practice embedment

and functional

 

733

00:37:56,430 --> 00:37:58,990

team centralization

or reassignment.

 

734

00:37:59,310 --> 00:38:02,970

Automate and QC these

models are really

 

735

00:38:02,970 --> 00:38:06,020

Important. And then finally

performance measurement

 

736

00:38:06,020 --> 00:38:08,990

We may have lost

you. Are you there?

 

737

00:38:10,110 --> 00:38:11,710

I am here.

 

738

00:38:14,250 --> 00:38:15,130

Katie,

 

739

00:38:24,550 --> 00:38:25,670

can you hear me now?

 

740

00:38:31,850 --> 00:38:33,290

Okay. It looks like Katie

 

741

00:38:33,290 --> 00:38:35,570

lost me. I'll

keep going here.

 

742

00:38:36,930 --> 00:38:39,070

And so in the next slide,

 

743

00:38:40,490 --> 00:38:44,010

vendor accountability.

So how are you

 

744

00:38:44,010 --> 00:38:46,570

assessing the accountability

of your vendors,

 

745

00:38:46,570 --> 00:38:48,770

compliance and

responsible use of AI?

 

746

00:38:48,830 --> 00:38:53,410

We at Optum have an

AI review board that

 

747

00:38:53,410 --> 00:38:57,450

we use to assess not

only our technology

 

748

00:38:57,450 --> 00:38:59,970

for the outcomes

we're committing to

 

749

00:38:59,970 --> 00:39:03,730

the market, but we

also assess for things

 

750

00:39:03,730 --> 00:39:06,690

like bias across

different populations,

 

751

00:39:06,850 --> 00:39:13,010

compliance with regulatory

terms. And

 

752

00:39:13,010 --> 00:39:16,450

Processes. So when

you think of responsible

 

753

00:39:16,450 --> 00:39:19,410

use of AI a few things.

Building, deploying,

 

754

00:39:19,410 --> 00:39:23,610

managing and scaling

responsibility review

 

755

00:39:23,610 --> 00:39:26,030

and approval by a

steering committee. AI

 

756

00:39:26,030 --> 00:39:29,410

Scalability, standard

operating procedures

 

757

00:39:29,410 --> 00:39:32,530

model inventory

management, responsible and

 

758

00:39:32,530 --> 00:39:36,580

sustainable AI to ensure

adaptability, resilience,

 

759

00:39:36,580 --> 00:39:39,490

reliability and

robustness. And alignment

 

760

00:39:39,490 --> 00:39:43,030

with the human value.

From an AI governance,

 

761

00:39:43,030 --> 00:39:45,050

it's about compliance

with data protection

 

762

00:39:45,050 --> 00:39:48,050

regulations, data

quality and security,

 

763

00:39:48,410 --> 00:39:51,090

explainability and

transparency, and

 

764

00:39:51,090 --> 00:39:53,770

ongoing evaluation

for hallucinations

 

765

00:39:53,770 --> 00:39:55,630

in these models.

We want to address

 

766

00:39:55,630 --> 00:39:57,830

bias and fairness,

accountability,

 

767

00:39:57,830 --> 00:40:01,250

and disciplinary measures

for noncompliance,

 

768

00:40:01,250 --> 00:40:03,990

mechanisms for

addressing concerns.

 

769

00:40:03,990 --> 00:40:06,770

And then the AI center

of excellence is

 

770

00:40:06,770 --> 00:40:09,210

really about leadership

alignment to provide

 

771

00:40:09,210 --> 00:40:12,030

advocacy, strategic

direction and oversight.

 

772

00:40:12,030 --> 00:40:17,250

Robust data infrastructure,

agile methodology,

 

773

00:40:17,250 --> 00:40:20,410

knowledge management

processes, seamless

 

774

00:40:20,410 --> 00:40:23,670

Collaboration, training

Programs, strategic

 

775

00:40:23,670 --> 00:40:26,190

partnerships,

stakeholder engagement,

 

776

00:40:26,190 --> 00:40:30,270

assigning your KPIs

and defining those. And

 

777

00:40:30,270 --> 00:40:34,050

continuous evaluation

of those and then the

 

778

00:40:34,050 --> 00:40:36,830

culture of innovation

and research. Katie,

 

779

00:40:36,830 --> 00:40:39,250

anything you would want

to add to this slide?

 

780

00:40:45,130 --> 00:40:46,850

I'm going to keep

us moving. We seem

 

781

00:40:46,850 --> 00:40:48,770

to have some technical

difficulties.

