Inside CAA Intell: André Vargas on Fighting AI Hype with Artist-First Data

Orlando Wood (00:05)

Andre, Thank you so much for taking the time to chat with us, because

I’ve been really wanted to talk to you for, mean, I always enjoy

talking to you, but I’ve been really wanted to talk to you on the

podcast for quite some time.

Andre Vargas (00:16)

Yeah, well, thank you for having me. This is very And yes, we’ve been

talking for a long time and can’t be more excited than being here with

you today and see what we can talk about.

Orlando Wood (00:27)

just to kind of kick it off, mean, as far as I see it, you your role

as chief data officer has become arguably one of the most important

strategic roles in any modern company. mean, especially with the

advent of generative AI, you know, and generative AI is getting a lot

of the attention. I know you’ve been working in data for so long and

machine learning for so long.

But I think that especially a company that sits as a cultural

crossroads like CAA and interacts with so many different types of

creative data, can you just take me through what your role is and as

well how it’s been changing in the last couple of years?

Andre Vargas (00:59)

For sure, definitely interesting times.

I came from management consulting, actually building data platforms

for big corporations. So it was a little bit of a culture shock for me

to enter media and entertainment and see how data was managed here and

how analytics was perceived. And especially in the arts and

representing artists, the uniqueness on combining that intuition, the

understanding of art, the personal relationships, and then how to

insert data into that whole process.

as

the world around our industry was becoming more data-driven and

technology-driven.

given the entrance to the business of entertainment being tech company

first. So it’s definitely been a challenge, but also a buffet of

opportunity, right? So imagine a dream come true to come to a place to

build a data competency and capability from the ground up in a very

difficult, non-traditionally data-driven environment. It’s been

nothing but a joy, to be honest, a challenge and a

joy and maybe really important for our conversation I think is

separating what generative AI is, which is kind of like overtaking the

word AI. Generative AI is a methodology inside the overarching machine

learning AI ecosystem, which we’ve been using for a really long time.

And if you allow me to tell you like a funny story, it’s…joy and maybe really important for our conversation I think is

separating what generative AI is, which is kind of like overtaking the

word AI. Generative AI is a methodology inside the overarching machine

learning AI ecosystem, which we’ve been using for a really long time.

And if you allow me to tell you like a funny story, it’s…

arriving at CAA, I had to quickly learn that talking about algorithms

and statistics and math and regression and dashboards, not really

catching fire on my early days of trying to bring data. It’s like,

what are you gonna tell me that I don’t already know, Mr. Data Guy?

Orlando Wood (02:41)

Just a lot of glazed over looks.

Yeah.

Andre Vargas (02:48)

But then what I realized is like, OK, I need to change my language. I

understand I’m not going to talk about machine learning. I’m not going

to talk about artificial intelligence or algorithms. I’m going to talk

about how do we supercharge how we serve our clients? How does the

company make money? How do we optimize deals for clients? How do we

discover the best talent? Rather than talking about the how, I just

wanted to understand what’s the workflow?

you know, how does this company serve its clients? So I never called

AI AI. In fact, I came up with this thing called augmented intuition

so that I would be able to be welcomed to these conversations about

how does data insert itself in that magnificent intuition the agents

have and the artists have. They have an intuition that you can’t

quantify. So I called AI augmented intuition.

was very careful making sure that I would be welcomed in the culture,

trying to introduce data. And then Chat GPT comes out and generative

AI is everywhere and everybody’s calling it AI. So I get a call from

the CEO of the company, literally, Andre, what are we doing with AI?

And I’m like, my God, from one day AI went from blasphemy to the word

of du jour. So that’s what happens, know, always looking for…

Orlando Wood (03:56)

Mm-hmm.

Yeah.

Andre Vargas (04:05)

better and improved ways to service our clients. And basically, being

the chief data officer, I’m in the nexus of what’s going on.

supporting not only the sales side, but supporting our business

development people, our corporate strategy. Again, it’s a buffet of

opportunity. So really exciting for sure.

