Frontiers: Exploring the World of Data | Revolutionizing Sports Betting Through Analytics

Sep 12, 2023

Frontiers: Exploring the World of Data | Revolutionizing Sports Betting Through Analytics

In this episode our guest host, Jazmin Furtado, speaks with Brian Beachkofski about his work with Rithmm, a company that provides predictive analytics for sports betting.

Brian shares about…

  • How he got into this industry.
  • How his team applies data and analytics to change the sports betting playing field.
  • How they’ve tackled common industry problems, from handling proprietary data to maintaining a competitive advantage.
  • The future of the field, including advice for people who are interested in entering the space.
  • And much more!

About today’s host: Jazmin Furtado has been a part of the software innovation realm for the Department of Defense where she has overseen large-scale Data and Artificial Intelligence programs in the Air Force and Space Force. She has also held various leadership and advisory roles with organizations such as Google, SpaceX, and Massachusetts General Hospital, where she designed and scaled AI, data, and education and training programs.

About today’s guest: …

About the series: Our new series, “Frontiers: Exploring the World of Data”, dives into how people are using their data science minds to shape organizations and change the landscape outside of “Big Tech”. In each episode, we explore the far-reaching corners of the world of data. If you’re curious about how data-minded individuals are making a difference in interesting, impactful and creative ways, then tune in!

Sign-Up for the Weekly hatchpad Newsletter: https://www.myhatchpad.com/newsletter/

Transcript
Tim Winkler:

Hey listeners, Tim Winkler here, your host of the pair program.

2

:

We've got exciting news introducing our

latest partner series beyond the program.

3

:

In these special episodes, we're passing

the mic to some of our savvy, former

4

:

guests who are returning as guest hosts.

5

:

Get ready for unfiltered

conversations, exclusive insights,

6

:

and unexpected twist as our alumni

pair up with their chosen guest.

7

:

Each guest host is a trailblazing

expert in a unique technical field.

8

:

Think data, product management,

and engineering, all with a keen

9

:

focus on startups and career growth.

10

:

Look out for these bonus episodes

dropping every other week,

11

:

bridging the gaps between our

traditional pair program episodes.

12

:

So buckle up and get ready to

venture beyond the program.

13

:

Enjoy.

14

:

Brian Beachkofski: Hello

15

:

Jazmin Furtado: everyone, and welcome to

Frontiers, exploring the world of data.

16

:

Frontiers dives into how people

are using their data science minds

17

:

to shape organizations and change

the landscape outside of big tech.

18

:

In each episode, we explore the far

reaching corners of the world of data.

19

:

My name is Jasmine and I'm

your host for this series.

20

:

I myself am passionate about

empowering people to make data

21

:

driven decisions, and I'm always

amazed at how others do it every day.

22

:

Today, we're exploring the

sea of sports analytics.

23

:

How advanced analytics can

reshape the sports landscape.

24

:

With that, I want to introduce our

guest for today, Brian Beachkovsky.

25

:

Brian is a leader in innovation

who has worked in a variety of

26

:

industries in both the private and

public sectors, but a common thread

27

:

has been his focus on creating value

through software, leveraging data

28

:

and analytics to empower people.

29

:

I had the pleasure of working for him

while at Kessel Run, an organization

30

:

building software for the Air Force.

31

:

And while Kessel Run was only a small part

of his career, he made a lasting impact

32

:

on not just the organization, but also the

team he led and the users he supported,

33

:

which I strongly suspect is a theme for

him with all the teams he has worked with.

34

:

He is currently co founder and CTO

of Rhythm, a tool that provides

35

:

predictive analytics for sports betting,

which is our main topic for today.

36

:

And I'm so curious about how he

got into this space, into this...

37

:

Arena, um, but more of

that to come in a bit.

38

:

So thank you so much

for being here, Brian.

39

:

It is a pleasure to say the least.

40

:

Brian Beachkofski: No,

thanks for having me.

41

:

I look forward to the

42

:

Jazmin Furtado: conversation.

43

:

It should be fun.

44

:

So start, uh, speaking of fun, our

icebreaker question, I wanted to start

45

:

off to, uh, get the ball rolling.

46

:

So the question is, if you were to

compete in one sport on the world

47

:

stage, does it matter whether you're

good at it or not you to choose one?

48

:

Which sport would it be?

49

:

And why?

50

:

So I'll start with, I would choose rock

climbing because I got into it while I

51

:

was in college and I really enjoyed it.

52

:

Uh, and I really liked how it's kind

of like a puzzle and you feel kind of

53

:

like a superhero and a ballerina at

the same time while you're doing it.

54

:

So it's like a really cool mix

of like brain strength and grace.

55

:

So I just think it'd be really

cool to just do that on the world

56

:

stage and not have to worry about

whether I'm good at it or not.

57

:

Just, I think it'd just be

58

:

Brian Beachkofski: really fun.

59

:

Yeah, that's awesome.

60

:

I love, I love climbing, but

I think I would pick hockey.

61

:

I, uh, I grew up playing hockey.

62

:

I'm I love, I love ice hockey.

63

:

I was a goalie growing up and,

uh, uh, there's just no better

64

:

sport in the world in my eyes.

65

:

It has everything from.

66

:

Speed and contact to, uh, uh, strategy

and, and all of it mixed together

67

:

and watching the elite athletes play

is, is just an amazing thing to do.

68

:

And to be able to be out there on

the ice with them, even if I was

69

:

the worst person there, amazing.

70

:

Are

71

:

Jazmin Furtado: you still

72

:

Brian Beachkofski: playing now?

73

:

Oh yeah.

74

:

I was still in men's league.

75

:

That's one of the things that has one

of the highest rates of people playing.

76

:

And we have a guy who's

65 on our, on our team.

77

:

So, uh, there's really no

upper, upper end of that.

78

:

You can, you can keep going with the,

with everyone far into your later years.

79

:

Jazmin Furtado: Wow.

80

:

Yeah.

81

:

I imagine on the East coast,

there's like a lot of that, a lot

82

:

of snow, especially Northeast.

83

:

So it's a big, big field out there.

84

:

So, Brian, our main topic for today is

to discuss how you've applied, like, your

85

:

data and analytics mindset to change the

playing field in sports, uh, specifically

86

:

when it comes to sports betting, but

you haven't always been in this space.

