After a week of AI conferences, vendor demos, and big promises, Chris Grimes and Reuven Gorsht are asking the questions lenders actually need answered.

In this episode of The Signal Podcast, Chris and Reuven compare notes from the HousingWire AI Mortgage Summit and the TMC event, where AI tools were everywhere — but clear answers were harder to find. Across demos and conversations, three questions kept coming up: which model is powering the tool, how can compliance explain it to an examiner, and can a lender train it on their own data?

The conversation moves from borrower trust and chatbot adoption to model risk, vendor due diligence, and the uncomfortable reality that traditional software assessment may not work when an AI vendor can change models overnight.

In this episode

• Why borrower trust in AI may be falling even as AI investment rises
• What lenders should ask before signing an AI vendor contract
• Why point-in-time vendor questionnaires are breaking down
• How compliance teams should think about explainability and model changes
• Where AI works best when scoped narrowly
• Why convenience often beats cost in consumer adoption
• What early adopters risk — and what they may gain

If you went back 1 year the trust in AI for borrowers was around

2

30%.

3

Reuven: Today it's 16%

4

Every conversation you're in

these days, whether you like

5

it or not, is being recorded.

6

Hello, hello.

7

Hey, Chris, how's it going?

8

Chris: Awesome.

9

How about you?

10

Reuven: Oh, good, good.

11

I'm, still, still recovering from

last week's travel, catching up

12

on emails, and it's, it's nice

to be in fair weather again

13

Chris: I think I had a, barbecue

food coma all weekend, so

14

Reuven: Oh.

15

Chris: just from being down in Dallas

and, we think we, we ended up eating

16

three, three different barbecue meals and

17

Reuven: Yeah.

18

Yeah, no, if, yeah, so, so Chris and I

was, we're, we're down in, Dallas for

19

a couple conferences, and we'll fill

you in on all the details shortly.

20

But yeah, we, we decided to

explore some barbecue spots.

21

And of course, you know, every, every meal

is the, is the gift that keeps on giving.

22

Delicious.

23

Delicious.

24

but I, I don't, I, I think I gotta go

on a barbecue detox here, Chris, right?

25

Chris: Yeah.

26

But it was interesting.

27

We had the normal barbecue, then we had

some Asian fusion barbecue, and then

28

whatever they served us at lunch, so

29

Reuven: Yeah.

30

Yeah, for sure.

31

And, the one thing, the one thing

that, this was my first time in a

32

Waymo and, and then, and then we

took that Tesla Cybercab, so they're,

33

they're all over, they're all over

the place, which is just fascinating,

34

like watching the car, you know, be

intuitive, cut people off, all that stuff.

35

Like, wow.

36

Chris: Yeah

37

Reuven: something else

38

Chris: I don't know.

39

Like I said to you in the car that day,

I still feel safer in these things than

40

I do in a regular taxi or an Uber today.

41

It's just, it seems to

see so much of the road.

42

And anyways,

43

Reuven: Yeah.

44

Yeah.

45

Same

46

Chris: till it's in our backyard

47

Reuven: Yeah, no, for sure.

48

Super, super convenient, and if

you haven't taken a, a Waymo or,

49

those Tesla Cybercabs, so they've

got, you know, they've got a whole

50

bunch of them in Dallas, these

gold Cybercabs, the doors go up.

51

they look like something from the future,

that someone I guess cobbled together.

52

But, fascinating nonetheless, because,

you know, here we are talking about

53

trusting AI, and I think we've got

some, some interesting topics today as,

54

as we look at the landscape and some

of the takeaways from the conference.

55

And on the other side, you know, you

step out of the conference room, and you,

56

you get into one of those, cabs that's,

like, fully self-driving and aware.

57

And I think I remember, Chris, we

got, we got stuck for about, like,

58

three minutes and- behind a car that

somebody decided to park in front of us.

59

And then, you know, this gentleman

comes on the, the voice and

60

says, "Hey, I see there's a car.

61

Let me clear that out,"

and then gave us a credit.

62

And it was, it, it, it was pretty

crazy to experience that level of

63

trust in transportation in your life

to an extent versus, you know, where

64

a lot of us are, are still struggling

with that fear, of exploring AI or

65

using it for anything beyond, you

know, a, a Google or writing emails.

66

So, so that being said, just, one,

one maybe pet peeve that, that

67

started to drive me crazy, but

I'm starting to see a pattern.

68

Chris, I know you and I both

use, these, these recorders,

69

like these meeting recorders.

70

So we're, I believe you and

I are both in, on Granola.

71

but the pattern I'm starting to

notice is, like, a lot of the

72

other providers that I'm using are

now stepping into that territory.

73

So one is, Whisperflow, which,

I use, for, for voice dictation.

74

So I don't type anymore.

75

I talk to my computer and

talk to my phone, and it's,

76

and it's pretty, pretty good.

77

So Whisperflow is now

in that AI detector…

78

or sorry, the, the, AI

transcriber business.

79

and then the other day,

I was on a Zoom call.

80

So, so you got li- literally,

you know, three things pop up.

81

It's like Zoom wants to record this,

and Whisperflow wants to record,

82

and of course, Granola wants in,

and Google's got its own variation.

83

So it seems like, you know, a lot of

these companies that have started as

84

kind of a one-trick pony like Granola,

which is, they're, they're good at

85

what they do, or Whisperflow, they're

now stepping on each other's toes

86

and, and converging into other, or

going into oth- other territories.

87

And then you got the traditional

players like a Zoom or a Gemini.

88

Like, call recorders are everywhere.

89

So, Chris, what I, what I'd

love to hear are your thoughts.

90

Like, is this, is this a convergence?