 

782

00:40:52,450 --> 00:40:55,990

So successfully achieving

ROI from automation

 

783

00:40:55,990 --> 00:40:58,790

is a carefully

calculated journey. It is

 

784

00:40:58,790 --> 00:41:02,060

so attractive when you

look at the potential

 

785

00:41:02,060 --> 00:41:05,590

of automation and

the temptation is to

 

786

00:41:05,590 --> 00:41:09,130

dive in. But over 50% of

automation investments

 

787

00:41:09,130 --> 00:41:13,250

fail to achieve their

desired ROI. You

 

788

00:41:13,250 --> 00:41:16,610

need to be strategic and

diligent in identifying

 

789

00:41:16,610 --> 00:41:19,870

your use cases, test

your hypotheses.

 

790

00:41:19,990 --> 00:41:25,190

You know, as we look

at automating coding,

 

791

00:41:25,190 --> 00:41:27,970

one of the things we see

is it's very different

 

792

00:41:27,970 --> 00:41:31,230

based on the volume

of cases, what your

 

793

00:41:31,230 --> 00:41:35,430

cost to code a case

is, and ultimately what

 

794

00:41:35,430 --> 00:41:38,850

the risks are if you

bypass the human.

 

795

00:41:38,850 --> 00:41:41,310

And so we talk a lot

with data scientists

 

796

00:41:41,310 --> 00:41:46,470

about it's not just about

how to automate what

 

797

00:41:46,470 --> 00:41:50,510

has traditionally

taken a human role to

 

798

00:41:50,510 --> 00:41:54,730

perform, but it's also

about training the

 

799

00:41:54,730 --> 00:41:59,130

models to understand

when a human should touch

 

800

00:41:59,190 --> 00:42:03,370

a case. So the best

AI strategy isn't just

 

801

00:42:03,370 --> 00:42:06,430

about AI, and I think

that's so important.

 

802

00:42:07,290 --> 00:42:09,570

Maximizing the value

of your investments

 

803

00:42:09,570 --> 00:42:12,390

in automation and unlocking

the full potential

 

804

00:42:12,390 --> 00:42:16,510

of the AI you're

investing in. Is early

 

805

00:42:16,510 --> 00:42:18,830

adoption of AI

technologies worth the

 

806

00:42:18,830 --> 00:42:21,070

risk for you? It's

changing the health care

 

807

00:42:21,070 --> 00:42:23,510

industry but not all

organizations are

 

808

00:42:23,510 --> 00:42:26,110

adopting at the same

pace. And Katie talked a

 

809

00:42:26,110 --> 00:42:28,970

little bit about that

earlier and learning

 

810

00:42:28,970 --> 00:42:31,210

from other industries

that may have been

 

811

00:42:31,210 --> 00:42:34,410

using AI longer. And

then AI will be a

 

812

00:42:34,410 --> 00:42:36,570

key differentiator

for healthcare. It's

 

813

00:42:36,570 --> 00:42:39,270

likely not going away.

But I do believe we'll

 

814

00:42:39,270 --> 00:42:41,950

see use cases that

work extremely well,

 

815

00:42:42,150 --> 00:42:45,250

use cases that are

not quite ready. And

 

816

00:42:45,250 --> 00:42:47,690

ultimately, I think whether

you're the vendor or

 

817

00:42:47,690 --> 00:42:50,250

the healthcare

organization, we'll see a

 

818

00:42:50,250 --> 00:42:52,690

need to evolve our

people, our process, and

 

819

00:42:52,690 --> 00:42:56,370

our technologies to

reach the final journey.

 

820

00:42:56,370 --> 00:42:58,750

Katie, would you like

to add anything here?

 

821

00:42:59,590 --> 00:43:02,370

Well, first, I'd like

to add to everyone.

 

822

00:43:02,370 --> 00:43:04,810

I apologize for the

technology gremlins

 

823

00:43:04,810 --> 00:43:06,710

that we've had today.

It seems that both

 

824

00:43:06,710 --> 00:43:09,690

Lorri and I have had

a bit of a hiccup.

 

825

00:43:10,150 --> 00:43:13,970

You know, if you'd

allow me to, I'd love to

 

826

00:43:13,970 --> 00:43:16,350

go back and talk a

little bit just about

 

827

00:43:16,870 --> 00:43:19,650

governance. I think this

is one of the things

 

828

00:43:19,650 --> 00:43:23,190

that I'd love to

share with the

 

829

00:43:23,190 --> 00:43:26,030

group, just where

we are along our

 

830

00:43:26,030 --> 00:43:28,630

journey because I

really think as an

 

831

00:43:28,630 --> 00:43:32,630

Industry, and what

I'll call early innings

 

832

00:43:32,630 --> 00:43:35,230

Here, and it's one of

the things that will

 

833

00:43:35,230 --> 00:43:38,490

get better if we

share. So from a

 

834

00:43:38,490 --> 00:43:42,550

Banner perspective what

we've done thus far,

 

835

00:43:42,550 --> 00:43:46,410

we've really created

a process where we

 

836

00:43:46,410 --> 00:43:48,890

have our process

start with vendor

 

837

00:43:48,890 --> 00:43:51,800

evaluation, an AI evaluation.

