Ryan Nauman hosts Zephyr’s Adjusted for Risk podcast from Lake Tahoe and interviews Jacob Ayres-Thomson, founder and CEO of 3AI, about how AI is shifting wealth management from understanding AI to implementing it for investing. Jacob explains 3AI’s “Alpha Intelligence,” which targets equity outperformance through predictive insights delivered for people, quant models, and AI-powered funds/indices. He describes his background in equities trading, stochastic asset modeling, and machine learning, and why humans struggle with stock picking due to noisy markets, insufficient data for reliable learning, and benchmark effects driven by market-cap concentration. They discuss AI’s role in increasing market efficiency, removing the analysis bottleneck by condensing vast data into forecasts and explainable research, and how advisors can use robust, statistically tested AI signals and indices, including products built with S&P Global.

Learn more about Zephyr here.

Learn more about 3AI here.

00:00 Welcome to the Podcast

01:18 Meet Jacob Ayres-Thomson

02:00 What 3AI Does

05:55 Why Stock Picking Is Hard

08:24 Learning From Noisy Markets

11:28 Speculation and Market Cycles

14:28 Stocks as a Never Ending Game

15:25 How AI Changes Investing

17:38 Alpha Intelligence in Practice

20:01 AI as the Analysis Engine

21:14 Wrapping Up the AI Thesis

22:03 AI Is Not One Thing

23:39 Markets Get More Efficient

25:44 Where Alpha Still Exists

26:52 Humans Plus AI Together

29:16 Alpha Intelligence Scoring

29:59 Stock Specific Factor Weights

33:47 Causality Versus Correlation

37:30 Moneyball Investing Analogy

39:26 Advisors Using AI Tools

41:25 Due Diligence On Forecasts

42:36 Where To Learn More

44:34 Podcast Wrap Up

Connect with Ryan Nauman:

LinkedIn

X

Let's go.

2

Ryan Nauman Market Strategist Zephyr:

everyone and welcome to zephyr's

3

adjusted for Risk Podcast

from the shores of Lake Tahoe.

4

I am Ryan Nauman, the market

strategist here at Zephyr.

5

Over the past couple years,

the wealth management space

6

has been grappling with ai.

7

First, it was about

understanding what AI was or is.

8

Now it is all about how to implement it

and what are the best ways of using it.

9

Well, I have on an industry expert

who's gonna share his thoughts on

10

the impact AI is having, on investing

in what his firm is doing to make

11

investing more efficient and improve it.

12

But first, today's episode is sponsored

by the award-winning Zephyr, which

13

helps investment professionals.

14

Make more informed investment

decisions on behalf of their clients.

15

Alright, I've already talked enough.

16

Let's move move on to

the star of the show.

17

I'd like to give a very warm

welcome to Jake Ayers Thompson.

18

Jake is the founder and chief

executive officer at three ai.

19

Jake, thank you so much for coming on.

20

It's an honor to have you on.

21

I'm really excited

about this conversation.

22

When I have conversations about ai, a lot

of it is more about improving processes,

23

workflows, and as we all know, I have

a passion for the investment side.

24

So really excited about this conversation.

25

Can you please tell us a little bit

more about yourself in three ai.

26

Jacob Ayres-Thomson Founder & CEO 3AI:

Yeah.

27

Thank you very much for having me

on and taking the time to give us

28

a chance to talk about what we do.

29

So, yeah, three AI basically our

product is Alpha Intelligence.

30

We're really focused on,

you mentioned before about

31

workflows, that kind of thing.

32

We're very much focused

on alpha outperformance,

33

predictive insights on equities.

34

And, and we deliver that in three

formats, sort of three types of

35

users, people, funds and models.

36

So for people, we empower them

models, we have various data sets

37

that quants use to, to enhance.

38

Their own strategies and for funds we

were able to partner with in industry

39

to launch sh I powered funds or indices

that they can wrap a fund around.

40

So that's basically in short what we do.

41

Ryan: Yeah.

42

Fantastic.

43

I love the fact that you

guys are working on Alpha.

44

Generating Alpha said a report

some research on how hard it is for

45

financial or for asset managers to

generate alpha in this environment

46

because, you know, whether it's

concentration risk or just data is more

47

available, markets are more efficient.

48

But we'll get into that shortly.

49

But why did you start three ai?

50

What was your thinking behind starting it?

51

Jake: It's that's a 17 year journey

after about five years in the city

52

and, and, and studying stats and, and

the markets and that kind of stuff.

53

So my own career had I, I've been

in equities trading, I've, I've.

54

Been specializing in what we

call stochastic asset modeling.

55

Generating asset simulation

systems as an actuary.

56

And I'd also worked as a, as as a

quant and basically across those three.

57

And also having been been very

interested in kind of Warren Buffet,

58

Ben Graham thinking and valuation.

59

So should I say, across those

four, I saw a lot of gaps.

60

Gaps between how fundamental investors

think versus how your typical PhD,

61

who's very statistical thinks.

62

Versus what's in data and

what's in information.

63

And so my own career was

kind of sidetracked into

64

obsessive research in this.

65

And I read a couple hundred books

and, and, and maybe 500 plus papers.

66

And then I, I recognized really

that the most advanced results

67

from papers were really originating

from stuff using machine learning.

68

So I went back to university to study

machine learning when they created

69

the first master's program in London.

70

And from, and, and it

should been on since then.

71

So I, I recognize there were lots of

gaps that could be closed and ways

72

to improve elements of, of, of the

kind of alpha generation process in

73

terms of actually starting through ai.