91

Is this, is this just, you know,

are they trying to get market…

92

get more market share?

93

Are they…

94

Is everyone trying to collect more

data, which is kind of the more,

95

Chris: Yeah, I

96

Reuven: sinister view of this, right?

97

Chris: I think that's where my mi- my

mind goes to I think two points, right?

98

and I think, I'll even go back to when

you and I first started talking about

99

doing some of these AI labs, and you

were doing some great presentations

100

at, and showing off some of the maybe

coolest AI products that were out there.

101

And, what-- and every time you went

on stage, it was a variation of

102

something else that you had maybe

presented, the quarter before.

103

I think what that told us was that,

there were a million writing tools.

104

Today, nobody uses any

of those writing tools.

105

Today, you're using-- if you want

something to write, you're gonna probably

106

use ChatGPT, you're gonna use Gemini,

you're gonna use Claude, and it's

107

gonna write for you if you want it to.

108

where go back twenty-four months and

you would r-- You know, you could do

109

it, but it wasn't as accurate, so you'd

use one of these different writing AI

110

tools that were popping up everywhere.

111

And I use that as an example because

I think what we're seeing now is these

112

other products that you've talked about,

these transcription apps that are really

113

trying to get stickier, find more ways

for you to operate within their ecosystem.

114

And then, of course, that leads to

an immense amount of data and, it's

115

no different than w- thinking about

these, these self-driving cars.

116

y- y-- we're sitting in the Waymo.

117

it's ch-- deciding what

type of music we want.

118

We change it.

119

It's now getting more…

120

it now understands that I like this type

of music over that type of music on top

121

of, where I like to go for dinner, and

it's capturing all this type of data.

122

and so I think there's, I think

that's the second part of it because

123

one-- the more and more data you can

accumulate both on, on how people

124

operate within our businesses, and you

think about these, these transcription

125

apps, it's really what they're doing.

126

They're taking a huge amount of

data on-- people aren't recording

127

their personal lives yet.

128

although I think I ha-- I

shared that story with you too.

129

somebody was telling me, I have

a device called Plaud, really

130

convenient for in-person meetings.

131

And, he was telling this story where

he accidentally left it on at home.

132

Reuven: Oh.

133

Chris: you can imagine

where that story goes.

134

Reuven: Yeah

135

Chris: he often doesn't win the

argument at home, we'll say.

136

But, but because he had this r-recording,

he decided to, to pull it up maybe two

137

weeks later and remind his significant

other that, no, this is what she said."

138

that didn't end so well.

139

Anyways, my-- a little off topic,

but my, my, my key point is that,

140

they're capturing a lot of this

data specifically around how we

141

work and what we're working on.

142

And, it's in their advantage and then

figure out, okay, what other services and

143

what other products can I provide for you?

144

so I think those are the two big

reasons why they're doing it.

145

I think us as business leaders and

people using these tools have to be

146

cautious about where and how many of

these things we want our data to go.

147

and then that also leads to the,

governance compliance side of that,

148

where, you know, lot of the tools,

especially if you're paying for them,

149

do give you options to turn off training

and things, and s- things like that,

150

which is always worth looking at

151

Reuven: Yeah.

152

Yeah, no, I, I think, there's

something to be said again about data

153

collection because a- as we talked

about previously, Chris, like part of

154

the kind of this exercise of creating

a second organizational brain is

155

capturing the conversations, right?

156

It's, it's not just, we're

not on keyboard all the time.

157

So th- those things you can capture.

158

There's a lot of great ways to capture,

how one interacts with a system or

159

different software or your, you know,

your computer or your phone in general.

160

But a lot of stuff happens, you know, in

meeting rooms, in hallway conversations,

161

in, at, at conferences, right?

162

so, so nonetheless, I mean, you bring

up a great point is that, you know, one

163

is, is the de facto is assume that every

conversation you're in these days, whether

164

you like it or not, is being recorded.

165

So, and, and that's not to say, you

know, don't say anything bad or put

166

a, a, a filter on it, but, but just

keep, keep that in mind, right?

167

Like, this whole notion of privacy is like

throw that out the window because whether,

168

you know, someone has a, you know, one

of these Plaud devices in their pocket,

169

or I've seen, you know, someone with a

little lapel microphone walking around,

170

or even if it's your phone, it's all being

recorded, transcribed, and, and it ends

171

up somewhere, whether it's training an

AI model or whether it's in a repository.

172

It doesn't really matter because

it, it's now it's conversation

173

turned to data, right?

174

the other, you know, the other side,

I couldn't agree more is like all of

175

these companies, and, and we've seen

that time and time again- are trying

176

desperately to stay relevant because

they know in, in one fell swoop,

177

either, you know, Anthropic or Gemini or

whatever becomes your de facto recorder.

178

It's, it's free, which is, like, you

can't compete with free, and that's

179

what, you know, that's what I'm seeing,

you know, with the Geminis and Zoom.

180

Like, they're only getting better.

181

Even your phone system, even your

computer natively will have all these

182

transcriptions and insights and all that.

183

It's, it's cheap to do.

184

but I think these companies, you know,

everyone's fighting for survival.

185

Everyone's fighting for, for valuation.

186

So, the, the defensibility is, you

know, expand and, like you said, become

187

sticky enough to, to the point that,

like, you, you'll pay that $16.99

188

a month as opposed to sticking with,

maybe something that you can get for free

189

'cause, 'cause it's a specialized tool.

190

But, speaking of, you know, speaking

of, the, the noise out there, so, you

191

know, so you and I went to a couple

conferences, both, tech or AI related,

192

in the, banking or mortgage industry.

193

So the first one was, HousingWire.

194

H- HousingWire had their mor- AI mor-

it was called the AI Mortgage Summit.