Orlando Wood (04:23)

That sounds like a very Hollywood story, doesn’t it? So you’ve been,

you’ve been banging on about something for a long time. Then suddenly

somebody hears it in the, the, the halls of power and you’re getting

called, do you know about this AI stuff? Yeah. it, that’s interesting

that you talk about, know, how you had to language it and, know, being

very astute of you to, sort of understand that at that stage, at least

before the generative AI conversation came out, that you had to.That sounds like a very Hollywood story, doesn’t it? So you’ve been,

you’ve been banging on about something for a long time. Then suddenly

somebody hears it in the, the, the halls of power and you’re getting

called, do you know about this AI stuff? Yeah. it, that’s interesting

that you talk about, know, how you had to language it and, know, being

very astute of you to, sort of understand that at that stage, at least

before the generative AI conversation came out, that you had to.

Andre Vargas (04:36)

That’s right. That’s right.

Orlando Wood (04:50)

speak in terms of outcomes rather than speaking in terms of the

technology being used. And that’s one of the things I really kind of

have noticed about you especially is that you’re always talking about

what it means to the artists and what you’re doing on the data end,

being really fundamentally connected to the artists and the job of the

agency, which is to represent artists. I feel that a lot of agencies,

pay lip service to it, but then when the rubber meets the road, the

people who are really doing that data-centric or technology-centric

work sort of start to forget what the overall purpose is. Does that

come top down? Is it something that just you really got into your

bones quite quickly?

Andre Vargas (05:32)

Yes. So it is part of the DNA of the company. We service clients, like

that’s period, right? So that is the, it’s in the DNA of the company

as a services organization. And I’ll tell you another story that hit

me really hard, which is what I’ve used as a North Star, right? So

it’s a combination of two things, like where I came from.

I always loved Clayton Christensen’s jobs to be done theory. What’s

the job to be done? Let’s start with the why and the job that needs to

be done versus, hey, here’s a tool, where can we use it? Right? So I

go the other way around. But what really hit me a few years ago before

the pandemic in a company retreat, one of our A-list superstar clients

was being interviewed as a guest, right?

And I was in awe. We know celebrities by the roles they play on

television or movies or when they’re playing sports or in music. But

then when you see them talk, you know, honestly about their dreams and

aspirations and how they build businesses, I was like completely

mesmerized by this particular client. But then at the end of this

interview, she she was asked.

Hey, so you have the whole company in front of you, right? What is one

thing you would like from us? And she said, I want to feel like I’m

your only client. And being a nerd that I am, I decided, well, I work

for these clients, like period. Everything I do and everyone that I

interact with, we all work for these clients. So whatever decisions we

make, what would she think? Does this benefit her?Hey, so you have the whole company in front of you, right? What is one

thing you would like from us? And she said, I want to feel like I’m

your only client. And being a nerd that I am, I decided, well, I work

for these clients, like period. Everything I do and everyone that I

interact with, we all work for these clients. So whatever decisions we

make, what would she think? Does this benefit her?

And that helps me combine the understanding that ultimately everything

we do is to supercharge the value of a client, the opportunities that

they can have with the job to be done. And that combination literally

helps drive how we design our systems that supports them, the agents,

because it all rolls up ultimately in the service of a client. So

that’s really what helped me.

and the team that we’ve built here, we all say the same thing. In

fact, a colleague of mine even gifted me a cardboard version of the

client that I work for because I started to tell this story all around

and everybody knows who Andre works for. Now this is the idea. That’s

right. It’s the boss.

Orlando Wood (07:42)

Yes, she watches down from you, doesn’t she? ⁓ Always in your office,

know, looking

over your work.

Andre Vargas (07:49)

I have a little sticker, it says the boss and in Spanish it says la

jefa. I’m a member of Ciel Latino. So definitely everything is there.

as a matter of fact, this client has this story. There has been

pictures and selfies sent to her via her agent. Finds it hilarious and

wonderful that that’s how we serve. So yeah, I think.

Once everybody centers around the job to be done and who do we

actually work for ultimately, it bypasses politics, it bypasses

everything else because at the end of the day, who do we work for and

what are we trying to accomplish is the North Star.

Orlando Wood (08:21)

Well, I think it also humanizes it as well, doesn’t it? that reminder

that these are people who have worked very hard for their careers and

there are people there. These aren’t just massive companies or

whatever else.