74

At the time I was running

a data science team at a, a

75

FTSE listed UK financial firm.

76

And we spent maybe 80% of our time looking

at equity forecasting y upload GPUs to,

77

to unlock new algorithms back in 2015.

78

Before most people knew what

NVIDIA was, unless you're a gamer

79

or, or using VR at that time.

80

And we had a series of three, kind

of six months to one year live

81

forecasting experiments that produced

statistical certainty pretty much that.

82

We were generating active

Alpha in our forecast.

83

So the decision to leave and, and

to form three AI at that time was

84

a, was a statistical decision.

85

Also based on understanding

how we're able to get Alpha and

86

what's missing in the market.

87

I.

88

Ryan: it.

89

That's fantastic, Jake.

90

And I love the the part, you know, about

machine learning and data and it's, it,

91

you know, interesting captivating stuff.

92

And, you know, I often

say on the show that.

93

stocks is really hard.

94

I don't do it.

95

I leave it up to the smart people to

manage my, you know, I'll pay the, you

96

know, 50 basis points or whatever it is

to have an asset manager, manager for me.

97

It, I just.

98

Takes too much time.

99

Struggle with it.

100

Why do humans struggle

with picking stocks?

101

It's hard.

102

It's not easy.

103

A lot of information out there,

A lot of stocks situ choose from,

104

whether it's individual stocks,

ETFs, mutual funds, you name it.

105

But why do you think humans

struggle to pick stocks?

106

Jake: It is funny actually, I'm,

I'm pleased you asked that question

107

because I've analyzed that.

108

And there's, there's, there's

two insights to share there.

109

The first is, if we, in observing,

like we see with s and p publish,

110

for example, 95% of active managers

are wanting to perform the benchmarks

111

that they're trying to beat.

112

If we look into that and, and try to

figure out why the, the first question one

113

might ask oneself is if humans struggle

to pick stocks, then perhaps they're

114

struggling to learn how to pick stocks.

115

And I took the s and p 500 stock

data, did a plot, earnings yield

116

versus return for the last 20 years.

117

Imagine that's your learning,

your experience of professional.

118

You work 20 years, you've got

perfect memory of 500 stocks, which

119

I would argue most humans don't have.

120

And you look at the pattern there

and there's next to no pattern.

121

So I think most people

do what Warren Buffet.

122

Recommended they don't do, he said,

don't let price be your teacher.

123

And the industry tends to be, I

think, too short term and gravitate

124

around short term performance.

125

Where, where in, in investing even

one year is short term performance.

126

So stocks are incredibly noisy and we

can do all the right things and all the

127

right research and, and have the right

frameworks for investing, but the market

128

will will teach us that we're wrong.

129

And if we believe the market, we may then.

130

Drop a winning strategy, a

winning way of, of, of, of

131

comparing and analyzing companies.

132

So I think, and, and then if

you think then within industry.

133

It's very hard to do what

Warren Buffett did, which was

134

to wonder to perform the.com

135

boom for two consecutive years without

looking like an idiot, even though the

136

tide went out between 0 0 0 3 and the.com

137

bust.

138

And we found out who was

swimming naked, right?

139

So that, that's like a five

year cycle where you had to

140

kind of hold onto your pants.

141

Some people threw the towel in.

142

And left industry people's kind of

grandparents were doing BET stock

143

investing, buying the latest.com

144

stock.

145

We've seen a period like

that not too long ago.

146

Things like Rivian reaching 140

billion, having never sold a car

147

that's now down, I think sub 10

billion or maybe it's gone up again now

148

'cause they're finally selling cars.

149

So.

150

The, the market is kind of crazy.

151

Warren Buffet describes it as a

schizophrenic, and you just turn

152

up and see if there's opportunity

in the crazy pricing that's there.

153

But really, at three ai, we really

think about stocks as companies, and

154

you're getting slices of companies.

155

So the first part is that people

tend to learn from experience.

156

It's a very human thing.

157

It's a very animal thing, right?

158

It's how our brains work.

159

We learn from experience, but

the stock market, the amount of

160

data that one gets in a career.

161

Is not enough to be

statistically reliable.

162

So if you have a very noisy system,

you need to look at hundreds of

163

thousands, even millions of data

points to see weak patterns that exist.

164

We look at a million years

stock data, that pattern of

165

earning shield becomes clear.

166

The second point, and I think this

is also probably misunderstood by.

167

Certainly the allocators to, to funds

is that there's periods in the stock

168

market where stock pickers look bad

or stock pickers look good because

169

we tend to benchmark them against the

major indices, the major benchmarks,

170

but their market cap weighted.

171

So if you look at the last

kind of five or so years.

172

The Mag seven have had a

huge weight in the index.

173

Arguably hard holding a higher proportion

of the index than the risk management

174

framework of a diversified portfolio

would allow a typical professional

175

manager to hold well, what does that mean?

176

It means even if they're bullish on

those stocks and were right, they may

177

have even held less and effectively been

short against them versus the benchmark.

178

So one of the interesting

things is if, if you randomly.

179

Sample the s and p 500.

180

There's been periods in recent times

where 199 out of 200 portfolios

181

of equal weight, 30 stocks has

underperformed the s and p 500.

182

And you go, well, how the

hell is that happening?

183

It's simply because of the, the

weighting difference and the,

184

the very largest companies have,

have performed so exceptionally

185

well in terms of stock returns.