195

and, it was pretty, pretty fascinating.

196

Really, you know, HousingWire

did a really great job, you know,

197

putting, putting together this event.

198

And, I think, Chris, you and I had quite

a few conversations with folks, and,

199

and the feedback was, you know, some

people wanted more in terms of tactical.

200

so there's a lot of, you know,

we saw a lot of really great

201

presentations from, you know, leaders

in large, small organizations share

202

what they've been doing with AI.

203

But, I feel, you know, maybe with the

exception of a couple folks, like, there

204

wasn't more of a tactical playbook.

205

Like, what, what do, what do I

need to do on Monday type of thing.

206

any, any thoughts on, you know, maybe

if we can kind of wrap up on, on

207

the, on the HousingWire conference.

208

What, what are your impressions, and,

and what does that tell you sort of about

209

the state of the market, both in lending

and the, uh, the AI intersection there?

210

Chris: Yeah, I agree with you.

211

I, when you're out in the hallway a-and

you're speaking to not the vendors, but

212

the actual, lenders that are trying to

get value out of it seems to be the same.

213

and I even go back to,

Digital Mortgage last year.

214

So I attended that.

215

I typically go to that one every three

years just to see what's happening.

216

and it was the first time where you

really saw AI vendors showing up on,

217

on s- on the stage and demonstrating

sort of the capability of what,

218

of what I think we saw this week.

219

there's a lot of similarities of

cou- course across what vendors are

220

pushing out right now, which I think's

probably a symptom of the fact that,

221

it's a mortgage processing, like most

regulated industries have a lot of

222

paper, have a lot of hum- human work, a

lot of it repetitive, and it, shows up

223

again throughout that, that workflow.

224

and I think seeing what we saw

was that, there was this- This,

225

approach from vendors to say,

"Look, we can solve your problem."

226

but nobody really wanted to say

how they solved the problem.

227

Nobody really wanted to get into the

weeds and I don't think we saw anything

228

this week that, that really may- maybe, I

don't know if move the needle is the right

229

word, but I almost feel like nothing out

there kinda shocked me and it's okay, I

230

can see a lender jumping on this tomorrow.

231

Reuven: Yeah

232

Chris: it seemed like a lot of

really interesting, technology.

233

and I think the other thing that,

that, that did stand out to me from

234

one or two of the vendors that were on

stage shared some of the stats, right?

235

Like it's, you…

236

We're seeing more and more of

these vendors pop up single day,

237

and we certainly saw it this week.

238

But the, I think it was, was it CoreLogic?

239

I think it's called Cotality now, right?

240

and they had a stat, I think it was

something like if you went back 1 year,

241

the trust in AI for borrowers was around

242

30%.

243

Reuven: Today it's 16

244

Chris: So you have more and more of these

AI vendors popping up, and the people

245

that it affects the most, the person

buying the home, is the one that, is

246

becoming fearful of it and really not

interested in, in, in participating in it.

247

So it kinda begs the question, if all

this money is pop- is pushing into

248

these types of AI platforms, and we

certainly saw a lot of AI platforms

249

geared towards borrower engagement,

where does that actually leave us?

250

And does it actually change anything?

251

Reuven: Yeah, yeah.

252

No, it's, it's…

253

I had a conversation, with, Dave Savage

who's, runs a, a really interesting

254

business in the US for, coaching,

you know, coaching loan officers.

255

And part of what he's, offering is, is an

AI chatbot, that is consumer engagement.

256

So it's actually a

pretty fascinating, tool.

257

And, Dave shared some, some very

astute findings, that actually

258

don't necessarily support that stat.

259

And, and, you know, where, where I

get a little bit, a little bit offside

260

here, Chris, is that we know that as

consumers, so I'm just putting myself

261

in the consumer, you know, in, in a

consumer's shoes, and I'm shopping

262

around for a mortgage, the, the

intuition tells me, and this is from

263

my own personal, use cases, is that

I, I would actually trust AI first.

264

Meaning, like, if I wanna know product,

I wanna do research, I wanna look for,

265

you know, who's got the best product,

who's got the best rates, all that stuff.

266

And we know for a fact, like, it's

happening with, aspiring homeowners.

267

I, I'm in the process of helping my

parents downsize, their place, and

268

yesterday we got a contract from the

realtor, and rather than reading this,

269

you know, 20-page masterpiece, I, I

literally popped it into Claude and,

270

and used that as, you know, not, not

necessarily, you know, not necessarily

271

reviewing the entire contract, but taking

a first pass for anything that might

272

not be standard and, and all that stuff.

273

So, the conversation that, that Dave

and I had co- coming back to that was

274

more around, you know, what happens

with adoption and, and what, I think

275

what Dave saw, and don't quote me on the

numbers, I've gotta, check back with him.

276

But the, the gap, the…

277

So top LOs had about, like, 30% adoption,

and this is again just a chatbot.

278

So instead of saying, "Hey, speak,

you know, speak to Marco, your loan

279

officer," speak to the chatbot, and the

chatbot should be able to, to answer,

280

like, 70, 80% of your questions, right?

281

And they help along the journey.

282

and then the, and, and then on

the lower tier, I think, you know,

283

it, it was, it was pretty low.

284

It was, like, 2% adoption.

285

And, and I think what, what Dave found

as part of his research there was

286

that, it was all about positioning.

287

And, the, the ones that, the loan

officers that didn't get, a lot of

288

traction with the tool, they got low

adoption, just didn't introduce the

289

tool, didn't set the expectations, didn't

mention that, "Hey, you know, I'm, I'm

290

always around, but, you know, if you

have a question at midnight, just chat

291

with Sally, the chatbot here," right?