87

:

So, uh, I'm curious if we could start

off with you speaking 1st about rhythm

88

:

and speak about, um, what you do and the,

the tool and then walk us through and

89

:

take us back a little bit how you got.

90

:

Brian Beachkofski: Yeah, absolutely.

91

:

Uh, so, so like you said in the

intro, rhythm helps, uh, helps the

92

:

average person make smarter bets.

93

:

What we realized is that people have been,

uh, betting the same way for an eternity.

94

:

Uh, they, they go with their gut.

95

:

They ask their buddy who they like

on tonight's game, or they just

96

:

think someone's due or, uh, have some

other hunch based way of doing it.

97

:

And that just seems out of place

in a world where that person

98

:

probably won't even pick a

restaurant based on a hunch, right?

99

:

They check, they check for reviews

and they look at the data and they,

100

:

they make a data informed decision

on, on where to get dinner, but they.

101

:

They go with their gut when it comes

to putting money down on a bet.

102

:

So that shouldn't be.

103

:

And like, why, why is there

not something out there?

104

:

And it looked like, uh, it, it hadn't

existed because there are a couple

105

:

of things that made it hard to, to

create models for people, um, uh,

106

:

and to let people have that insight

into their own modeling approach.

107

:

So that's basically our goal is to

make, uh, to make modeling easier for

108

:

people to understand sports betting.

109

:

Uh, Ultimately trying to get to a no

code analytic service for folks to

110

:

make their own models to play around

with things and have access to tools

111

:

that will enable them to make data

informed decisions when they're,

112

:

when they're doing sports betting.

113

:

Uh, so the other part of the question

is how do, how do I end up there from,

114

:

uh, all the other things I've done?

115

:

Um, well.

116

:

I started at, in the air force research

labs doing analysis of jet engine

117

:

reliability and like, uh, how do we

make maintenance decisions better?

118

:

When can you predict, uh, maintenance

will be needed or when there

119

:

could be a higher probability.

120

:

Of an incident with a, with an engine,

uh, but believe it or not, uh, even

121

:

within the air force, it's hard to

get engine data, uh, the, the, the

122

:

companies, uh, who manufactured the

data on that data, there's usage, right?

123

:

So I found it was much easier for

me to go out and scrape sports data.

124

:

Like I said, I grew up playing sports.

125

:

I like sports.

126

:

Uh, so being able to learn some of the

analytical, uh, approaches and apply it on

127

:

sports data was just a way for me to teach

myself and have that internal professional

128

:

development, uh, while, uh, while trying

to apply that within the Air Force.

129

:

So that was my, my first approach into

sports analytics was, was just using

130

:

that to teach myself the techniques

that I was using for my day job.

131

:

Um, from there, I worked in the Pentagon

for a while, doing some budget analysis,

132

:

doing, uh, uh, uh, more analytics

around, um, patent and scientific

133

:

journal publication data to identify

emerging tech, uh, research areas.

134

:

So that got more into, uh, the

language modeling there and in

135

:

predictive analytics from that side.

136

:

Uh, and then after that job, I went

to private industry and worked at a

137

:

startup, uh, in the nonprofit sector.

138

:

Thank you.

139

:

That was doing data analytics to

understand the impact of human

140

:

service delivery in a local community.

141

:

So how can we use that community's

data to understand if work

142

:

programs are actually developing

people into a career as intended?

143

:

Are programs for homeless services

getting people self sufficient and

144

:

in stable housing down the road?

145

:

But still using that to understand,

using the data of outcomes to understand.

146

:

How to make those

decisions up front better.

147

:

So that was like the path I got here.

148

:

And then I met my co founder, Megan,

who said, what if we did something

149

:

similar, uh, for sports betting?

150

:

And I said, yeah, I think, you know,

at that time I was then doing college

151

:

basketball analytics as my side

project, trying to teach myself stuff.

152

:

And, uh, she was a division one

basketball coach in a former life.

153

:

So being able to, uh, connect there.

154

:

And, uh, put what was, you know, a

hobby at the time into practice was

155

:

Jazmin Furtado: awesome.

156

:

And I think when people talk about their

careers today, in retrospect, you're like,

157

:

Oh, these were like the common things

are like, you know, it just seems to all

158

:

piece together when you look back at it.

159

:

Is that how it was kind of unfolding

to you as you were experiencing

160

:

it, or were there a lot of.

161

:

Whether it's a be dead ends or a lot of

exploration and other fields to see what's

162

:

stuck, you know, what was the journey

like, as you were going through it?

163

:

Brian Beachkofski: Yeah, that's

a great question because no,

164

:

that story did not play out.

165

:

Like, it does have a common

theme in retrospect, but it was

166

:

not part of a grand strategy.

167

:

Uh, uh, I.

168

:

I left the research lab

because I was at that point.

169

:

A lot of us get to, uh, when, when

you get around 30 of, am I going

170

:

to be a deep technical expert

or go into a managerial track?

171

:

And, uh, I realized that I was

more interested in making an

172

:

impact in a way that was possible

from the management side of.

173

:

The, uh, the business.

174

:

So that's where I went into the

Pentagon to get a breadth of

175

:

experience within the, my air force

career to be able to set that up.

176

:

Uh, and then when I, when I went

to third sector capital partners,

177

:

uh, that, that nonprofit, uh,

social impact team, uh, that was.

178

:

Uh, relationship I had developed

at business school that they were

179

:

looking for someone to expand

their quantitative analysis.

180

:

And I had that experience.

181

:

So part of that career track was

based on my quantitative experience

182

:

bringing it in, but it was, it was there

because it was an interesting problem.

183

:

Uh, and, and I wanted to solve hard and

interesting problems and it was there.

184

:

And that's, that's a continuing theme too,

is just take something that's interesting.

185

:

And I think your career.

186

:

Story will write itself in retrospect.

187

:

If you keep challenging yourself with,

uh, interesting and challenging problem,

188

:

meaningful problems that actually address,

uh, some, something someone needs fixed.

189

:

Jazmin Furtado: And it, but with sports

betting or with sports generally,

190

:

you've always had an interest.