One of things I really appreciate about you is that, you know, very

often if we were talking about a piece of technology, you would very

often reorient me, be like, Orlando, remember, there are people here

and these are artists here. And how would you, you know, and with my

mother, she was a novelist and her books were all, all of her books,

she wrote 20 romance novels and they were all in the books, three

corpus. And so when you kind of, that orienting thought of, you know,

what would an artist say about this?

I really respect what came out of that is CAA’s overall and your

personal approach to the provenance of data that these language models

and AI models have been trained on. You you guys really walk the walk

when it comes to not engaging with models that have the original sin

of copyright theft at their heart.I really respect what came out of that is CAA’s overall and your

personal approach to the provenance of data that these language models

and AI models have been trained on. You you guys really walk the walk

when it comes to not engaging with models that have the original sin

of copyright theft at their heart.

It’s these artists that you represent, It’s their work that has been

stolen to make these models. for me, I think that especially the

language models, it would have been very easy to train a reasonable

model on all of the publicly available legal data and courthouse data

and all of the language data that was open to the market. But I think

that what these models and the people who train them realized was,

The thing that’s gonna make people want to use it on a daily basis is

art and poetry and filmmaking and television and literature. And that

it’s actually the art created by human artists that’s gonna give it

that secret sauce that’s gonna make people want to interact with them.

And so, I guess it’s a two-part question. One is, how do you retain…

that standard when this technology is moving so fast? And well, I

guess we’ll start with that question.

Andre Vargas (10:15)

Well, so remember again, who do we work for? Right? So great that you

framed it that way. And then CAA values its integrity, right? Our

North Star in servicing clients means that we have to be ethical in

everything we do, right? And we’ve also been very openly about

defending the rights of our clients.

Right? So I think the solution to this should have been just easily

compensation, credit, and consent. The fact that they just took it, I

think that is the difference. Right? And there is like, I think, over

40 lawsuits going on at the moment around copyright and the defense of

fair use versus non-fair use and so on. Then the next part of this

answer, I would say, there is a lot of hype.

around AI, which causes this, you’ll fall behind. There is this FOMO

frenzy going on. And sometimes you need to also be courageous to be a

little bit, okay, well, can we first study this thing and see what is

it versus just jumping in the pool? You should just, know, what is it

really? And then obviously looking at the data sets, the pile and the

books three and all of those things, obviously it’s a red flag and an

alarm.

in that you cannot, you know, at the same time, you know, publicly be

on the side of artists, be ethical, and then also utilize some of

these LLMs, right? What CNA has done, and you know, you can look this

up, you know, there are efforts that we are doing with YouTube, you

know, in terms of identifying fakes and helping our clients have those

removed, for instance, once identified.

And the vault, if you look at CAA’s the vault, that is also a way for

clients to digitize their image and likeness, voice and so forth, so

that those things can be legally licensed and monetized, right? So I

think LLMs per se are unfortunately taking over the word AI when

they’re actually language models.And the vault, if you look at CAA’s the vault, that is also a way for

clients to digitize their image and likeness, voice and so forth, so

that those things can be legally licensed and monetized, right? So I

think LLMs per se are unfortunately taking over the word AI when

they’re actually language models.

But there many things that you can do about it, right? So I think

ultimately protecting the rights of clients are the right thing to do.

It is a very complex world that we’re living in because you can’t undo

the stealing. I think the tech companies, they have to, you know, they

should settle somehow, right? And pay artists or rebuild all of their

LLMs without including copyright material. So who knows what that’s

going to be like.

So that means that we have to be very, very careful when opportunities

arise to utilize this technology. Again, what’s the job to be done?

How does this benefit a client? Does it fit in the workflow? And are

there models, and there’s I think about four or five now that

publicize their training data sets and specifically differentiate

themselves by

We don’t use copyrighted materials and we respect robots.txt files and

you can maximize the ethics of how these models have been trained and

we’ll only utilize those. If that means that they are a little bit

smaller and I believe that smaller models work better anyways, but

let’s say if my model is not as good as the latest OpenAI model or

cloud model,

but it gets the job done. I’d rather do that. I’d rather have my data

science team work harder to make it work and maintain that integrity

and ethics than jumping on the latest, latest version, which by the

way, there is diminishing returns in every new version of these LLMs

anyways, as all the data has been already swallowed. at this point,

it’s problem solving more than anything else.