It's a questionnaire.

 

838

00:43:51,800 --> 00:43:54,470

It's really 25

questions across five

 

839

00:43:54,470 --> 00:43:57,290

different categories.

And those categories

 

840

00:43:57,290 --> 00:44:01,160

are around usefulness,

usability, and efficacy.

 

841

00:44:01,210 --> 00:44:04,110

So we ask in that

section questions

 

842

00:44:04,110 --> 00:44:06,670

like, how have you

addressed potential concerns

 

843

00:44:06,670 --> 00:44:09,890

around AI reliability,

bias, fairness.

 

844

00:44:10,050 --> 00:44:12,930

Explain the measures

that are implemented

 

845

00:44:12,930 --> 00:44:15,150

to build user trust.

So a lot of the things

 

846

00:44:15,150 --> 00:44:17,310

that we've talked about

on this webinar

 

847

00:44:17,310 --> 00:44:20,750

thus far, around

trust and

 

848

00:44:20,750 --> 00:44:25,170

ethical uses of AI.

Another category we look

 

849

00:44:25,170 --> 00:44:28,290

at is fairness equity

and bias management

 

850

00:44:28,290 --> 00:44:31,710

so asking questions

like, how have you

 

851

00:44:31,710 --> 00:44:34,210

comprehensively identified

potential sources

 

852

00:44:34,210 --> 00:44:36,790

of bias,

mapped out socio

 

853

00:44:36,790 --> 00:44:39,670

-demographic groups at

Risk, developed robust

 

854

00:44:39,670 --> 00:44:42,270

strategies to proactively

monitor, detect

 

855

00:44:42,270 --> 00:44:45,410

and mitigate potential

discriminatory impacts

 

856

00:44:45,410 --> 00:44:48,190

across different

population segments.

 

857

00:44:48,410 --> 00:44:51,630

So, you know, Lorri

talked about, you

 

858

00:44:51,630 --> 00:44:55,150

know, before, it's not

just about, you know,

 

859

00:44:55,150 --> 00:44:57,190

do we have

hallucination, but are

 

860

00:44:57,190 --> 00:45:01,450

there any biases that

were not intended but

 

861

00:45:01,450 --> 00:45:03,910

are showing up

because of the use of

 

862

00:45:03,910 --> 00:45:07,050

that AI? How do we

monitor for that so

 

863

00:45:07,050 --> 00:45:10,450

that we have awareness

before we get to

 

864

00:45:10,450 --> 00:45:13,130

an outcome that we

weren't anticipating.

 

865

00:45:13,470 --> 00:45:16,370

Another area we look

at is safety and

 

866

00:45:16,370 --> 00:45:19,690

reliability. So questions

around like user

 

867

00:45:19,690 --> 00:45:21,910

controls and safety

mechanisms. What

 

868

00:45:21,910 --> 00:45:24,430

safety guards have

you put in place to

 

869

00:45:24,430 --> 00:45:26,710

make sure that there's

human oversight?

 

870

00:45:27,580 --> 00:45:29,620

We also look at

transparency,

 

871

00:45:30,310 --> 00:45:33,090

interoperability,

and accountability.

 

872

00:45:33,670 --> 00:45:35,830

Questions like

describing your

 

873

00:45:35,830 --> 00:45:38,130

comprehensive documentation

strategy. Lorri

 

874

00:45:38,130 --> 00:45:40,610

talked about that

earlier, including

 

875

00:45:40,610 --> 00:45:42,510

how will you

maintain transparency

 

876

00:45:42,510 --> 00:45:45,470

about the provenance

of that data,

 

877

00:45:45,470 --> 00:45:48,730

your model limitations,

decision thresholds.

 

878

00:45:48,730 --> 00:45:51,570

And then finally,

a lot of questions

 

879

00:45:51,570 --> 00:45:54,770

around security and

privacy. So ensuring

 

880

00:45:54,770 --> 00:45:57,590

that we've got third

-party risk management.