191

:

It's something that has been something

that's kind of been on the outskirts

192

:

and then you've been able to like

tap into, um, as you follow your more

193

:

professional or like formal career, I

guess, path, um, for lack of better terms.

194

:

Uh, The sports interest, though,

became more of a reality later

195

:

on in your career in terms of

an actual career move or pivot.

196

:

What was it about the timing of that

or was there like, what other factors

197

:

at play that made you realize, okay,

this is a good time to actually get

198

:

into this space and actually start to

capitalize on these interests of mine.

199

:

Outside of just something that's

more on the side of more of a hobby.

200

:

Yeah,

201

:

Brian Beachkofski: it's a,

it's an interesting story

202

:

because the timing was sort of.

203

:

Uh, fortuitous, I guess.

204

:

So it started out with me and Megan,

uh, talking about, uh, is this possible?

205

:

And it was a hobby.

206

:

So on nights and weekends wrote some

Python code and, uh, uh, was making

207

:

predictions, uh, for games and, and

tracking out how they were actually.

208

:

Working in a spreadsheet, and we did

that enough and said, actually, it

209

:

looks like this is working pretty

well and outsourced a very rough first

210

:

iteration of the web app to go out there.

211

:

And we actually spent about 2 years.

212

:

Just collecting data with about 50 friends

and family users that were tracking things

213

:on there, uh, and got about:

of predictions and showed that there

214

:

was, that it was working and Megan was in

business school at the time and wrote it

215

:

up as a, as a project plan for one of, uh,

one of her classes to, to build out a, uh,

216

:

a pitch deck and business plan and through

a friend that got into, uh, uh, Um, A VC

217

:

seed company, uh, who wanted to take a

look and got interested and made an offer

218

:

and the timing worked out pretty well that

it was right at the time my command tour

219

:

at Kessel Run was coming to an end too.

220

:

So, uh, it was actually a little earlier

than we were planning it to be, uh,

221

:

but that's how, you know, if, if you

have a hobby and you're developing

222

:

your skills and the data is there and

someone else is making a data driven.

223

:

Decision, it can, uh, it

can, it can drive your time.

224

:

Jazmin Furtado: Were there other, uh,

like more macro influences at play in

225

:

terms of like the landscape of sports

betting or the landscape of technology

226

:

that really enabled or jumpstarted the

development or the fruition of this tool?

227

:

Brian Beachkofski: Yeah, there was a

Supreme Court decision with PASPA a few

228

:

years ago, which was the one that took

away the federal prohibition against

229

:

states, uh, making sports betting legal.

230

:

Uh, so that, that

happened a few years ago.

231

:

And so you saw more states coming on and

in passing legalization of sports betting.

232

:

Um, and I think that drove a lot more

investment into this area of tools

233

:

and, and how to foster that what people

knew was going to be a growing sector.

234

:

Uh, so they're making investments there.

235

:

So that was definitely

part of the timing as well.

236

:

Yeah, I can

237

:

Jazmin Furtado: imagine that with all of

the, the increased focus or the ability

238

:

to be able to, um, put more out into the

market and the space that there's a lot

239

:

of competition is probably like a lot

of people that are trying to, you know,

240

:

it's like a little bit of a new frontier.

241

:

Um, I was wondering what.

242

:

You found and how you came upon

your value proposition and your

243

:

niche, like, how to differentiate

yourself in such a growing field.

244

:

Brian Beachkofski: Yeah, no.

245

:

So, so this is where focusing on the user

problem is something that you hinted.

246

:

We probably get to, uh,

was really important.

247

:

Um, because, uh, the, the vast majority

of companies out there that are providing,

248

:

uh, pick support are just telling you

what to pick and are not transparent

249

:

at all about how they got to that side.

250

:

Um, so people follow them until

they don't win and then they move

251

:

on and you really don't know.

252

:

And there's, there's quite

a bit of fraud, right?

253

:

Right.

254

:

Famously, a lot of pick services in the

past would call and tell, uh, let's say

255

:

there was a game on tonight, they would,

the first person to call, they'd say, uh,

256

:

take team a, the second person to call,

they'd say, take team B and they know that

257

:

half their customers would win, right?

258

:

So, like, that is, uh, uh, fraud.

259

:

I think is a good way to say it, but when

somebody wins their first bet, they're

260

:

more likely to pay for your service.

261

:

And, uh, but, but you could do that

when there's nothing behind it, right?

262

:

Like when there's no transparency

in what your record is.

263

:

So like part of this is what is,

what is the core user problem?

264

:

One, the user doesn't have any

idea of whether or not it's

265

:

working or it's not working.

266

:

Um, and, and they are, are still not

solving the core problem, which is.

267

:

I know when people who bet sports are

usually interested in sports and have

268

:

an opinion, but they're calling a pick

service because they don't have a way to

269

:

put that opinion into their own model.

270

:

Right?

271

:

So if they think, I think offense

is very important, or I think

272

:

high tempo is very important.

273

:

They.

274

:

The average person doesn't have

the time or the skills to make an

275

:

analytical model that would turn that,

that, uh, theory into, into practice.

276

:

That's what we're trying to solve

is how do you take someone who's

277

:

interested in sports, has a theory

about sports, but doesn't have the

278

:

modeling skills or the time to maintain

their, their data feeds and to, to

279

:

have their server up and running and

constantly doing the analysis on it.

280

:

If we offload that and just make

it very easy for someone to say,

281

:

here's what I think is important.

282

:

And then get that.

283

:

What does that mean in terms of my

expected value on all these plays?

284

:

That's what we're trying to solve.

285

:

And that was actually, uh, there's not,

there's not anyone out there in the

286

:

market that is truly focused on that

problem of how do you provide no code

287

:

analytical model development for users?

288

:

Jazmin Furtado: How do you find that

sweet spot when it comes to transparency?

289

:

Versus accessibility, or, like,

understanding, you know, when it comes to

290

:

these models, they get very complicated.

291

:

And.

292

:

I imagine when you're, you know, trying

to tell people what your model is doing,

293

:

you have to figure out how much to be

able to, you know, what's, what are

294

:

the important things to get across?

295

:

So how did you figure that out?

296

:

And then what, where did

you find that sweet spot?