186

During that time period, but

you know, if history teaches us

187

anything, these things turn around.

188

If we then see a crash, we see if, if we

were to see the mag seven significantly

189

decrease in their market cap and

their weighting of the the index.

190

Then the average stock picker might

suddenly look like they're producing alpha

191

against that benchmark again, dot com.

192

Boom.

193

If you dig into the data in the.com

194

boom and crash, it was really at an index

level, but that was really prominent.

195

But if you randomly picked stocks

equally weighted from a wider US stock

196

universe, you would've looked like

you're underperforming in the boom and

197

you're not seeing as much of the crash.

198

So it looked like you were

negative alpha and positive.

199

So there's a difference

between stock selection skill.

200

And how well you perform

against a benchmark.

201

So I think there's two elements there.

202

One is learning.

203

We're not, I don't think the, we

are naturally the right brain type.

204

The way that we learn, I think,

is for a constant physics system.

205

We learn how to walk and talk.

206

No surprise neural nets.

207

Were able to learn to

do this stuff very well.

208

You see robots talking and walking

and, and all of this kind of thing.

209

And we could see that even 10 years

ago that they were very good at mammal

210

like tasks because they're designed

approximately like mammal brain

211

structures, albeit over overly simplified.

212

But so I think, I think

that's the two key reasons.

213

Ryan: No, I love that, Jake.

214

And you're right.

215

There is maybe that education

gap or it's just hard to learn

216

and I feel as if there's.

217

Always those periods of, we had it

after, well, during COVID and directly

218

after it when money was free and we

had to, you know, FOMO runs rapid

219

and meme stocks and those trends or

those periods of time, they kind of

220

throw everything that we've learned

fundamentals out the window, and all of

221

a sudden we're focusing on fundamentals.

222

Those aren't working because

everyone's shooting to the moon.

223

So what impact does that have?

224

Jake: Yeah.

225

I think that the, one of the things

to think about that if you're, if

226

you're training systems to learn

how stocks work is to, is to think

227

what's the correct way to kind of

approach how you learn from stocks.

228

Our approach at three AI really is, is,

is to learn from all stocks, from all

229

time, from all regions and all sectors.

230

And we, and we take a kind of first

principles based approach to stocks that

231

from the perspective of a shareholder,

they're colorless cash printing machines.

232

And so if you had infinite time and in,

and therefore had infinite stocks to

233

evaluate the averages would be the truth.

234

And so that's what we seek to

learn too, which effectively gives

235

us a very long-term focus system

that becomes deep on company

236

measurement, not on speculation.

237

So, so in times when you see like

correlations go to one like COVID or

238

Credit Crunch, everything goes down

in, you know, obviously in COVID it

239

was worse for airlines than, than

other things, but you get these fear

240

spikes or you sometimes also get.

241

In reverse these booms

that are speculative.

242

I would say we've had a crash

that's speculative somewhat recently

243

in, in, in, in software stocks.

244

Think people thinking AI

will eat all software.

245

It's a hypothesis, but it's getting

priced in as, as probably more

246

realistically probable than it is.

247

'cause if you're a software

engineer, maybe it means you, you

248

need less engineers, but you're

still gonna be developing your

249

software just albeit more rapid.

250

So.

251

There's a, there's so many different

weird and wonderful things that

252

happen in the stock market.

253

I do think that each company is its own

story and so that we think a bit more

254

like that and we seek to deeply measure

the financials of companies knowing that

255

the stock market will be consistently

kind of doing the wrong thing.

256

But over time it should pay

off if you do the right thing.

257

But if you think that doing the right

thing will get the right result.

258

Frequently then that's a

misunderstanding of the stock market

259

Ryan: Yeah.

260

Jake: it, because it

won't forward like that.

261

So it is a bit like Warren

Buffet says buy stock.

262

I, I I, I buy stocks where I don't mind

if the stock market shuts for 10 years.

263

'cause I want to own the

company and sit on the company.

264

Ryan: Y That's fantastic.

265

I love that you said the stock

market's weird and wonderful.

266

Can I can I

267

Jake: Yeah.

268

Ryan: post that?

269

I think that's a fantastic quote.

270

I love it because I think that's

why we love it too, right?

271

That's why every day it's something

different and why I think so many

272

people love the stock market.

273

It's weird and wonderful at the same time,

274

Jake: Yeah, it is.

275

It is.

276

It is a, it is a, it is

a very interesting game.

277

I think I'm very into game playing

all different types of games,

278

playing chess or, or card games.

279

And ideally things with

probability in like cards or dice.

280

But the stock market for me is the

one where I feel I can just keep going

281

deeper, deeper, deeper, deeper, deeper.

282

It's like an ion where you never

fully have unraveled all of it.

283

The deeper you look in,

the more there is to learn.

284

And the interesting thing about

the stock market is it connects

285

to nearly all human activities.

286

Nearly all of them will ricochet through

the stock market somewhere or other, or

287

the financial market somewhere or other.

288

So if you're an alien, you come to Earth

and you wanted to kind of see like the,

289

the, the activity and behavior of, of

homo sapiens, you might go to the stock

290

market to actually see what's really

going on at a kind of fundamental level.

291

Both in how we perceive things, but

on actually also what's happening.

292

The two, which are not the same thing.

293

So.

294

Ryan: I love it.

295

So let's talk about AI with three ai.

296

know, Zephyr was created, you

know, years and years ago,

297

technology, and it was based on Dr.

298

Bill Sharp's, returns

based style analysis.