292

and, and of course on, on the

other side of the spectrum, the

293

loan officers that did make an

intro, saw tremendous adoption.

294

So, so maybe, you know, part of

me is thinking- It's, it's that

295

human layer on top of an AI.

296

It's not just like chat with a chatbot.

297

It's, it's, "Hey, this is,

this is my knowledge, my, my

298

experience, my reputation that,

you know, sits in this interface

299

that's now available to you 24/7.

300

Of course, I'm always here,

I'm always behind the scenes."

301

But, you know, don't…

302

I- it comes down to don't outsource

your, your core value proposition, which

303

is, you know, to a loan officer it's,

it's talking with customers, right?

304

Like, that is your, that

is your bread and butter.

305

So if you look at an AI solution

like we saw on, on stage, Chris,

306

and say, "Hey, you know what?

307

I'm, I'm just gonna outsource

the conversation to an AI bot."

308

Yeah, you might get some scale, but you,

you're gonna get a lot more effectiveness,

309

and, and again, going by the data that,

that Dave shared, you're gonna get a

310

lot, a lot more effectiveness if you

introduce it, if you're still there,

311

if you're behind the scenes as opposed

to just saying, "Okay, my, my clients

312

will be handled by a bot," right?

313

and I've…

314

You know, I think you and I have

experienced, like, a lot of companies that

315

are, you know, using bots the right way.

316

and then there's a lot of companies

that are using bots in a way that you

317

just wanna pull your hair out, right?

318

I, I had a, a, a business I was inquiring,

with, yesterday, and I actually, you

319

know, decided to give them a call.

320

Oddly enough, their, their online form

didn't work, so I gave them a call.

321

I get the CI chatbot,

of course, here we go.

322

but it turned out to be phenomenal.

323

Like, it, it, I was looking for some

information, found the information.

324

it realized that, you know, at the end

of the day it didn't have everything.

325

but what really blew me away, Chris,

was the, was even the follow-up.

326

So as soon as I, you know, as

soon as I got off the phone, I

327

got a text, from, from this bot.

328

An email came in as well.

329

Then the company called me, a real human

called me about three minutes later,

330

which was, I've never seen that before.

331

Like, it's, it's, you

know, they're, they're…

332

We talk about, you know, speed to lead,

whether, you know, you're in software,

333

you're in the mortgage business, or

you're, you're in any sales, like speed

334

to lead is, is, has been critical,

but this is a whole new level.

335

So getting, you know, getting someone to

answer the phone, you know, 100% of the

336

time, getting the, the text follow-up,

getting an email, and of course, you

337

know, knowing when to bring in a human and

having the human there to literally dial

338

you up and say, "Hey, I, I just, you know,

I understand you spoke to Jennifer, our AI

339

assistant, and you had a couple questions.

340

By the way, I've got the answers for you."

341

So I'm, I'm, you know, I'm personally

like I'm sold on that company.

342

I love the response.

343

Just curious, to, to hear maybe some

of your experiences with some of

344

these, either voice chatbots or others.

345

I know we've had some, some good ones.

346

We've had some awful ones as well.

347

Chris: Yeah, for sure.

348

It's, I've had a few interesting

experiences recently.

349

and you're starting to see, I think,

that more and more with some of

350

the call center, apps actually.

351

one of, one of our customers was sharing

a story recently with me that they've

352

recently put this into their front line.

353

they rolled it out the, the, response

rate was way more than they expected.

354

So they really were conservative.

355

we deal with banks and appropriately

and they had rolled it out to, o-

356

only allow it to answer about 20% of

the, the typical questions that they

357

would ask and, nothing that would be

client-sensitive or anything like that.

358

Just the generic type questions

that someone might have.

359

could be about how do I open a

bank account, things like this.

360

And, and what they found

was that there were…

361

How people reacted and how their

customers reacted was phenomenal.

362

That, they loved the experience.

363

There was no hold time.

364

It's instant answers.

365

it, it's trained again, like you said,

it's within the context of what they're

366

supposed to be talking ab- or their own,

their own, customer service and training

367

data that, that they shared, they trained

it on so that when the customer engages

368

with it, it's, it's as close to that

customer or that bank as it could be.

369

and then for more complex stuff,

it actually live transfers.

370

Reuven: Mm-hmm.

371

Chris: it's the same experience we

all deal with today is, we get on the

372

front line agent doesn't have a call

or doesn't have the answer, then they

373

transfer you to the next, the second

line person, and then eventually maybe a

374

manager, and you go through this really

frustrating sequence of, of events to

375

get to the answer you need, where, you

know what I think they're experiencing

376

at least is that 50% of those questions

now are being answered at, by the AI

377

bot and, and customers are loving it.

378

I think it comes back to your original

point here which was around, you

379

know, the stat we heard from Cotality,

which was, basically this consumer

380

confidence, at least borrowers are

becoming less confident, and I think

381

you laid out some good points here.

382

and maybe where this really lies

then is, it's the fear factor

383

that's kinda surrounding it.

384

I think when people actually engage with

it, they're seeing it's really positive.

385

and maybe that's the, that's the, the

part that's missing from both those

386

studies that, you know, you mentioned

the conversation with Dave Savage and

387

where we hear these stats on stage.

388

and that, until people are really

engulfed in it, then it, you don't

389

really understand what's capable.

390

Because, we've talked about this in

other episodes, the reality is most

391

people are still using or Gemini

to write an email for you, right?

392

and to, to experience it a

different way, it's an unknown.

393

And we're all the same.

394

We're all creatures of habit and,

the, the fear of the unknown is always

395

the thing that holds people back.