 

881

00:45:58,050 --> 00:46:01,730

These are things that

I think it's important

 

882

00:46:01,730 --> 00:46:05,810

from a governance

perspective that, A) we're

 

883

00:46:05,810 --> 00:46:09,470

understanding how

our vendors are

 

884

00:46:09,470 --> 00:46:13,470

leveraging this AI,

what are the controls that

 

885

00:46:13,470 --> 00:46:16,230

they have in place

and then from there we

 

886

00:46:16,230 --> 00:46:20,730

create a score on all

the information that we

 

887

00:46:20,730 --> 00:46:25,030

gleaned from this

questionnaire, we create

 

888

00:46:25,030 --> 00:46:27,630

a score so that we

understand where we're in

 

889

00:46:27,630 --> 00:46:30,110

alignment with our

standards and where there's

 

890

00:46:30,110 --> 00:46:33,410

some gaps and we have

an AI committee that

 

891

00:46:33,410 --> 00:46:36,650

reviews all of the

use of AI. They review

 

892

00:46:36,650 --> 00:46:41,250

everything that comes

through for new contracting.

 

893

00:46:41,270 --> 00:46:44,390

And there's robust

conversation around

 

894

00:46:44,390 --> 00:46:47,170

where those gaps are,

the use of our data,

 

895

00:46:47,390 --> 00:46:51,290

any gaps and

controls. So I wanted

 

896

00:46:51,290 --> 00:46:53,800

to share this

because I think it's

 

897

00:46:53,800 --> 00:46:56,250

incumbent upon

all of us to make

 

898

00:46:56,250 --> 00:46:58,860

sure that we're

continually tooling

 

899

00:46:59,690 --> 00:47:03,850

our governance,

it's for AI and

 

900

00:47:03,850 --> 00:47:05,750

for all of the

newness, right,

 

901

00:47:05,750 --> 00:47:08,590

that's coming into

our atmosphere.

 

902

00:47:08,590 --> 00:47:11,310

It's not just about,

you know, AI is

 

903

00:47:11,310 --> 00:47:13,550

going to happen. So

how do we do it right?

 

904

00:47:13,550 --> 00:47:15,490

How do we do it with

integrity? And how

 

905

00:47:15,490 --> 00:47:17,510

do we have the right

controls in place?

 

906

00:47:18,090 --> 00:47:19,330

Thank you for allowing

 

907

00:47:19,330 --> 00:47:20,790

me to go backwards, Lorri.

 

908

00:47:21,450 --> 00:47:22,710

Thank you, Katie.

 

909

00:47:26,700 --> 00:47:27,910

So as we come to the

 

910

00:47:27,910 --> 00:47:30,470

conclusion of

today's webinar,

 

911

00:47:30,470 --> 00:47:33,290

I just wanted Katie

to chime in and give

 

912

00:47:33,290 --> 00:47:35,410

us some tips and

tricks around

 

913

00:47:35,410 --> 00:47:38,110

evaluating your

automation capabilities.

 

914

00:47:39,570 --> 00:47:42,190

So, you know, we've

talked a lot about

 

915

00:47:42,190 --> 00:47:46,970

these notions of where

we can use AI to overcome

 

916

00:47:46,970 --> 00:47:49,290

some of our

documentation challenges.

 

917

00:47:49,410 --> 00:47:53,930

Lorri talked a lot

about the trust in the

 

918

00:47:53,930 --> 00:47:57,860

data. She gave a great

example of when AI

 

919

00:47:57,860 --> 00:48:02,410

might be reading things

in a way that leads

 

920

00:48:02,410 --> 00:48:05,690

to degradation in

trust, and frankly,

 

921

00:48:05,690 --> 00:48:10,510

beyond that, and outcomes

from a data perspective

 

922

00:48:10,510 --> 00:48:12,110

that we weren't

anticipating. So how

 

923

00:48:12,110 --> 00:48:14,850

do we put the right

controls in there to

 

924

00:48:14,850 --> 00:48:17,550

drive the outcomes

that we're looking for?

 

925

00:48:18,230 --> 00:48:21,110

We've talked about this

notion of resistance

 

926

00:48:21,110 --> 00:48:25,710

to change and how we

can overcome some of

 

927

00:48:25,710 --> 00:48:29,330

that, how we can help

our teams to lean

 

928

00:48:29,330 --> 00:48:33,990

into AI to make it

less scary and really

 

929

00:48:33,990 --> 00:48:37,830

identify ways that our

teams can engage in a

 

930

00:48:37,830 --> 00:48:42,930

way that opens up that

creativity and thinking

 

931

00:48:42,930 --> 00:48:45,290

about how do we really

use these things

 

932

00:48:45,290 --> 00:48:47,390

to transform the way

we work. It's not

 

933

00:48:47,390 --> 00:48:50,510

replacing, it's transforming

the way we work.