299

At the time, it was

pretty, you know, moving.

300

It was like, wow, we can determine the.

301

The style, the behavioral, through

just tracking, you know, re

302

returns, regression, and time over.

303

A manager for a manager.

304

So some great stuff there.

305

Now we're advanced to ai.

306

What impact can AI have on investment

management space and, and picking stocks?

307

Jake: A lot.

308

I'll, I'll, I'll, I'll be honest.

309

So if, if we, if we think

about the terminals.

310

That traders, investors,

the, the financial market

311

community have access to today.

312

They're really an information layer.

313

So they're, they're kind of their

database with a ton of different

314

data sets, nice, gooey, and,

and, and very usable analytics.

315

It's been really developed over

many, multiple decades four decades

316

in case of Bloomberg since 81.

317

But they're really an information layer.

318

So whilst they, they got rid of the.

319

Data bottleneck.

320

You go back to ticket tape,

you've got data bottleneck.

321

You've gotta go and contact the

company by telephone and, and hope

322

that they send you the accounts

and look in the post every day.

323

Yay, we've got the accounts.

324

Pass it to somebody who's then gonna

write it all up on paper and you

325

haven't even got a spreadsheet, right?

326

And then you're analyzing

that, et cetera, et cetera.

327

So there was a, a data sourcing

bottleneck, just, just probably having

328

access to information was the edge.

329

Think about Warren Buffet when he started.

330

He was buying the investors Almanac

and literally looking at things like

331

price to earnings ratio and going,

whoa, this thing looks so cheap.

332

It's half book value with a 25%

yield, and profits seem stable.

333

So it is a wonderful bond at 25% yield

and or even if it got acquired or

334

exited, there's gonna, there, there's

a lot of realizable value too dollars

335

for every dollar I'm buying on it,

but she called cigar by investing.

336

But the fundamental.

337

Thing that he was doing was effectively

stock screening with his own mind,

338

looking at basic metrics, right?

339

So he was a step ahead of most,

most people don't like looking

340

at data, they find it yucky.

341

But what we're doing now at three ai,

so if we think of the people part of

342

our empowerment, we are essentially

taking all that kind of data that's

343

in, in a terminal, that financial data,

the analyst reports, the financial

344

forecast of the future, technical price,

behavior risk, et cetera, et cetera.

345

And we're using machine learning

methods to analyze the entire

346

history of the, of, of, of all

stocks and learn from all of it.

347

And then look at all of today's

data for all stocks to be able to

348

then forecast them for tomorrow or

in our case, the next 12 months.

349

So what that's offering, I think, as a

solution is, is we're going from data

350

sourcing being the bottleneck to, and but.

351

Being then solved by terminals

so that analysis then becomes

352

the bottleneck, right?

353

What we are doing is, is essentially

then taking all that data and condensing

354

it into a single number that people

care about, which is an alpha forecast.

355

And also, most recently we

have effectively quite deep

356

stock research reports.

357

So about 15 pages long

per stock, 20,000 stocks.

358

That's 300,000 pages of equity

research being generated a week.

359

Which is readable by PMs

and, and, and, and, and also.

360

Then deeper, even analyzes our own AI

to look at what to watch out for, what

361

could, what could really move the needle

or flip the view on a particular stock?

362

What's driving it, what's not driving it?

363

And we found our alpha intelligence

is basically incredibly intelligent.

364

It, it's, it's, it's unlike a human

research report, it's incredibly specific.

365

You know, think about the future.

366

An AI doctor, a human doctor,

human doctor says, Ooh, I think

367

we need to give you a scan there.

368

There's risk.

369

You go, oh my God, I've

definitely got cancer.

370

The AI doctor will say, I, I, I

371

chance that you have this and

a 4% chance of this, and a 1%

372

chance of that type of cancer.

373

And, and the rest.

374

It's a property of no cancer.

375

That's what we find with

our alpha intelligence.

376

It's completely specific and

it's able to unlock a, a very

377

deep analysis of each stock.

378

In our case, we've got over 300 factors

or models that the AI looks at on

379

every single stock across over 20,000.

380

But at the micro level, we also

have global macro sorry, macro,

381

so macroeconomics and a kind of

top down alpha as well as a bottom

382

up alpha all combined together.

383

So.

384

What then I think that unlocks for

the investment manager is it gets

385

rid of the analysis bottleneck.

386

It really turns all of that

information in a way that a human

387

would not have the time to do.

388

E even a quant would not have

the time to do traditionally.

389

So I think for a an active manager,

this is effectively a quant shield.

390

So the, the, all that alpha that was being

eaten by quants, we see the massive rise

391

in the systematic hedge fund industry.

392

There were, there were basically

the evidence-based approach to

393

data and using data science and

statistics and, and algorithms.

394

It's that on the plate.

395

Without needing to necessarily

know why, but but also having the

396

information and explainability to how

it relates to that stock and how to

397

analyze it and how to think about it.

398

That's one aspect.

399

The other two are the other two

products that we spoke about.

400

So for quants, we're able to give them

our forecast, ingest it into their

401

systems to enhance them, and finally,

for the creation of products, like

402

wanna create a new ETFA new index,

then it's very easy to use our data.

403

It's all stock mapped,

it's ranked it's forecast.

404

It's very easy to construct.

405

AI enhanced portfolios.

406

So I think that's, that, that's

the three that we're immediately

407

seeing that we're working on and,

and that we're, we're delivering.

408

Ryan: Yeah, that's fantastic.