396

So until you've been forced into it, and

I think that's what we're seeing with

397

these, whether it's the loan officer

saying, "Hey, I'm not here on, on

398

Sundays and Sa- Saturdays and Sundays

anymore, but, Bob the, the avatar is.

399

talk

400

to him."

401

and, you're sitting at the open

house and you just wanted to ask

402

the question, so you give in and you

talk to Bob the avatar, and you're

403

amazed that you can get 80% of the

answers, that you were hoping for.

404

And, and so I think that's probably

part of the reason we're s- we're

405

seeing these, the shift and the changes

406

Reuven: Yeah, I think, I think, Chris,

a couple, couple things, and again,

407

I don't, don't necessarily disagree

with the data, but, but I do think

408

there's an adoption curve, right?

409

So, every e- every time, you

know, th- there's, there's

410

couple, couple points here, right?

411

One is Doing bad AI, right?

412

So, so there's a lot of companies

that just get pressure to do AI.

413

Like, they, they just, you know, they

hear it in the industry, "We gotta do

414

something," and then they do a bad job.

415

And, and, and that's, you know,

ultimately the results, what,

416

what results is a bad experience.

417

You talk to a bot that doesn't even

understand what you're asking, and

418

it just keeps repeating and repeating

and repeating, and then you just get

419

frustrated, and you start yelling,

"Human, get me to a human," right?

420

and then there's, you know, the

other, you know, the other side

421

that are doing it really well and,

and they're deploying good tools.

422

And like you said, they're

being very selective on, the

423

automation versus augmentation.

424

So yeah, if I can carve out, you know,

15, 20% of my frontline, you know, 20

425

most commonly asked questions and never

have to take a call again, it's, it's a

426

benefit to my business for sure, right?

427

'Cause I…

428

That means, you know, less, less staffing,

and, and more efficiencies on the,

429

on the frontline, which is critical.

430

That's our line to the customer.

431

but on, on, on the flip side as

well, it's a much better customer

432

experience 'cause they get…

433

You know, nobody wants to

spend 30 minutes on hold.

434

now on the flip side of, you

know, this, this adoption

435

curve, we, we've seen it before.

436

I mean, we've seen it with IVRs.

437

I, I remember, this was, you know,

probably 20 years ago, I was at a

438

conference in San Francisco, and it

was, it was in a remote part of town.

439

It was some, you know,

industrial building.

440

And, the conference finished around 5:00

or 5:30-ish, and you've got 300 people

441

walking out to an area that is remote.

442

There's no public

transportation whatsoever.

443

And I still remember I was

with couple, couple of guys

444

that I met at the conference.

445

One of them pulls out, you know,

pulls out their phone and says,

446

"Oh, don't worry, I got you.

447

We'll Uber it."

448

And I'm like, "What?

449

What does that mean?"

450

and, you know, presses a button.

451

A, a black car shows

up, like one of those…

452

That, that was the early days of Uber.

453

The limo, a limo shows up.

454

I'm, I'm in the limo with,

you know, three other guys.

455

And says, "Well, where, where do

you want, where do you want, where

456

do you wanna be dropped off?"

457

And I tell him, you know, the hotel,

and he doesn't tell the driver.

458

He puts it in his phone.

459

I'm like, "What the, what

the heck's going on here?"

460

but once I got over that experience

and I got, you know, dropped off at

461

the hotel and realized that, you know,

I didn't have to, you know, pull out

462

my credit card or cash and all that

stuff, and it was, it was really slick.

463

That was, like, my first exposure, uh,

to rideshare altogether, a while back.

464

now I don't see it any other way, right?

465

Like, it, it just…

466

You know, taking a cab is, is, is passe.

467

Like, we don't flag a cab anymore.

468

same thing with, our earlier experience

of, like, you know, taking a Waymo.

469

It, it's so much more convenient,

than, than, than taking your

470

standard, Standard Uber or rideshare.

471

So, so all these evolutions will

happen, I mean, ti- at least in my view.

472

You know, once, once somebody experiences

a, a really, you know, a really decent AI

473

experience, at the end of the day what we

gotta remember, even, you know, ourselves

474

as consumers, we're motivated by outcome.

475

Like, we, we don't really care how it gets

done, and, and a lot of, a lot of people

476

I find still, still misconstrue that.

477

There, there's a lot of pride and joy,

you know, even when we talk about,

478

you know, the mortgage process of,

oh, we're manufacturing something.

479

Well, a- as a consumer, I don't really

care what you're manufacturing or how

480

you're manufacturing as long as you're,

as long as you're meeting my need.

481

It's the same story as like, you know,

when we went to, to get barbecue.

482

We're not nosing around in the kitchen.

483

We just wanna get our meal.

484

We're hungry.

485

There's a job to be done, right?

486

Chris: Yeah

487

Reuven: really no different

in any other outcome.

488

If I wanna, if I wanna call a loan

officer, or I wanna call, you know,

489

the government, or I wanna call

my bank, I just want an outcome.

490

And, and I think I mentioned to you,

Chris, I use, uh, Wealthsimple as a, as

491

a good example of really figuring it out.

492

I, I had a, I had a, an odd question

about, updating a beneficiary

493

on a, on a certain account,

and we all know how it rolls.

494

Like, you know, you have to, you

have to call the bank, probably

495

get through five departments.

496

Then they'll send you some paperwork.

497

Then you have to fax it back.

498

and then probably have to make

another couple calls just to make

499

sure that that change went through.

500

Well, with, with Wealthsimple, they-

they're using a chatbot on their website,

501

and I literally went in and said, "I,

I wanna … How do I change beneficiary

502

on, on, on this type of account?"

503

And I, I kid you not, within 10

seconds it sent me a prefilled DocuSign

504

already filled in the account number,

filled in all, all the information

505

that, that I mentioned earlier.