Orlando Wood (13:43)

Well, I think it’s also, you know, understanding that just because the

language is model is large doesn’t mean it’s the right tool for the

job. You know, that, you know, making sure that you understand that

there are now loads of tools out in the market and using the right

tool for the right job and as well putting it all through that filter

of, you know, copyright provenance and, and, you know, clean data, I

think is, you know, another option that’s available now that that

wasn’t necessarily a

Andre Vargas (13:52)

Exactly.

Orlando Wood (14:11)

you know, a year or two ago.Andre Vargas (14:12)

Yeah. And one of the advantages we have, as we remember how I had to

change the language and how do I embed data in the workflows and how

the agents work, that meant that we had to come up with this notion of

co-design, right? So we don’t just get an idea from an agent and then

go into our data lab and come up with a solution, right? So we work

really, really closely and together with the agents. So…

In addition to a smaller model potentially being better, what makes it

work way better than a bigger model is that feedback loop that we have

with the people who actually are going to use it, specific to how they

want to use it. So externally looking at it, it’ll be harder to match

how a CA agent does their work if you don’t have

hundreds of them providing you continuous feedback on how to improve

those business processes. So there is a competitive advantage in

actually understanding the use case and the job to be done and then

apply the right tool cell to that. So it’s that way versus let’s, you

you can put an LLM on top of a broken process. It will just amplify

the problems of that broken process. So you kind of have to fix the

process.

understand the job to be done and then apply the right tool for it.

Not everything is LLM. That’s the issue. It’s the frenzy about LLMs is

that, we can use AI and, let’s use AI. Okay, let’s answer that

question first and see what’s the right tool.

Orlando Wood (15:38)

Yeah. And you talked about not succumbing to the fear as well. think

that, you know, what, what’s been so interesting about you guys is

you’ve had the bravery to be a bit slow when it comes to making the

right decisions about generative AI. But a lot of that bravery as well

comes because you’ve been doing it with machine learning for so long,

like CAA Intel, which, you know, I’ll ask you to explain it to

everybody in a bit, but

one time you explained it to me and you said, you know, Apple, Amazon,

Netflix, these are tech forward companies that we’ve been negotiating

with for quite some time. If you think that they’re not showing up to

a negotiation with the receipts and the data that they’ve prepared,

we’ve had to combat and outfit our agents with that data.

for quite some time. So yeah, can you talk to us about CAA Intel and

what you’ve been doing and how robust that is from a machine learning

perspective?

Andre Vargas (16:31)

Totally. So I’m an economist. So I used to work on operations

research, statistics, game theory. And I’ve never really worked as an

economist, but I’ve been able to use a lot of those ways of thinking

to kind of build the solution that we’ve done over here. And the main

problem that I saw that we needed to solve was this information

asymmetry.Totally. So I’m an economist. So I used to work on operations

research, statistics, game theory. And I’ve never really worked as an

economist, but I’ve been able to use a lot of those ways of thinking

to kind of build the solution that we’ve done over here. And the main

problem that I saw that we needed to solve was this information

asymmetry.

we call a buyer would be somebody who is looking for the services of

one of our clients, right? So if a buyer knows more about the project,

the market, the popularity and audiences than our agents do, they are

at a disadvantage, right? So, you know, when you’re negotiating or

like if you and I were playing poker and I know your cards and you

don’t know my cards, I’m sure I’m going to bet in a way that I’m going

to win, right? Because I know more than you. So

There is no perfect information because as you know a lot of these

companies do not really share all of their data. So the beginnings of

Intel for us, we first called it how do you measure a black hole,

which is by the light and movement around it.

And the idea was there is no single data source that is going to

answer the questions the company has or how do we reduce this

information asymmetry gap. So the idea was like, but wait, if we start

combining data sources, unifying data sources from many, different

angles, that is going to create that light and movement around these

black holes, right?