409

And sometimes I feel as if, and you

mentioned it numerous times, or about

410

AI just in general, we think of AI in

a broader sense of just making things

411

more efficient, processes, you know,

given us some time back in our day.

412

But at the end of the day, what

AI best, and I might be probably

413

oversimplifying it here, is just taking

huge sums of data, synthesizing it,

414

and making it intelligible, correct.

415

Like that's at the end of

the day, what AI is doing.

416

So like you said, humans aren't

good at just creating, looking

417

at millions of lines of data.

418

Whereas AI can do that

and make sense of it all,

419

and that's really what you guys have done.

420

Jake: Yes.

421

Yeah, it's a bit more complic.

422

I mean, yeah, it's, what I will

say about AI is that AI is,

423

is almost as unique as people.

424

So people think AI is one

thing, but it's really not.

425

It's a, so.

426

In, in the financial system.

427

We would even today find, I

suspect, thousands of AI systems,

428

depending on how we classify ai.

429

I, I think of it as machine learning.

430

We'd find thousands of them distributed

across the world, primarily in the

431

us It's the most advanced sector

it's most advanced region, sorry.

432

But, but EE each one is, is

as unique as the designers

433

and, and we are, we're quite.

434

I think in, in, in, in really

approaching the, the Ben Graham, the

435

Warren Buffet kind of learning and

bringing that into the AI world or,

436

or, or should we say, beginning with

the hypothesis of is it learnable

437

through data, how this stuff works?

438

Ryan: Yeah,

439

Jake: so.

440

Ryan: that's So.

441

of touched on it earlier, and

it's part of, you know, the name

442

of your scoring system, alpha.

443

So if machine learning, if machines all,

whether they are, you know, each ai tool

444

is unique and the methodology is unique.

445

Doesn't it make markets

much more efficient?

446

And if it makes markets more efficient,

how do investment managers, even

447

if they're using technology, how do

they generate alpha in such a, you

448

know, that's much more efficient?

449

Do the technology and data

being more accessible?

450

Jake: Okay.

451

Okay.

452

I'm writing some notes down there.

453

There's a couple of questions in there.

454

So in terms of efficiency I

think the short answer is yes.

455

Artificial al intelligence machine

learning based approaches and, and to

456

be honest, anything that, that brings

greater forecasting accuracy, better

457

allocation of capital, plays a role.

458

In humanity's effort to

build the efficient market.

459

A long time ago there was a

hypothesis of an efficient market.

460

They called it a hypothesis

'cause it was, it didn't exist.

461

It'd be very easy if your progressor to

prove that you'd find in a liquid stock

462

call up $5,000 a stock and move its price.

463

There you go.

464

That wasn't an efficient

idea that, to prove a point.

465

You couldn't do that in an

efficient market, could you?

466

An efficient market, everything

would be perfectly priced.

467

You couldn't move prices.

468

So so yes, I definitely believe that.

469

Artificial intelligence is going to play

a role in, in enhancing the efficiency.

470

I think if you look at your home

markets in the us, we in our data see

471

them to be and from our client base

see them to be the most efficient

472

stock market in the world today.

473

And you have the largest gathering

of highly sophisticated quants,

474

armed with tons of compute

and access to tons of data.

475

And I think a result of that is that your

stock market actually goes up quicker.

476

So pension funds are

larger in the long run.

477

If you put the stock market up by

one or 2% a year, every person's

478

lifetime, it's a, it makes a huge

impact on the size of their pension.

479

And so I think the societies' a

beneficiary of a more efficient capital

480

allocation system in the stock market.

481

If we go to, if we imagine a completely

inefficient stock market it's not going

482

to go up as fast because capital's

going to sit, idly being used poorly.

483

Rather than being allocated to companies

that use it really, really efficiently,

484

and actually in the US you have that

all the way around, starting with VC and

485

Angels and the risk tolerance right up

through into the stock market itself.

486

The whole thing is more efficient.

487

It's why you have, and I think that's

actually why you put those two together,

488

why you have the world's biggest and

most successful companies, certainly that

489

have been created with new technology,

excluding things like Saudi Aramco,

490

which was just sitting on liquid gold.

491

In terms of generating alpha.

492

So if I talk about two different types,

firstly, hedge funds and systematic, they,

493

our clients don't tell us what they do.

494

Having been a trader two of our team,

were, we're former traders as well.

495

We know certain areas of trading very

different approaches, harnessing and

496

working in very different timeframes

often than our own forecasts.

497

So they, there's so much

out for in the market.

498

That you can have tens and tens

of thousands of people all, all

499

becoming specialists in different

areas and exploiting alpha.

500

That's orthogonal to our own and ours,

ours is long term and slow moving alpha.

501

In terms of active managers, this,

this is a very interesting point.

502

So if we think about a,

a portfolio manager, the

503

our view is that the Warren

Buffets of this century will,

504

will have AI enhancement.

505

In the same way that you wouldn't

find a trader on the trading floor

506

today without a Bloomberg terminal.

507

Imagine giving one just a, a TV screen

and a phone and saying Good luck.

508

They're gonna be, they're gonna

be half blind, if not more than

509

compared to everybody else.

510

The thing for a human active manager that

we've recognized is the limitations of ai.

511

It's important to understand them.

512

So yes, it's brilliant in the kind

of data such you spoke about before

513

where you've got millions and billions

of lines of data and you just can't

514

read it, you can't even look at it.

515

It's a waste of time.

516

You're sitting there and you're

not seeing any pattern, right?