506

And, and, and within probably

90 seconds I was done.

507

And did I care that I

didn't chat with a human?

508

Not at all.

509

In fact, you know, probably saved

me a couple hours right there.

510

Chris: Yeah, I think you're,

you raised a really good point.

511

I think it's something I think we've

seen over the last even decade, is this

512

shift in, in how people buy, right?

513

you always need to have a problem, right?

514

You needed a need.

515

That's why you buy something.

516

And then, where I think- where I

think the next piece always came into

517

was, you wanted to have trust and

then, ultimately the cost, right?

518

Reuven: Yeah

519

Chris: go in that order.

520

I think there's a fourth dimension or it's

ch- starting-- it's really changed, right?

521

And, I use that e- the example that it…

522

Cost isn't the-- everything being equal,

cost isn't the reason why you go one

523

way or the other anymore, which is

524

Reuven: Mm-hmm.

525

Chris: s- mind-boggling in itself.

526

But, I-- you, you talked about the,

the taxi versus Uber versus now

527

Waymo, and I remember sitting at a, a

founder dinner maybe three years ago,

528

and there was 12 CEOs in the room.

529

we were invited by a-another

vendor that we all used and, we

530

all showed up different ways.

531

I took a taxi 'cause sometimes in Toronto,

downtown, look at the, the Uber app and it

532

says some crazy price, and then you flag

the taxi down and you go f- over for $12.

533

Or you pay the $35 'cause

that's what Uber said.

534

But, there's a convenience

factor in there, right?

535

It's, I trust it, it's really easy to use,

and I don't wanna have to find a way to…

536

I don't carry cash.

537

I don't wanna have to

pull my credit card out.

538

I understand all that.

539

But for two to three X the price, for

me, I still think that's ridiculous.

540

I'll get in the taxi.

541

but there's this shift where in, in this

buying pattern in how people are thinking

542

about it today, and I think, AI's only

going to, probably exacer-exacerbate it.

543

But, you're sitting there and saying,

if it's easy, don't care about price.

544

Reuven: Yeah

545

Chris: much."

546

And, so I-- and so this is where

this, this shift is happening and,

547

I think you, you raised a few good

points there that I think is just

548

telling the story more and more.

549

And so it, this is going to be the future.

550

If it's convenient and it's easy, will

take that over having-- whe-whether

551

it costs me a dollar more or not.

552

so it's, and I think, w- e-even coming

back to, where we wanted to talk about

553

this episode and thinking about, these,

these two events, we talked a little about

554

the AI housing conference, but then we

saw it at TM-- the, the TMC event, right?

555

You had these, I think we

saw 15 demos through the, the

556

Reuven: Yeah

557

Chris: two days.

558

Again, a lot of similarities with

what these vendors were offering.

559

But, you made this comment earlier

in this episode now, and you said,

560

look, with the, in talking about

the tr-transcription apps, right?

561

how soon is it before Claw-- Anthropic

or Claw or, Gemini, whatever it is,

562

just Makes this part of their overall

package and, we don't use these anymore

563

because all that-- We're using these

transcription apps typically to feed

564

our AI agents to do work for us anyways.

565

So why, what's stopping Anthropic

from just turning on the same feature

566

and letting all the data sit there?

567

A-and then, we just run it directly.

568

it's probably not much, but we'll see.

569

But I think the same thing we saw

on stage is that there were a lot

570

of vendors that were popping up that

were selling a convenience story.

571

And, that it…

572

the idea that, look, it makes it easier

for you to do your work this tool.

573

and, you could probably execute the same

thing today using some of these AI models.

574

I don't think I'm talking out of turn.

575

I think there were several things

on that stage that could've

576

been replicated that way.

577

But you need to go and hire the teams.

578

you need to understand

how to use these models.

579

You need to find free

time in, in your workday.

580

And, as we see in our AI lending

labs when we run them, few lending

581

executives have time to actually-- go

and learn how to operate these things.

582

So if they don't have time, their teams

don't have time, and the outcome of

583

it is they're not actually getting the

benefit from it, so they'll just go

584

and pay for an out-of-the-box solution,

which there's nothing wrong with.

585

that's how most of the

economy works anyways.

586

And, but I just think it's a full c- full

kinda full circle, comment there that,

587

this, the outcome and the ease of that

outcome is going to trump everything.

588

And maybe this is the reason why

once you've adopted one of these

589

vendors, whether it's us using

Granola, even if Claude one day

590

turns it on, we keep paying Granola

591

Reuven: Yep.

592

Yeah.

593

No, I, I c- couldn't agree

with you more, Chris.

594

I, I think, eh, we, we

talked about it previously.

595

There's, eh, there's an infinite

amount of opportunity, right?

596

Because you've taken the cost.

597

You know, when, when you and I

started our companies, we spent

598

millions on development, right?

599

And we, you know, we had developers

in-house and all, and, and all the

600

infrastructure of engineering that comes

across, you know, that comes with that.

601

So if you wanted to enter the software

business, if you and I had an idea,

602

even five years ago, we would have

to basically, you know, put together

603

a nice pitch deck, go to the market,

raise a couple million bucks, build

604

it, hope that it sticks, right?

605

and now, I mean, at least the,

the majority of these companies

606

that we saw last week, that,

that barrier's, like, gone.

607

It's, it's mostly, like, you know, a

founder, maybe someone semi-technical,

608

that are building a prototype.

609

they're shopping around the prototype.

610

Even, even their slides

are made by Claude, right?

611

So, so the, the barrier to entry

is almost, like, nonexistent.