And then we call that because obviously again with the language,

right? So let’s not call this measuring a black hole. Let’s call it

data for selling. You know, how do we provide data to the agency that

helps us sell projects, identify opportunities for our clients. And

that was kind of like the entry point. So what we’ve done differently,

and I think, you know, this is obviously the way we thought coming

from the outside, we didn’t come from the traditional

media and entertainment industry, right? So what I’ve noticed, there’s

a lot of focus on research inside media companies. Me and my team,

came in.

with the background of building data capabilities for very large

corporations. And this was all about building data platforms, multi-

dimensional cubes, and using machine learning to answer business

questions. So once we understood what the business questions were, we

realized.

My goodness, we’ve never seen a more eclectic business than CAA,

right? So if you work in a car company, well, then you’re gonna do

manufacturing and parts and supply chain, and it becomes very much

that, but.

Orlando Wood (18:52)

Yeah.Andre Vargas (19:00)

popularity and it’s in movies and TV and digital and then there’s

music touring and then there’s like all these brand endorsements and

then there’s like the variety of data is tremendous. So I keep telling

people, know, I’m gonna die of old age before we’re done doing half of

what the company needs to do. But the strategy of Intel was we

actually need

a unified space where we can have an audience 360 that is unified with

a talent 360. And this is all data around them and a buyer 360 and a

project 360. So it’s almost creating a multiverse of 360s that are

unified so that I can cover almost like a cube that you can go back

and forth. So I know, you know, I have data around audiences, but also

what they, you know, what’s their purchase behaviors are.

Orlando Wood (19:37)

Right. Yeah.

Mm-hmm.

Andre Vargas (19:49)

their consumptions that they have, what do they view, what are their

brand affinities, what are their demographics. There’s all kinds of

data that combine together, which is what is the hard part.

and all using data that everybody else has access to. The difference

is that we’ve decided to ingest all of it and unified in our data

platform, which is really what makes it very unique in that our data

insights teams have access to data just at their fingertips. We also

open it up to the whole agency.

Orlando Wood (20:09)

Mm-hmm.

Andre Vargas (20:20)

So any agent, assistant, coordinator has access to our Intel platform

in case they want to look up things very quickly. So it’s definitely a

supercharger in that regard because the data is at everybody’s

fingertips and available to all.

Orlando Wood (20:35)

And I, it’s interesting, just to kind of go into, because that was

such an interesting thing you said about, you know, measuring the

light as it refracts around a black hole can give you information

about the mass at the center of the black hole. You know, similarly, I

imagine that, you know, so if, if an artist has a 2 % gross profit

participation on the back end of a film and you’re not getting the

clear

box office data or reporting data from the studio on the film, you can

generally surmise what the overall box office was, you know, or what

it meant to them in terms of their profitability on the film. If it’s

much higher than was anticipated, you then have the ability to

negotiate more strongly for their role in the sequel. They can’t come

back and go, well, that movie barely made any money. We’re doing you a

favor by even calling you, you know, if your artist is going to pass,

we are not that bothered And thenbox office data or reporting data from the studio on the film, you can

generally surmise what the overall box office was, you know, or what

it meant to them in terms of their profitability on the film. If it’s

much higher than was anticipated, you then have the ability to

negotiate more strongly for their role in the sequel. They can’t come

back and go, well, that movie barely made any money. We’re doing you a

favor by even calling you, you know, if your artist is going to pass,

we are not that bothered. And then

You guys can presumably turn around and say, actually we know this,

this, and this. These three metrics allow us to, you know, game out

how much you guys made from this movie. And I think you’re more

interested in making the sequel than you’re letting on.

Andre Vargas (21:38)

Yeah, and what’s interesting is sometimes we get feedback from our

agents that says, they dismissed the data. They didn’t care about the

data, but they renewed for season four. Sometimes it’s knowing that we

also know, right? And giving the enough information.

to an agent that’s already an expert negotiator to have that

confidence to know, actually, the size of this black hole is 3.2 times

the size of this other black hole who made this much, right? So in the

world that we live in with not having complete transparency about

performance of shows,

we feel that providing these insights in multiple angles, right?