517

It's like, okay, great.

518

Unless you're brain man, and even

then you wouldn't have the, the.

519

Arm strength to, you know, flick

through the 10,000 page volume to see

520

all that data anyway, you'd be sitting

there for weeks and weeks and weeks.

521

But that works where you have a lot

of data, but there's a lot of areas

522

where you don't have a lot of data.

523

And so machine learning, particularly

in the stock market where it's

524

so noisy, needs lots of data.

525

And at three hour we call the, the

data that we work with homogenous data,

526

data that is the same for all stocks.

527

So we can use it like dollar profit

or, or price moves, analyst ratings,

528

that's, that's kind of homogenized

across all stocks has the same meaning.

529

But if we think about certain other

companies that, you know, it, maybe

530

you're an app company, maybe there's only

five o other listed app companies, you

531

haven't got enough, enough price behavior

to get any statistical significance

532

on learning the interrelationship

with that data to the stock.

533

However, the analyst, the PM who look

at that, can very clearly understand

534

its meaning and relate it and feed

it into their financial forecast.

535

How they view the stock, they, there's,

there's lots of data sets that just

536

aren't yet ready for machine learning.

537

And may never be.

538

So one of the areas where we work

on with active managers is, is

539

this kind of yin yang relationship.

540

It's not one or other.

541

The optimal is both a

bit like planes today.

542

Yeah.

543

We can fly planes without a pilot.

544

We see it with drones,

but we still have a pilot.

545

Why that?

546

So.

547

So, yes, I, it's I think there's an

education that will come with ai.

548

We're certainly invested in that and

working on that with our clients to

549

support the, the active managers to, to

realize, okay, there's a whole load of

550

lifting that you couldn't do before that

we now do that that was never getting

551

done, but really go and focus to these

other areas that are not so suited.

552

To, to AI or machine learning that

we're not gonna deal with, we're not

553

gonna be able to add value and add

additional value on top of the ai.

554

The AI just really forms a

base layer of enhancement.

555

Ryan: Yeah, Jake, that's great.

556

So let's go to your alpha int

intelligence scoring methodology and tool.

557

you've talked a lot about,

benjamin Graham, Warren Buffet, and

558

they're, you know, that's, you've

taken a lot from them the years.

559

So with your scoring, is it more, got more

of a value tilt to it than value emphasis?

560

Does it only work on just than of

value names, or can it, you know, has

561

got some growth prospects to it too?

562

Or is it all looking for, you know,

good companies at a discounted price?

563

Jake: All of the above and in, and

it's a, it is a piece of string answer.

564

So one of the things that our alpha

intelligence does is seek to work out to

565

what degree those things are important.

566

Point in time on a stock by stock basis.

567

So in 2021, we began

developing explainability.

568

This was an original just in data format.

569

So it's ugly.

570

You know, we're talking over 300 factors.

571

So there's like a, a bar chart of

over 300 factors of alpha being

572

attributed back to in underlying models.

573

Of course, within that would sit

value, would sit, growth, would sit

574

momentum, and lots of other data.

575

What we, what we source pretty much

straight away was the degree of difference

576

that the system was putting onto these

different things, across different

577

stocks at the same point in time.

578

So.

579

Some stocks are more of a value

play and, and when they're more of

580

a value play, the pattern that we're

seeing from the alpha intelligence

581

is that prior financials are more

predictable, the more reliable companies.

582

So you can begin to look

at them a bit like a bond.

583

And I think that if we think of Warren

Buffet, he avoids cyclical companies

584

and he also avoids banks with, with

derivatives in and things he can't

585

understand, or at least he said he used

to, and that that may have changed.

586

But, so you have to play to

your circle of competence.

587

And the alpha intelligence learns

that again, with, with, with growth.

588

We, in recent time could see there was

a lot more weight being put into Tesla.

589

If we go back a few years into analyst

reports and what analysts were thinking.

590

And that made than, than say Butcher

had the way which was being viewed much

591

more as with more importance and weigh

into valuation, value based information.

592

Valuation by the way we call valuation

value and growth because I think

593

you need both things to value.

594

I think so.

595

But.

596

So it, it, it is very stock specific.

597

You mentioned meme stocks earlier

we actually went back and looked at

598

some meme stocks and we saw actually,

despite the company being horrible

599

one of those particular meme stocks

as it started booming up, became very

600

highly rated, but it was really just a

technical volatility trend-based play.

601

And, and it was spotting that.

602

And, and so the way that.

603

I've gotta jump you back The way

that our Alpha intelligence learns

604

and how it differs from a model.

605

If we think about a model

input equals output, right?

606

You put the input in.

607

The model is always gonna give

you the same output that is.

608

If we think about how our Alpha

intelligence works, it's a bit more like

609

a a thing that goes back and reviews

all the movies that have ever been made.

610

So it looks at, it looks at today's movie,

you give it an outline, the script, the

611

cast, the director, the length of the film

who did the sound, et cetera, et cetera.

612

Goes back all through its history, does

pattern recognition, looks at all of

613

the movies, works out the relevance

of all the past movies and forms an

614

average of how that movie's likely

to go based on all that information.

615

And so it's looking at different

past, in the present, depending on

616

the stock that it looks at, you see?

617

So it's, it's a lot more advanced

than, than the classical kind of

618

modeling based approaches or, or

factor tilting based approaches.

619

It is really taking a very stock

specific, every stock tells

620

a story approach to stocks.

621

So yeah, and, and really it's

just learning from the past.