612

And, and of course, anyone smart

that's jumping on, the bandwagon

613

of, of learning AI and kinda going

neck-deep is now saying, "Well, I can,

614

you know, I can have something ready

by Sunday that I can probably get on

615

stage and demo and see if it sticks.

616

And if I get a couple customers,

it snowballs from there."

617

Now, I'm not trying to stop that, but

what, what I'm saying is, there's so

618

much noise that comes with that, that

it's, it's, it's literally overwhelming.

619

You got a lot of companies

essentially doing very similar things.

620

from a buyer perspective, it's super

confusing because to your point,

621

they're thinking, "Do I develop it?

622

Do I buy something?"

623

You gotta start thinking about

the longevity of these companies.

624

Like, if you're, you know, if you're in,

running a, a, any, any, any operation

625

of sig- significant size and scale,

you can't afford for these tools that,

626

you know, your frontline is using or

your back office to, to go down or

627

even worse, go out of business, right?

628

The other, you know, the other angle, um,

I think there were questions that were

629

asked of, of the audience, of the, you

know, buyers, multiple times, and, I don't

630

think we heard a clear answer, did we?

631

Like, what, what, what

are those questions, Paul?

632

Chris: no, I think you're right.

633

I think there's…

634

the questions really boil down to three.

635

I think we heard it every single time.

636

what models are you using

to run this AI tool?

637

And I didn't hear an answer.

638

Now, to be fair, I don't think I

caught every single demo, but the ones,

639

you know, the, the 12 or 13 that I

saw I didn't hear an answer to that.

640

I think the second one that came

out was, how do they explain

641

this to their compliance team?

642

when the state examiner comes in,

or in Canada, when OSFI comes in,

643

Reuven: Yeah

644

Chris: do they explain what rules

were followed, even if there's

645

some transparency around it?

646

And, we heard one, one of the,

the vendors on stage say, "Look,

647

deterministic AI is impossible."

648

And, you know, this might be a for

another conver-- for a future episode.

649

this generative AI is all that we have,

which is effectively a prediction model.

650

And, so-- and if you can't have the

de-deterministic meaning that every

651

single time, the, the guardrails are

structured in a way that the output

652

is always the same, you can't bring

that the, to the state examiner or

653

frankly, even meet Fannie's, guidelines

that they put forth on, about

654

Reuven: Mm-hmm.

655

Chris: days ago or the

beginning of August anyways.

656

and then I think the third thing

that we really heard was, what if I

657

wanted you to use my data to train?

658

and really, use my secret sauce to

get that-- to really hone in on that.

659

And, I…

660

It was frankly, blank faces.

661

It was…

662

probably tells you, we're too-- In

some of these cases, they're, they

663

were either too early, they didn't come

from industry, so they didn't have the

664

experience behind to understand it fully

and, I think sometimes we see that a lot.

665

I think we see vendors pop up in spaces

that, they're not entirely, familiar with,

666

but it's-- but when it's super complicated

and there is a real use case for AI,

667

which we know in mortgage lending, that it

seems like an easy hill to kinda climb up.

668

not trying to disrespect anyone, but I

think we, maybe we saw that a little bit

669

this week across those two conferences.

670

and then, ultimately, to your

point a minute ago, all of this

671

has to be presented to a board.

672

if you're the ops leader and you've

made the decision, "Look, I'm, I

673

need to make this easier for my team

because I'm hearing everyone else

674

is jumping on the AI bandwagon."

675

But what are you going to tell your

board that's gonna give them the

676

confidence that, look, this isn't

technically revenue generating.

677

This is gonna reduce our costs,

and we're gonna get this benefit.

678

but on top of that- there's the unknown

679

Reuven: the risk

680

Chris: of the risk because

none of-- nobody on the stage

681

can answer the questions.

682

Reuven: Yeah

683

Chris: what is the next step?

684

for sure, having seen this 12 months

ago and seeing it 12 months later

685

and with different conferences,

the mo- the needle's moved.

686

I will give everyone credit on that.

687

it's…

688

we're getting to the point where

the same day mortgage will happen.

689

I don't know if it's gonna happen

this year, maybe two year, two

690

more years we're probably there.

691

there's still a lot that's unknown

that we still need to move to be

692

able to get to that, that outcome

693

Reuven: Yeah.

694

Yeah.

695

I mean, the, the, y- y- you're,

you're absolutely right, Chris.

696

I mean, there's the, there's a cost,

and then there's always a risk.

697

And especially, you know, you and I

got a chance to, to sit in on, the,

698

the compliance, committee, which

was, which was really phenomenal.

699

Like, I mean, you know, it's, it's great

to see, any industry getting ahead of

700

understanding, you know, what challenges

are there with risk and compliance.

701

'Cause at the end of the day, we don't

know what we don't know, to your point.

702

And, and to the credit of some of

these companies, like, you know, yeah,

703

I can, I can tell you which model I'm

using today, but just the speed of,

704

how quickly things are changing here,

I can have a different model installed

705

by this afternoon, which, you know,

brought up an interesting conversation

706

in that room around compliance.

707

'Cause, you know, coming from, you

know, enterprise software and, and,

708

you as well, Chris, coming from,

you know, having, having your, your

709

platform assessed by numerous companies,

the way to think about it is it's

710

always been a point in time, right?

711

You get an RFP, you get the questionnaire,

you clear, you clear the different

712

hurdles, you show them that, you know,

you're SOC certified or you're NIST

713

or you're ISO and, and, and so on.

714

And th- those, those used to be the

barriers and, you know, it, it was

715

just a questionnaire at a point in

time and, you know, once we go forward

716

with the implementation, we kind of,

you know, we set it and forget it.

717

Of course, you get updates around, you

know, privacy and, you know, part of

718

your contract with, with the vendor.