That’s kind of like the light around it, enriches the way our agency

then negotiates, right? Or identifies these opportunities. So it is

literally, we might never close the information asymmetry gap entirely

because we don’t have first party data from each of the companies that

we do business with, but we’re closing it, right?

blindness at all like our agents go in armed you know with ammunition

in terms of you know I know the popularity of my clients right I know

the relative value they’re providing to this particular show we know

the audiences that it’s bringing you like

If you go to Amazon, we might not go in and say, you’re going to get

more subscribers to Amazon Prime, but we’re going to bring an audience

that’s going to buy more shoes. So what does the buyer really want

from a particular show? What are the audiences they’re trying to

reach? And we’ve been really, really good at that and very successful.

Also demonstrating the value of clients helps our agents negotiate

bigger, better deals. Also parity.

of compensation when there’s been disparity of compensation for

particular clients.

Knowing that just energizes our team so much because we know the

impact it’s doing because it ends up with our clients, right? So they

get the deal. They get their dream of renewing a season gets renewed.

And we celebrate it as we have been in the negotiating room because we

know we participated. we feel very connected in the work that we do in

the results that the agents get when they’re doing, you know, their

business on behalf of our clients. It’s really terrific.Knowing that just energizes our team so much because we know the

impact it’s doing because it ends up with our clients, right? So they

get the deal. They get their dream of renewing a season gets renewed.

And we celebrate it as we have been in the negotiating room because we

know we participated. we feel very connected in the work that we do in

the results that the agents get when they’re doing, you know, their

business on behalf of our clients. It’s really terrific.

Orlando Wood (23:48)

Yeah, that’s phenomenal. when you sense check your, your data and the

way that you’re using it, do you come at it, sort of user or outcome

first, or do you look at new pockets of data and go, we, we now have

access to this new sports data or this new repository of information

that we’re actually getting through our sports division. And we need

to.

Bring it in close, examine it from all sides and figure out what the

opportunities are coming out of it. Or are you more engaged with your

sports agents, for example, and going, what do you need? What can we

find for you? we think we have this data. know, what are you, how much

of it is user first? How much of it is data first?

Andre Vargas (24:26)

It is always user first. What is happening is the following. So

imagine that, hypothetically speaking, the maturity level of the

agency in terms of utilizing data and selling with data was, let’s

say, level one, which means that you can only ask level one questions.

So in the beginning, as the data team, we kind of know.

a little bit more so we can say, here’s the art of the possible. You

could do this, you could do this, you could do this, you could do

this. We prototype some ideas in how we understand the business. So

with a starting point, that code assigned feedback loop begins because

the agents are elite in what they do. So the moment they see,

But could you do this? So I’ll give you one example from our touring

department because one of the early prototypes we built was for the

touring group, specifically our comedy department, with a good friend

of mine who’s the head of the comedy department, Matt Blake. This is

actually been, he’s written about this too, which is hilarious how he

explains it versus how I remember it.

But we were introduced when I was trying to understand how the touring

group worked, because I had this idea that

this is one place that I can visually show them something that they’ve

never seen before. It was like a great opportunity. So I was

introduced to Matt and this is my goodness, like over 10 years ago. So

this is really embarrassing that it’s so long ago, but he was forward

thinking in terms of optimizing tours based on what he called recency,

right? And how, how recent one of his act had played in a particular

market to not return to that market too soon.to maximize ticket sales. And he was using a map with some red pins

and yellow pins and green pins. And I came to him and I said, hey,

explain what are you doing? He did explain exactly what he was trying

to do. So again, user first, right? But he didn’t know what the art of

the possible was. So I said, hey, I can automate that. And then he

said, no, you can’t. And I said, yes, I can. And he said, no, you

can’t. And I said, yes, I can. And then he said, well, knock yourself

out.

Orlando Wood (26:16)

Mm-hmm, yeah.

Andre Vargas (26:27)

So I came back literally like two days after that with a interactive

map that was doing exactly what he was doing without the red, green,

yellow dots. And he said, well, let me explain to you how the red,

green, yellow dots work and why. And I’m like, OK, you don’t want to

overfish the pond. So let’s find out what’s, is it 12 months, 18

months, 36 months? What’s the cutoffs? So I came back and all my map,

heat map of dots around

Play Dates came back red, green, And then he says, that’s really

great. But can we do this by Venue Type OK. So let’s add the Venue

Type slicer. So it just started creating a feedback loop. And if you

saw our Intel Live Touring system today, you’ll get a heart attack.