622

So we're just getting told

back the truth of the past.

623

It's like a super research

mechanism in a way.

624

Ryan: Yeah, that's great.

625

I was gonna bring up too, like how do you.

626

Shift through that noise of like during

COVID with the meme stocks and like you

627

said, it was all momentum technical or

you know, just, it, it, it different

628

completely forgot about fundamentals.

629

Like how do you shift through that and how

does a machine be able to like, kind of.

630

And ignore it and focus on

fundamentals, but it's hard to ignore

631

when a MC is shooting to the moon

and you know what's driving it.

632

Jake: Yeah, so we, we,

we, we don't ignore it.

633

We consider that that everything

that ever happened was truthful

634

and the key is to try and work out

what that truth was and where you

635

can't work out what that truth was.

636

It just becomes noise in the system.

637

And we accommodate for that.

638

So, which gets you to the real core

question of, of, of everything,

639

which is trying to search for

causality, not correlation.

640

What's the, so, so we think of the

stock market as a, as a highly noisy,

641

almost like statistical physics system.

642

And our mission really, or our

vision is to solve investing,

643

which means solving that system.

644

What are all the parts

that move it and why?

645

Connecting the dots, but it's very,

very important to be able to have

646

methods that recognize they're

not connecting the dots and have

647

elements of doubt in what they do.

648

I don't think I've answered your question

specifically in the way that it would

649

be discussed in, in investing community,

but I don't think the way the investing

650

community is to think is the right

way to solve the problem after 20 odd

651

years of concentrating on the problem.

652

If it was, we, we, we wouldn't

exist today through ai.

653

So, but yeah, yeah.

654

We're, we're always

gonna have these things.

655

Like I said before, I think the stock

market is a bit like, did you play poker?

656

Ryan: A little bit.

657

Not very good, Jake, not very good.

658

Jake: Okay.

659

Yeah.

660

So, so if you think about playing

poker, a bad poker player, when

661

the flop comes, will say, oh, crap,

I've, or, or they go all in or

662

something, and then the flop comes.

663

It wasn't how they had hoped it might be.

664

They lose, they think,

oh, I made a mistake.

665

A really good play player might win a

hand and think they made a mistake because

666

they didn't play the distribution right.

667

I think stock returns are as

noisy or, or even perhaps, noisier

668

than, than, than, than the flop.

669

In, I say they are noisier

than the Flopping Poker.

670

And so the important thing is to really

learn to, or what we seek through AI is to

671

learn to the long-term truth of of stocks.

672

So that means that we see the full

lifecycle of over 150,000 stocks.

673

And through enough data, if we see a

stock from birth through to death, and

674

the average stock life, by the way,

is seven years, if you go through all

675

the data, the data that we've been

through, which is not long, but at

676

the same time, it takes all the atoms

in our body apparently to replace.

677

Although I've looked that up online.

678

I'm not sure if that's actually

true, but, but it's a commonly

679

assumed thing that we say.

680

Then it's not really that long, but

the, our, our simple principle is that.

681

Like I said at the beginning,

companies are owe us colorless

682

cash printing machines.

683

And from the perspective of a shareholder,

if I could know the rate at which

684

it prints cash or needs to eat it to

keep going the rate of change in the

685

printing of that cash and or whether

or not that cash printing machine was

686

vibrating with smoke coming out of it,

IE defaulting not gonna exist tomorrow.

687

And what you want, you want, and if

the price of it is how much floor

688

space it takes and your floor space

is your capital, you really want small

689

printers that print cash rapidly are

increasing the rate at which they

690

print cash and look sturdy as hell.

691

They're gonna be there a hundred

years and, and it's, and it's that

692

simple and everything else is noise.

693

You see, if you think mathematically,

the Warren Buffet talks about this

694

as well, you know, when talks about

things like Clayton Holmes who didn't.

695

Visit management.

696

When we bought them, we looked at the

record to see what management was like.

697

We, the focus here is, is toward

not just measurement of the past and

698

Warren Buffet, which use his wonderful

mind to estimate into the future.

699

We, we are money balling that in a

way, using all the data possible to

700

estimate the future in the present.

701

In the past, focus much

more to measurement.

702

And then comparing all the world

stocks and, and, and that's how we're

703

sourcing and generating Alpha, which

is, which is kind of money balling.

704

And it is, and actually it's like the

film Moneyball, if you've seen that.

705

Because in that film, when I watched it

back, my jaw almost dropped because it was

706

almost a direct parallel to what we do.

707

We cover 20,000 stocks in that film,

they took about 20,000 baseball players.

708

In that film, they find the baseball

players that maybe aren't gonna sell

709

t-shirts or, or, or, or trading cards.

710

They're not necessarily gonna be on the

cover of a magazine, but statistically

711

we're good and certainly undervalued.

712

A lot of the world's best stocks

are hiding in plain sight.

713

They're just not sexy.

714

People are not interested in them.

715

Might be like a lift company

out in Germany that sells lifts.

716

Might be some industrial

chemical manufacturing

717

company you've never heard of.

718

That are serious cash machines and trading

at a decent price and, and growing profit

719

nicely, and looking after shareholders

doing buybacks, you're ending up with more

720

of the same stock, et cetera, et cetera.

721

So we've sought to go much, much deeper

into what we make visible to our systems

722

than, than what we see out there in the

systematic investing space, enabling AI

723

to answer and ask, or ask and answer.

724

Lots and lots and lots of questions.