297

:

Brian Beachkofski: Yeah, so this

is, this is a real importance

298

:

of having a great team.

299

:

So I am a quantitative

background kind of person.

300

:

So my default position is actually to over

explain and to get into too much detail.

301

:

Megan comes from a different

background that hasn't had that

302

:

quantitative, uh, uh, experience.

303

:

So she's always the tension between us

is, is, is how do you get it so that

304

:

you are explaining what's happening

without getting so technical that you

305

:

lose people and actually having, uh, two

folks with very different perspectives on

306

:

the same problem, trying to communicate

it to people is incredibly helpful.

307

:

So that's been really good and

how we've balanced that what we

308

:

don't have that tension around us.

309

:

We always want the UI to

be as simple as possible.

310

:

A lot of a lot of stats pages out

there give the error of give the error

311

:

of being, um, Technical by making

the UI, like actually terrible, they

312

:

overwhelm you with numbers and graphs

and plots and things to make it feel

313

:

like something's going on there.

314

:

But actually pulling information

out is hard when they just

315

:

dump raw numbers at you.

316

:

So that's that's the key for

us is how do you make the UI?

317

:

Simple, sleek, easy to understand,

while also having the technical level of

318

:

support be really high while not confusing

people by overwhelming them with detail.

319

:

So that, the tension is where do

you, how do you explain it in a way

320

:

that doesn't overwhelm people, uh,

while keeping that UI very clean,

321

:

very straightforward and simplified.

322

:

Jazmin Furtado: So there's, I can

see there's, there's really a level

323

:

of creativity when it comes to like.

324

:

What visually does this need to look

like, but I imagine there's also a

325

:

level of creativity on the back end in

terms of how everything's put together.

326

:

Um, how do you like translate

user inputs into something that's

327

:

actually valuable to the model?

328

:

And then, you know, what do you wait

most important with, you know, that

329

:

stuff going on in the background.

330

:

So how did you come up, come to what

were the important things to consider?

331

:

What, or the, what, what do

you need to get from your users

332

:

versus where do you need to get

from else more automated sources?

333

:

So can you talk a little bit more

about how you come to the backside, the

334

:

Brian Beachkofski: backend?

335

:

This was a really

complicated problem as well.

336

:

And I think this is one where if, if

there's core proprietary information that

337

:

we have, it is in that, how do we simplify

it while still having a complex model?

338

:

So, because, um, so in the basic

sense, what we we've done is for every

339

:

sport, we do the feature engineering

as a whole, uh, and do a lot of the

340

:

testing to make sure that the baseline.

341

:

Models are predictive, uh, and then

to make, which would be great, but

342

:

now you only have one model, right?

343

:

So now how do you get personalization when

you've done all of the feature engineering

344

:

to, to meet the baseline problem?

345

:

That's that actually reduces

the flexibility in the model.

346

:

So how do you reintroduce the flexibility?

347

:

We got our inspiration from another

data scientist I was talking to about

348

:

these issues, and he said, well, for

computer vision, they can change this

349

:

sensitivity for cancer screenings.

350

:

If you take a biopsy and you have

a slide of the biopsy, the way they

351

:

train it to be sensitive to the

cancer cells is in the training

352

:

set, they overrepresent Thank you.

353

:

Thank you.

354

:

The cancerous cells so that it becomes

more sensitive to those pieces.

355

:

Now we took that as inspiration.

356

:

And so we take our baseline model

and we have the sensitivity for the

357

:

classification of every, of every

game in history, uh, that in our

358

:

dataset to each of those factors.

359

:

So now if the user says, I

think offense is more important.

360

:

We take the games where offense

was more determinative of how

361

:

that game was classified by the

model, and over represent games

362

:

like that in the training set.

363

:

So we make a bespoke training set for

every individual user, where the model

364

:

dynamics underneath it are the same,

but the differences in training make it

365

:

more sensitive to different features.

366

:

Per the per the users, um, intuition

367

:

Jazmin Furtado: that we're, um, you're

talking about features and I get very

368

:

excited whenever we're talking about

features because I think it's just.

369

:

The hard part and a lot of data science

applications is, like, finding the

370

:

data and, like, getting good features.

371

:

Do you have that issue when it comes

to sports and for sports betting?

372

:

Do you have a problem with, like,

finding good data or finding enough data?

373

:

Uh, where are your challenges

if any, when it comes to.

374

:

Obtaining the database.

375

:

Brian Beachkofski: So, so the good

part in sports, there are a lot of,

376

:

uh, good data providers out there.

377

:

So we actually work with a

data provider company, uh,

378

:

who gives us historical stats.

379

:

Uh, and that's all been, um.

380

:

Curated and so we don't have to

do a lot of the time consuming of

381

:

data cleansing and keeping your

scraping feeds and things open.

382

:

So that that has been very helpful for us.

383

:

Um, we do we do do a little

bit of scraping here and there.

384

:

And that takes a little bit of time.

385

:

For example, like we know what we know.

386

:

It's very important to get how far a

a road team travel to get to a game,

387

:

uh, both in terms of miles traveled,

but elevation change is really A

388

:

factor, uh, having gone to the Air

Force Academy, which is 7, 258 feet

389

:

above sea level, I can tell you,

uh, altitude or elevation matters.

390

:

So, uh, we have that.

391

:

It's like

392

:

Jazmin Furtado: you

memorized that or something.

393

:

Brian Beachkofski: Yeah, yeah.

394

:

Uh, so they, uh, so we have to

scrape some of that and use,

395

:

uh, some other API interfaces to

turn that long into an elevation.

396

:

Um, and, and those can be a bit tricky.

397

:

But basically, we were able to

procure the information we need

398

:

from data providers, uh, and then we

also, um, are able to make our own

399

:

proprietary, um, uh, stats internally.

400

:

So, um, we use a Bayesian ranking process

to be able to take, uh, the results of

401

:

games plus, uh, uh, Initial estimates of

team rankings and then take the initial

402

:

estimates as a prior, you, you put the

actual game results in, uh, as observed,

403

:

and you can get a Bayesian ranking

formula out of that and how we, what,

404

:

how we actually use those parameters

is how we make our own ranking system.