 

934

00:48:51,050 --> 00:48:54,010

And then in

terms of risk,

 

935

00:48:54,010 --> 00:48:57,130

we just talked about

some of the things

 

936

00:48:57,130 --> 00:48:59,190

that from a governance

perspective,

 

937

00:48:59,190 --> 00:49:03,390

some of the ways that

we can really open up

 

938

00:49:03,390 --> 00:49:06,590

deeper understanding

of the AI that we're

 

939

00:49:06,590 --> 00:49:09,930

looking to put into

place, really scrutinizing

 

940

00:49:09,930 --> 00:49:13,790

do we have the right

controls in place do

 

941

00:49:13,790 --> 00:49:16,430

we have the right

awareness of how this

 

942

00:49:16,430 --> 00:49:19,570

is going to integrate

into our workflows

 

943

00:49:19,840 --> 00:49:22,290

what are the change

management steps that

 

944

00:49:22,290 --> 00:49:25,550

need to occur as we

deploy some of this AI

 

945

00:49:26,070 --> 00:49:28,470

and then i think

most importantly

 

946

00:49:28,470 --> 00:49:32,190

talking about the

ethical use of AI. This

 

947

00:49:32,190 --> 00:49:36,090

is something that

again, we are

 

948

00:49:36,090 --> 00:49:40,050

all evolving in terms

of understanding

 

949

00:49:40,050 --> 00:49:42,050

all of the

different use cases.

 

950

00:49:42,550 --> 00:49:45,330

There's been a lot

of focus on revenue

 

951

00:49:45,330 --> 00:49:47,610

cycle because this

is one of the areas

 

952

00:49:47,610 --> 00:49:50,210

that I think has a

lot of early promise

 

953

00:49:50,210 --> 00:49:53,930

for how we leverage

these technologies

 

954

00:49:53,930 --> 00:49:57,430

to really transform

the way we work and

 

955

00:49:57,430 --> 00:49:59,530

making sure that

we're doing it, again,

 

956

00:49:59,530 --> 00:50:02,630

in a very ethical

and responsible way.

 

957

00:50:03,270 --> 00:50:05,950

So I appreciate

everyone's time today.

 

958

00:50:05,950 --> 00:50:08,310

Thank you, Lorri, for

partnering with me.

 

959

00:50:08,310 --> 00:50:11,190

And I will turn it

back over to Marie.

 

960

00:50:12,170 --> 00:50:14,290

Awesome. Thank you,

Katie. Thank you both

 

961

00:50:14,290 --> 00:50:16,790

for an excellent

presentation. And just as a

 

962

00:50:16,790 --> 00:50:18,850

reminder to our audience,

you can still submit

 

963

00:50:18,850 --> 00:50:20,830

questions through

the Q&A widget at the

 

964

00:50:20,830 --> 00:50:23,510

bottom of your console

player. It already

 

965

00:50:23,580 --> 00:50:26,450

may be open and appear

on the left side of

 

966

00:50:26,450 --> 00:50:28,870

your screen. And just

note that all questions

 

967

00:50:28,870 --> 00:50:30,630

are going to be

collected and shared with

 

968

00:50:30,630 --> 00:50:33,170

our presenters to follow

up after the event.

 

969

00:50:33,350 --> 00:50:35,570

So that's all the time

we have for today.

 

970

00:50:35,570 --> 00:50:37,850

I want to thank Lorri

Sides and Katie LeBlanc

 

971

00:50:37,850 --> 00:50:39,930

once again for an

excellent presentation

 

972

00:50:39,930 --> 00:50:42,110

and to our sponsor,

Optum, for making

 

973

00:50:42,110 --> 00:50:44,690

the whole program

possible. And finally,

 

974

00:50:44,690 --> 00:50:46,570

thank you to our audience

for participating

 

975

00:50:46,570 --> 00:50:48,630

today. We hope you'll

join us in the future

 

976

00:50:48,630 --> 00:50:50,630

for another HealthLeaders webinar. And

 

977

00:50:50,630 --> 00:50:52,630

that concludes today's

program. Thanks.

Related healthcare insights

View all

Article

Optum wins 2023 enabling technology leadership award

Optum won the 2023 North American Enabling Technology Leadership Award. 

Article

Developing evidence-based tools for accessible and quality care

Explore how evidence-based tools like The ASAM Criteria® Navigator and InterQual® Substance Use Disorders Criteria are helping payers and providers efficiently and effectively navigate the complex substance use disorder landscape today.

Article

Navigating the challenges of outpatient evaluation for patients with acute coronary syndrome

Learn how the InterQual level of care team tackled a content development challenge following updated societal guidelines for acute coronary syndrome.