725

So effectively our systems

will ask different questions

726

depending on what they get.

727

The answers back they, it

is investigative as well.

728

Again, you wouldn't get

that in classical modeling.

729

You wouldn't have that level of

intelligence in how it thinks.

730

Ryan: I love your comparison

to the movie in sports.

731

I love it.

732

It is a great comparison.

733

Let's.

734

Jake: It is really the, it is,

it is a very much a a similar,

735

similar thing, really similar thing.

736

Ryan: Let's finish.

737

Great conversation.

738

I learned a ton.

739

I'll probably go back and

watch this a couple times.

740

Has taken it all in.

741

All the great insight.

742

Let's just finish.

743

So financial advisors out there

that are watching, how can they

744

implement strategies like this?

745

Like they find it really interesting.

746

They wanna implement some AI and more

quant into their client portfolios.

747

Where can they start?

748

Jake: So are we talking about a

financial advisor that's actually

749

creating strategies for their client

or allocating their clients into funds?

750

Ryan: If

751

Jake: Both.

752

Ryan: that are are creating investment

portfolios, whether it's mutual funds,

753

ETFs, individual stocks for their clients.

754

Jake: Okay, so I mean, I'm

working for three ai so I'll

755

tell you about three AI side.

756

We can effectively provide

an alpha intelligence.

757

Interface with pretty much a global

coverage of every stock in the

758

world, over $50 million market cap.

759

It is instantly searchable

to find the world our highest

760

rated, lowest rated stocks.

761

We have.

762

And deep explainability, like

I said, about 300,000 pages of

763

equity research updated weekly.

764

So it's, I think the average analyst

765

So I guess you'd say it's about a thousand

analysts, but we tested it against the

766

institutional brokers estimate system and.

767

The, which covers 18,000 sales side and

our forecasting accuracy, we're explaining

768

alpha for the following year, circa

20 times higher than analyst ratings.

769

In the, in the, since we've been

770

If the for funds we're, we're

launching AI powered funds with s and

771

p Global, so should we say indices?

772

So we, we, we, we, we provide our

forecast data to s and p Global.

773

They're creating indices with it to

create enhanced versions, alpha seeking

774

versions of the s and p 500 s and p world.

775

And there'll be more to come.

776

Actually, we've got.

777

Clients in the Middle East and, and,

and Asia now are showing interest.

778

And I think we'll be global

with those products in the

779

not, not too distant future.

780

Should we speak again?

781

I think we might well be so, but

I, I, I would recommend to, to, to

782

do due diligence if, if for example

people are using AI systems for Alpha.

783

I think it's really important to

statistically analyze the performance.

784

Of the of the forecasts.

785

And if the, and if the statistical

performance of the forecast is robust,

786

then, and you understand why the apples

and pears of it, there's actually

787

fundamental reasoning, grounded in

reality, not in data, but in the

788

real world that we can see with our

eyes, then, then, then sit with it.

789

Like all good, like all proper

systematic approaches or, and stuff.

790

You can sit with that and you

should get paid out in time and

791

outperforming in time, but obviously

nothing's a guarantee on anything.

792

The market can be, you know, the

old saying the market can be crazier

793

than we've got the wallet to kind

of stomach, but which is true.

794

Ryan: Wow, Jake, fantastic conversation.

795

I, like I said earlier, I learned

a ton of really interesting stuff.

796

Thank you so much for coming on.

797

Really an honor to have you on.

798

Where can our audience get more

information about three ai.

799

Jake: We've got a website like everyone.

800

So you can go to our

website, www.threeai.co.

801

And see us there.

802

I must be honest with you, we have

very limited information on our website

803

because our clients are on the buy side.

804

We're an alpha unit, right?

805

Like, like we see ourselves as an, as

a kind of a team with our, with, with

806

our clients that we're, we're almost

like an, you could imagine like an out

807

outsourced like research alpha r and d

center and trying and supplying that and

808

distributing that back into industry.

809

So our website is, doesn't contain a

great deal of information on there, but

810

if, but if you have institutional clients

that wish to work with us, then they

811

could reach out to us via the website.

812

There's videos on there too, but

813

I did a talk at when we were

814

the university at Harvard in I

think 22 or UCL, the same year.

815

As well.

816

And you can see in those, actually some

of the things I spoke about today, what

817

we, what we were seeing in forecast,

how, how we were seeing an scur in

818

the future of stock returns with our

lowest rated, highest rated, looking

819

somewhat like an S with significant

alpha being generated across a

820

hundred trillion dollars of stocks.

821

So if you have, you have clients

that have active managers they

822

wish to, to produce new, innovative

strategies that seek to beat the

823

market, not conventional, thematic.

824

You come up with thematic,

we'll turn up and power it.

825

We'll put the engine in it.

826

And and if you've got quants, you

listen, by all means reach out as well.

827

'cause we have unique and

proprietary data sets.

828

So our quant customers have told

us that roughly 50% of the alpha

829

are unlocking is on scene to them.

830

And I think, I think we're really

unlocking the company behind the stock in

831

a way for, for the kind of quant AI era.

832

Ryan: I love it.

833

Jake, thank you so much.

834

Like I said, thank you.

835

It was an honor.

836

thank you so much for listening

to this episode of zephyr's

837

Adjusted for Risk Podcast.

838

You can watch all of our other episodes

on the Zephyr YouTube channel and Spotify.

839

Please be sure to like and

subscribe to those channels

840

and give us follow on LinkedIn.

841

you very much and have a

great rest of your week.