405

:

But so we do that as well, where

we take those source inputs that we

406

:

get from our data provider and then

build our own derived statistics

407

:

off of, off of those as well.

408

:

Jazmin Furtado: Yeah, when you can

offload that, like, data wrangling and

409

:

data cleaning, it's amazing how much

free time that frees up so you're able

410

:

to actually, like, really focus on,

like, the meat and potatoes of the work.

411

:

Brian Beachkofski: Well, and it's still

like, as they famously say, 90% of your

412

:

time is in the data cleansing, right?

413

:

And the treating of it.

414

:

And then what we found is the other,

still, even when that's taken care of, 90%

415

:

of our time is still in those edge cases.

416

:

So like in football, wacky things happen

all the time where like, uh, this, this

417

:

play didn't end right, or this happened.

418

:

So there's always weird things that happen

in sports that you still are like, Oh,

419

:

I got to take, I got to figure that out.

420

:

I got to solve for this edge case.

421

:

And you go, why, why

does that not look right?

422

:

And then you still have to go back.

423

:

So we still spend about 90% of our time

looking at the edge cases, even when

424

:

we, when we get clean data that come in.

425

:

Jazmin Furtado: Interesting, I'm

wondering what your observations

426

:

have been when it comes to sports and

sports data and the sports analytics.

427

:

So, what have you seen as unique

to this field that you haven't seen

428

:

in the other work that you've done?

429

:

Brian Beachkofski: Yeah, so I'll say.

430

:

The thing that has been most different is

the rate of uptake of insights from data.

431

:

Um, and what I mean by that,

I can give a couple examples.

432

:

One is, uh, when you find someone

who's doing very interesting analytics

433

:

work, and oftentimes it's like on

Twitter, you just find people who put

434

:

interesting results out there on Twitter.

435

:

Um, Some of them get picked up by the

pro teams to be on their analytics team.

436

:

So, uh, uh, Namita was a great

follow and she's now on the

437

:

Kraken staff, uh, in the NHL.

438

:

So like you, you see, um, you see a

quick adoption of, if somebody has

439

:

something that is actually providing

real value to an organization, or it

440

:

could be useful to an organization,

they go after it very quickly.

441

:

And you see that.

442

:

Um, from the beginning of, uh, uh,

Moneyball all through today, I was

443

:

listening to a podcast before where

they said, you know, um, An advantage in

444

:

draft order now persists a year or two.

445

:

So if you're an organization and you

find some little metric that gives you

446

:

an advantage to be able to get someone of

higher quality than you should be getting.

447

:

You can maintain that advantage

for a year, maybe two.

448

:

And then other people start asking,

why would they have picked that person

449

:

and can actually back into what you

were doing and they copy each other.

450

:

So that speed of adoption is

incredibly different than it was.

451

:

Like when I was working with state

and local governments and they said,

452

:

Hey, can you, can you get us an AI

algorithm to predict what's the best

453

:

practice to be able to, uh, to help

children, uh, children in our community?

454

:

And I say, how do you use data today?

455

:

And they go, well.

456

:

We have one person with a

spreadsheet in the back office.

457

:

I'll go again.

458

:

So we can't, we can't take you there

overnight, but that's like the difference

459

:

is, um, the desires there, but the actual

adoption rate is so much faster in sports

460

:

than it was in other areas I've worked.

461

:

Jazmin Furtado: Yeah, it really

requires you to stay on your toes.

462

:

Yeah, does that mean that the competition

is just really cutthroat, like, it's

463

:

hard to, like, maintain a competitive

advantage, um, because the, your

464

:

algorithm is only as good as its last run.

465

:

Brian Beachkofski: Yeah, that's right.

466

:

It's it's a classic Red Queen

problem where you have to run

467

:

faster to stay in place, right?

468

:

So that's what we do all the time.

469

:

We're always looking at our

model predictions and making sure

470

:

that there's still predictive

that things are still working.

471

:

So this is we've had over 60, 000

track bets within the rhythm platform,

472

:

and we have a positive almost 1%

ROI across all track bets, which is

473

:

great when you compare it to you.

474

:

The minus 5% that a naive better usually

gets so that, uh, we, we track that and

475

:

we want to make sure that we're still

offering massive value to our, uh, to our

476

:

users, but that's, you can't just say I

had a good model and then next season,

477

:

assume it's going to work every, before

every season, we have to go back in and

478

:

retune and tweak and see if there's any

learnings that we can incorporate in

479

:

because you have to make those advantages

just to keep the same position you've had.

480

:

Jazmin Furtado: Yeah, you

have to really stay on top of.

481

:

On the latest and greatest, like, outside

of just like, the, the, you can't just

482

:

look at your model to see how it performs.

483

:

You really have to keep a not

awareness of the whole industry

484

:

in terms of how that affects your,

your team makeup and the people that

485

:

need to come together to build this.

486

:

Do you need to have a

diversity in, you know.

487

:

Backgrounds, or do you find that

people really are just like clumps?

488

:

You know, you have a lot of you are

able to get a lot of diversity and

489

:

like, you know, few core individuals.

490

:

What does your team makeup have to do

to reflect the challenges of the field?

491

:

Brian Beachkofski: Yeah, it's, it's

interesting because we, uh, we talked a

492

:

lot about this up front and, and I think,

I mean, I, I've even brought it up here.

493

:

The diversity of viewpoints,

even between the CEO and me

494

:

has made our product better.

495

:

And we definitely are, are very interested

in getting great people who have a great

496

:

Uh, skill set on the team and having that

domain knowledge is kind of secondary

497

:

because people can learn a new domain.

498

:

It's having the things that are

most important is, is being curious.

499

:

It is having a right breadth of

experience to be able to handle

500

:

new problems because when.

501

:

When you do have to reinvent every season

to be able to maintain that advantage,

502

:

you need someone on your team who is okay

with having to scrap the old and start

503

:

over and, and not know exactly what it's

going to look like when you're done,

504

:

because there is no rinse repeat here.

505

:

It is always pushing the frontier.

506

:

So that's what has been the thing

that's been most important for us.

507

:

A group of people who are a curious

problem solvers that are okay

508

:

working in ambiguity, uh, rather

than people who know the ins and

509

:

outs of sports, uh, or betting.