719

but nowadays, and, and, you know, this

whole conversation started with, "Oh,

720

we gotta, we gotta do more education.

721

We gotta do more, you know, collaboration,

because AI is changing so quickly."

722

And in the back of my mind, I was

almost kind of jumping out of my

723

chair to say like, "You know what?

724

Like, this is not…

725

Like, what got us here over the last

30 years of how we select software

726

and go through the questions and

assume that we're always having a

727

conversation is not the way forward."

728

Now, I don't have a good solution.

729

I know, I know there's probably someone,

someone will solve it, but at the, the

730

pace that things are going, So yeah,

you can, I could fill the questionnaire

731

and say, you know, Chris, that you're

using the appropriate model, and I've

732

tested it and the regulator's given

the blessing and all that stuff.

733

But we've all seen what happens.

734

The new model gets released, or a

new frontier model, and every vendor

735

just being, you know, wanting to be

competitive, wanting to, to lower, their,

736

their operating costs, 'cause that's a

whole other discussion around, you know,

737

the token costs and stuff like that.

738

So you sell me something, I sign off, and

tomorrow you're installing a new model.

739

Now- In hindsight, you might not even

know, Chris, as a vendor, what that

740

model actually has or, or doesn't have.

741

We saw that with Fable 5, like it was

classic, you know, put it on the shelf.

742

No, no, no.

743

Take it back.

744

so how, how do we wanna think about that?

745

Like, so what does, you know, if, if we're

746

Or is there, is there governance

or is it, you know, is it a bunch

747

of AI agents updating each other?

748

Because we're at the point, at least

that I feel, that we're, we're beyond

749

human cognitive capacity to manage that,

let alone some of these organizations

750

are, are smaller in, in size that,

you know, they don't have the time to

751

manage every time there's a, there's

a new model being installed, right?

752

Chris: Yeah.

753

Look, I think that's, you

made some great points.

754

And, I think number one, even the

regulators aren't sure around it.

755

So if you're a ve- if you're

the, if you're the lender, how

756

do you actually that decision?

757

Because you're gonna be beholden to what

these guidelines are put down, whether

758

that be, I, I know OSFI's E-23 is coming

out and, there's some framework for

759

that, but even that's not 100% set yet.

760

I think that's gonna…

761

that will evolve.

762

And then you take, or in the US, we, we

met with, the head of, of MISMO and they

763

were talking about, the, this frame,

this idea that they basically took

764

Freddies and Fannies and, and some of

the state regulations and they're trying

765

to put a, an industry standard together

to say, "Look, as long as we adhere to

766

more or less this, we should be okay."

767

but I think you're right.

768

It's the You know, and actually I'll…

769

I think Brian said it on stage, right?

770

It's how do we allow…

771

How do lenders get comfortable by,

assessing the vendor without necessarily

772

opening up, the hood of the car?

773

And I think it was almost his

word for word verba- verbatim what

774

Reuven: Mm-hmm.

775

Chris: because,

776

Reuven: Mm-hmm

777

Chris: how I build my engine is what

makes, me valuable as a company.

778

Reuven: Yeah

779

Chris: I don't want to

expose that to everyone.

780

Yet, at the same time, if I'm on

the other side of the table and

781

I'm the lender, kinda wanna know

what's in that engine because,

782

Reuven: You have to

783

Chris: I wanna know where my data's going.

784

I wanna know where, how and

what models are being used.

785

a lot of these questions are gonna get

answered over the next little while.

786

But, I think the point that I would wanna

make, and I think this is, we've seen this

787

with every technology curve ever, right?

788

There is going to be early adopters.

789

I still think we're in that early

adopter phase where we're nowhere near

790

s- s- sitting at the top of the peak.

791

and the, the ones that adopt early are

gonna probably have some bruises, but

792

they're also gonna have tremendous wins.

793

Reuven: Yeah

794

Chris: and I think it's being

cautious about where you deploy that

795

information and where you layer it.

796

I think if I'm sitting, if I'm sitting

back and thinking about it from a

797

lender's perspective today, I…

798

The part that I'm worried about

is where my PII data is going.

799

And I think that's the question that

I'd be most concerned about because

800

that's where the real risk lies.

801

The rest of it, y- you can

recover from more or less.

802

The, because I don't think there's

any AI model out there yet that's…

803

any AI tool out there yet that's fully

running operations from front to back.

804

and the humans are still gonna

be in the loop, and I think it's

805

probably a mistake to remove the

human from the loop at this stage.

806

They, you know, giving them the

power to make that decision as long

807

as the, the, the tool that you're

buying gives you that flexibility.

808

you may eventually decide, "You know what?

809

I'm happy with the AI running more

of this operations, but in the short

810

term, I still want, I want my people

to be augmented, not replaced."

811

I think s- I think it's how you get

around some of these, the compliance

812

at, at, at risk in the short term.

813

anyways, that, that's how I think

about it and I don't know if you have

814

Reuven: Yeah.

815

Chris: kind of

816

Reuven: No,

817

that's, that's awesome, Chris.

818

I, I'm, I'm, I'm gonna have to sign off,

but, thank you for joining us this week.

819

We'll come back.

820

I mean, there's a lot more of

these topics to, to talk about.

821

All of them, you know, all of them are

meaty topics where, you know, Chris

822

and I are trying to stay on top of

everything, that's new, and that's, that

823

is proving a full-time job on its own.

824

But what we're hoping to do is bring you

more, you know, more of those practical

825

perspectives and our own experiences in

terms of what we're seeing, what we're

826

hearing, data, dispel it, argue it a

little bit, and, we'll go from there.

827

Hopefully add some value

to your own, AI journey

828

Chris: Have a good week