510

:

Jazmin Furtado: Yeah, I can, I could, I

could see how, how many new things you

511

:

can pull and learn while you're actually

just like building the model in of itself.

512

:

Things that you insights that

maybe even didn't know before that

513

:

you found out the model showed.

514

:

I think it's like, both ways.

515

:

You need to, you learn and you

provide input and it's just like,

516

:

there's like, positive feedback loop.

517

:

Talking about feedback loops,

actually, I'm curious how in your.

518

:

Your experiences to date.

519

:

Being at rhythm or being in sports,

what are the skills that you've

520

:

had to pull on the most in this

field at this point in your career?

521

:

.

Brian Beachkofski: Yeah, great question.

522

:

So I would say the, the, the ones that

matter the most right now have been.

523

:

Kind of that team building and problem

solving because like any startup,

524

:

uh, you're, you're solving problems

more than you are implementing the

525

:

pristine plan you had ahead of time.

526

:

So having, uh, having an

ability to, um, understand.

527

:

What's most important and stay

focused on the thing that matters

528

:

most is critically important.

529

:

Um, and bringing the team together

so that you can have, uh, diverse.

530

:

Ways of solving problems,

but still be a cohesive team.

531

:

That's one that like, uh, cause there

are plenty of stories out there.

532

:

People who have people with

fundamentally different views on

533

:

how to solve a problem that can't

work together because they can't.

534

:

See themselves on the same team.

535

:

So that teamwork and

problem solving number one.

536

:

The second is having that,

um, technical curiosity.

537

:

So like you said, the exploring and

finding insights that you didn't

538

:

know, uh, when you're building the

models, there's, you know, two phases.

539

:

The first is exploratory where you're just

in a Python notebook and you're looking at

540

:

the data and trying to figure out, okay,

is this predictive is that near exploring

541

:

and making combinations and being able

to find those insights and just being.

542

:

Continuously curious about what's there

and then saying, why would this happen?

543

:

Is it spurious?

544

:

Is it just a random correlation or

is there truly an insight there?

545

:

And then being able to talk through

that without it becoming, I'm always in

546

:

exploration and never actually moving

it to prod because that you can run

547

:

into that instance as well, where, and

there's a lot of technical people who.

548

:

Always say, but I could make this better

and this better and this better, and

549

:

then it never actually gets the prod.

550

:

Right?

551

:

So having that balance is another one.

552

:

That's really important of saying

yes, but right now we're good enough

553

:

and there can be another iteration.

554

:

Jazmin Furtado: Yeah,

it's it's I think that.

555

:

Is always a tough bridge to cross

when you get people that are.

556

:

Just they're used to maybe

building something in a lab.

557

:

Or does it just continuing

on the research?

558

:

Oh, I just want to make it a

little bit more better a little bit

559

:

better when it comes to actually

points thing out into production.

560

:

You need to.

561

:

Find something that's going to, you

have all these other things to consider

562

:

in terms of like the timeliness and

the how much, how much overhead that

563

:

provides that gives you and you're

creating and you're deploying the app.

564

:

There's there's not just

the algorithm itself.

565

:

The algorithm performance

itself is not the only factor.

566

:

So,

567

:

Brian Beachkofski: uh, it's

not providing user value if

568

:

it's not in the user's hands.

569

:

Jazmin Furtado: Right, right.

570

:

Uh, the last question I wanted

to post to you was, um, talking

571

:

about the future of your field.

572

:

Um, where do you see sports betting going?

573

:

Obviously, you're talking about,

you know, the just such a fast

574

:

pace of iteration and advancement.

575

:

Where do you see it heading

in the next, you know, in the

576

:

near future or more long term?

577

:

And then what advice do you have for

folks that are interested in this field?

578

:

And what advice do you

have to get them into it?

579

:

Brian Beachkofski: Yeah, so I

think we're at a really fun and

580

:

interesting part in this space, right?

581

:

We, we.

582

:

We're betting the same way we

have for 150 years at least.

583

:

I think I said earlier in eternity, but

in Casey at the bat, the poem, they say,

584

:

I put up even money with Casey at the bat.

585

:

So somebody is, is, uh, wants to

live bet that that guy's going

586

:

to get a hit at a baseball game.

587

:

Like this could be what my buddies

are doing this weekend is, is

588

:

betting on a hunch that this guy's

going to get a hit this weekend.

589

:

Right?

590

:

So things are going to change.

591

:

Uh, and how do you give people

the tools to make that happen?

592

:

And, and.

593

:

The space, the, the, the, the sports

betting space has been relatively

594

:

resistant to real fundamental change,

like it changes with now it's on an

595

:

app, but, or we use more parlays or

we, it changes on the margins, but

596

:

at its core, it's largely the same.

597

:

And I think we're going to

see that continue to evolve.

598

:

So there, there's going to be more,

uh, more diversity in what people can

599

:

play on, but I think more importantly.

600

:

Like I said before, no one chooses

where to have dinner now without

601

:

consulting reviews and, and no

one picks out what, uh, what thing

602

:

they're going to buy online without

seeing the reviews and the ratings.

603

:

We're going to see more of that in sports

betting, where you use tools to make

604

:

picks because it just seems backwards

to go with your gut or to just use what

605

:

somebody flashed up in front of you.

606

:

And in what we've seen are there two big.

607

:

Different groups, so 1 is the

person who is is just want some

608

:

support the equivalent of of of a

rating and say, I trust this thing.

609

:

I'm going to go do it.

610

:

So that's 1 group.

611

:

And they're going to they're

they're want to model.

612

:

They want some sort of reason for doing

it, but they don't want to go deep.

613

:

But there are a lot of people who

really want to go that extra level.

614

:

And that's what's going to be most

exciting for where we're going is having.

615

:

This premium experience.

616

:

That's a no code modeling.

617

:

So rather than us do all the feature

engineering, let people go in and

618

:

see here are the base level stats.

619

:

How can I group them together in a way

that makes sense to me, make my own

620

:

features, get get feedback on how, how,

uh, effective they are in and build

621

:

their own models, change different,

uh Uh, modeling approaches switch from

622

:

a linear model to a tree based model.

623

:

And maybe some people, like I said,

don't want to go into that depth, but

624

:

there's a lot of people who are very

interested in doing that, but just don't

625

:

have the time or training to do it.

626

:

And if we can reduce that barrier,

I think it's going to be amazing.

627

:

The number of people that will have

the tools and be able to make their

628

:

own insights and be more engaged, not

only with sports, but within analytics

629

:

itself and get excited about it.

630

:

The potential like Saber metrics is

huge with baseball analytics and people

631

:

doing that and imagine if that same

kind of engagement can happen, but with

632

:

with without having to have your own

coding solution there, you can just.

633

:

Go in and pick what you

want and get those results.

634

:

So I think the future is amazing.

635

:

I think we're at a point where

everyone will be using some sort of,

636

:

uh, uh, analytical tool to help them

get engaged in sports betting and in

637

:

the amount of engagements people are

going to have with sports betting in

638

:

general is going to grow because it's

such a great fan engagement tool for

639

:

the leagues and the teams themselves.

640

:

So how would someone get into this?

641

:

I, I would do it just kind of like I did

is the great thing about sports data is

642

:

there's so much out there that's free.

643

:

Um, you can go to baseball reference

dot com and scrape their data.

644

:

And I think most sports now have a

football reference, soccer reference,

645

:

like the basketball reference.

646

:

They're all out there and you can scrape

the data and, and just like how that's

647

:

where I started because it was easier to

get sports data than it was engine data.

648

:

Um, I was, I was just talking to

a recent grad today who was asking

649

:

for, for mentorship advice and.

650

:

And that's what I said to

him is just go find problems.

651

:

If you listen to sports talk radio,

there's 50 great research topics said

652

:

every hour about like, Oh, what's

the value of this, uh, draft pick?

653

:

Oh, this guy's going to be injured.

654

:

What's that going to cost us

in terms of playoff position?

655

:

People throw out these

questions and then move on.

656

:

But if you're an analyst, you're

like, man, that's awesome.

657

:

I should go figure that out.

658

:

And in all the data you

need is publicly available.

659

:

So that's what I would do.

660

:

I'd spend a half hour listening to

sports talk, hear the, hear the guys

661

:

spout off questions they never really

want to answer, but then go in and

662

:

get the actual analytical solution to

those questions and, and, uh, put it to

663

:

Jazmin Furtado: practice.

664

:

Yeah, it's such a low barrier to entry,

uh, to just start getting interested

665

:

and start tinkering in the space.

666

:

And I think that, uh, is a breath of

fresh air for people that have trouble

667

:

just trying to get in, get information.

668

:

What you were talking about earlier

with, uh, people being interested

669

:

in the space and the future of

this field kind of just exploding.

670

:

It's just so interesting that

you're looking at the user.

671

:

In the future, you know, people

you're expecting for people to

672

:

be more digitally competent.

673

:

You're expecting for people to expect.

674

:

Certain features, be able to dive a

little bit deeper under the hood of

675

:

these algorithms, because people are

just have more of that knowledge in

676

:

this digital age that they're in.

677

:

So it's.

678

:

Great have to look forward, looking

not just at the industry, but also

679

:

the users who's going to be like,

what do they expect from your tools?

680

:

And that really brings

back at the end of the day.

681

:

It's the user.

682

:

It's what.

683

:

You know, it's you don't have a product

if it's not valuable to them, you can

684

:

have the coolest thing under the hood.

685

:

But at the end of the day,

they want to be able to.

686

:

People interact with it and use

it as useful for them where their

687

:

skill levels at and everything

needs to be centered around there.

688

:

So I think it.

689

:

That is just very exciting.

690

:

It's so fast paced.

691

:

That is also very different.

692

:

It's like a great arena where you have

technology that you're able to rapidly

693

:

iterate on, but then also rapidly show to

users just rapidly iterate the whole loop.

694

:

So that's just such a cool,

cool space that you're in.

695

:

I love it.

696

:

Thank you for doing that.

697

:

No, thank you for having me.

698

:

It was great.

699

:

So, for the last part of this

episode, I wanted to close with this

700

:

game I'm calling Fact or Fiction.

701

:

So, I have a few statements, uh,

here about sports, and I'm not

702

:

expecting you to know the answers.

703

:

I was hoping to make it so hard so

that you wouldn't know the answers,

704

:

uh, and I want you to tell me if you

think that they are fact or fiction.

705

:

So, are you ready?

706

:

I am ready.

707

:

All right, so the first one.

708

:

There are over 8, 000 sports

played around the world.

709

:

That's not the, that's not the,

that's not the one in question.

710

:

But about 4 of them, 4% of them

are represented in the Olympics.

711

:

Is that fact or fiction?

712

:

Brian Beachkofski: Uh, well, 4% of 8, 000?

713

:

Is, uh, what?

714

:

320.

715

:

So I'm going to say that seems fact.

716

:

Jazmin Furtado: That is fiction.

717

:

There, there are 33 sports

played at the Olympics.

718

:In:

719

:

Brian Beachkofski: Okay.

720

:

Yeah.

721

:

Was leading events, not sports.

722

:

Jazmin Furtado: So less than 0.4% of

sports are represented in the Olympics.

723

:

Yes.

724

:

Brian Beachkofski: That makes sense.

725

:

Yeah.

726

:

I was thinking like, uh, like track,

uh, like running would be one.

727

:

Not like the 100, 200, 400, 800.

728

:

Yeah.

729

:

Yeah.

730

:

Jazmin Furtado: Okay.

731

:

I need to find some machine to

recalculate this for advance.

732

:

. . Alright.

733

:

A second.

734

:

About 20% of people in the U.

735

:

S.

736

:

participate in sports on a daily

basis, or like, on any given day.

737

:

Brian Beachkofski: Oh,

that's, that's too high.

738

:

I'm saying false.

739

:

Jazmin Furtado: That is fact!

740

:

Sports, as in sports,

exercise, and recreation.

741

:

So, based off the Bureau

of Labor Statistics.

742

:

So, about 20%.

743

:

Doesn't mean that 20% of

people exercise daily.

744

:

It's on any given day,

about 20% of folks do.

745

:

All right, you're 0 for 2.

746

:

Or 2 for 2, it depends.

747

:

Or 2 for 2.

748

:

Number 3.

749

:

Basketball has the highest estimated

rate of sports related injuries among

750

:

individuals above the age of 25.

751

:

Brian Beachkofski: I would say true.

752

:

I've seen so many angles roll.

753

:

I'm going to say true.

754

:

That

755

:

Jazmin Furtado: is...

756

:

False.

757

:

I'm just, I'm like, so proud of myself.

758

:

Um, cycling, bicycling has the highest

rate of sports related injuries.

759

:

Then basketball is next.

760

:

Brian Beachkofski:

Basketball's number two.

761

:

It's true.

762

:

I, uh, I had a cycling injury myself.

763

:

I was going downhill, hit a pothole,

flew over my handlebars, broke

764

:

my helmet on, uh, right, right

next to, uh, a telephone pole.

765

:

That was, uh, it was a bad day.

766

:

That

767

:

Jazmin Furtado: was, was that in Boston?

768

:

It was in DC.

769

:

Oh, in DC.

770

:

Oh, that's interesting.

771

:

Oh, I'm glad you're okay.

772

:

No, those cycling accidents

are, they're, they're real.

773

:

And, um, apparently the biggest

sports related injury is in the knees.

774

:

So like knees are like

the most popular, I guess.

775

:

Most most injuries are in the knees.

776

:

All right.

777

:

4th 1 in the U.

778

:

S.

779

:

crime rates can drop during

sporting events, but up to 25%.

780

:

But that's not the fact in question.

781

:

However, in the capital of the

Philippines, when boxer, many fights,

782

:

the crime rate can drop up to 100%.

783

:

Or reach 0%.

784

:

I say true.

785

:

Yes, that is true.

786

:

Benny Pacquiao is a hero

to many in the Philippines.

787

:

And so, I found that very interesting.

788

:

I was like, really?

789

:

Crime rate was zero.

790

:

At least reported crime rates were zero.

791

:

For the whole 7 hours that he was playing.

792

:

Which is interesting because that's...

793

:

In the capital, which is so dense, that's

so many people, but yeah, yeah, the

794

:

correlation to sporting games and crime,

everyone should just go see more sporting

795

:

events for the betterment of humanity.

796

:

All right, so you got 1.

797

:

All right, so last 1, let's

see if we make that 2.

798

:

MMA is the fastest growing sport in

America, as it has consistently grown

799

:

about 20% annually for the past 3 years.

800

:

Brian Beachkofski: Is this in viewership?

801

:

Jazmin Furtado: This is in

just like the Or participation.

802

:

The dollar value of the industry.

803

:

It's not participation,

it's just like of the whole

804

:

Brian Beachkofski: industry.

805

:

Okay.

806

:

Then I'll say yes.

807

:

Jazmin Furtado: That is fiction.

808

:

Pickleball.

809

:

Oh, that's right.

810

:

Brian Beachkofski: I forgot pickleball.

811

:

Of course it is.

812

:

Of course it's pickleball.

813

:

Jazmin Furtado: Pickleball is the

fastest growing sport in America.

814

:

I have done, my parents introduced

me to pickleball, and it's fun!

815

:

It's addicting.

816

:

I don't know if you've tried it yourself.

817

:

I haven't, I haven't tried it.

818

:

You need to get you may get on a court.

819

:

There's like, I think

10, 000 courts in the U.

820

:

S.

821

:

I know all these things now because

I've been on the depths of the

822

:

Internet for the past a couple of days.

823

:

Brian Beachkofski: Well, maybe we'll get

coverage and rhythm here pretty soon.

824

:

Jazmin Furtado: Yeah,

you should look into it.

825

:

I don't know what the stats are

behind it or like how much data

826

:

is being captured on it yet.

827

:

But, you know, I'm sure

it's only going to grow.

828

:

Yeah.

829

:

Thank you for playing and

for entertaining that game.

830

:

I know it's like really tough,

but hopefully you're able to

831

:

learn a new a couple new facts.

832

:

Oh, of course.

833

:

Thank you as well for being my

guest and a fellow data explorer

834

:

in this world of data we're in.

835

:

I know firsthand how

you've inspired people.

836

:

All around you and all the teams that

you've been a part of, and your focus

837

:

on people has been really reflected

in every step of your journey.

838

:

And I know that your story really

resonates with folks who feel like

839

:

there's such a breadth of opportunities

out there, but haven't seen someone or

840

:

seen too many people that have really

gone out and explored and ventured

841

:

all those opportunities in real life.

842

:

So.

843

:

Thank you for being that example.

844

:

Well,

845

:

Brian Beachkofski: thank you.

846

:

I know you've been a great

example to many people, too, and

847

:

congratulations on your post service

career and where that takes you.

848

:

It's going to be amazing.

849

:

Jazmin Furtado: Thank you so much.

850

:

That's the one thing I love about tech

is that you're just constantly learning.

851

:

So, uh, I'm excited just to

explore more of the world and

852

:

just be enveloped in it even more.

853

:

Uh, I also wanted to thank Hatch

IT for sponsoring this episode and

854

:

allowing me to host this series.

855

:

And as always, I'd like to thank you,

the listener for tuning into this episode

856

:

and exploring the world of data with us.

857

:

Thanks everyone.

858

:

Take care.

859

:

Tim Winkler: Calling all

startup technologists.

860

:

podcast, but don't know where to start?

861

:

Well, here's your chance to shine.

862

:

We're thrilled to introduce beyond

the program, our exclusive mini

863

:

series, and we want you to be a part

of it as tech leaders and mentors.

864

:

You'll get the exclusive opportunity to

become a guest host right here on the pair

865

:

program podcast, share your expertise,

insights, and stories with our audience

866

:

of startup focused technologist excited.

867

:

We knew you would be.

868

:

To be considered, head over to myhatchpad.

869

:

com backslash contribute.

870

:

Fill out a brief form

and submit it our way.

871

:

Let's co create something

amazing together.

872

:

Don't miss this chance to elevate your

voice and expand your personal brand.

873

:

Visit myhatchpad.

874

:

com backslash contribute.

LET’S DISCUSS YOUR HIRING NEEDS

Build a custom hiring solution to grow your product, data, and
engineering teams.