Like, we do radius clauses. It does the math to understand, you know,

Let’s not play within 50 miles of a market, but not further away than

300 miles so that the driving of the trucks don’t go beyond, know,

guild agreements and so forth. It does recommendations of venues based

on who you are and where you are in your career. So that’s machine

learning. know, large language model is not going to be able to

predict that. You need a different methodology for that. from, knock

yourself out.

Orlando Wood (27:21)

Mm-hmm.

Yeah.

Andre Vargas (27:38)

you can’t automate this to being providing machine learning venue

recommendations through multiple iterations. And then more agents get

added and more of them start asking new questions. So what’s happening

is the maturity of the agency grows on every iteration of Intel, which

means they ask better and bigger questions to us, which then forces us

to

Orlando Wood (27:38)Yeah. Yeah.

Andre Vargas (28:01)

At some point we can say, we have the data and if we don’t have it,

now we have to go search. Like who would have these kinds of data?

Right? So it is basically co-design. I would say it goes both ways. If

we learned about a new methodology or a new tool, we would go back to

the agents and say, Hey, if we could do this, how would you use it?

And then they would explain it to us or say, yeah, I don’t think maybe

the timing is wrong. Right? So we come to them with ideas, but

I would say it’s 80-20, it’s demand driven from the agents and how

they are trying to do their business. And the 20 % of us is trying to

expand that horizon a little bit. So new ideas come in. When I

interview these agents, I sometimes tell them, hey, if you had a magic

wand and there were no technological constraints, what would you like

to have from data? And sometimes they ask for things that we can do

right away.

they just don’t know what the limits of our competencies and

capabilities are. So sometimes it’s just expanding, breaking a little

bit of the potential understanding of our limits and just say, just

ask and we will tell you when we can’t, right? Versus I don’t think

you can do it, right? So it’s been, you know, we released Intel in

2018. If you saw Intel today versus what it was in 2018, you would

almost say that it’s a completely different app.

Orlando Wood (28:57)

Mm-hmm.

Yeah.

Andre Vargas (29:17)

But it’s that continuous evolution through feedback loops from every

department. Music does it one way, motion picture is a different way,

the brand groups a different way, the podcast group does it a

different way, the books group. So all of that feedback loops means

that kind of like this thing morphs and adjusts and adapts to the

different needs of the different departments.

Orlando Wood (29:17)

Yeah.

Andre Vargas (29:38)

quite specialized. And that’s the power of Intel is that it answers

questions the way the agents ask them.

Orlando Wood (29:44)

Yeah, I think that one of the things that I have found quite

beneficial about the generative AI sort of wave of interest is that

what we’re finding is a lot of people are coming to us asking for

solutions that don’t actually require AI, you know, but they didn’t

think to ask a little bit like you were saying about the agents, you

know, they just.Yeah, I think that one of the things that I have found quite

beneficial about the generative AI sort of wave of interest is that

what we’re finding is a lot of people are coming to us asking for

solutions that don’t actually require AI, you know, but they didn’t

think to ask a little bit like you were saying about the agents, you

know, they just.

the art of the possible they didn’t know to ask. And now that AI has

come out, it seems anything is possible. So they ask the question that

they’ve been sitting on for quite some time. And I find that that’s

actually, you know, of quite some benefit. And then on the other end,

what I find is that people are a little bit more aware of the term

MVP, you know, or POC, and it makes them comfortable going well. If it

does one thing really well, we can iterate and we can build, you know.

And as you say,

You know, 10 10 years ago you had that conversation. I can’t imagine

that you know your first iteration and where it is now even look like

they have the same DNA.

Andre Vargas (30:44)

Now that we go back a little bit on generative AI, it’s the same

approach, right? So the agency comes back, it’s like, can you do,

know, could this do this for us? Could it do this other thing for us?

And we do have a data lab, right? So our data lab gets to do

experiments, right? And one of our key values inside the data lab is

no-sum-cause bias, right?

So if something doesn’t work, we’re not going to double down and say,

we have to make it work. If it doesn’t work, we are very honest, open,

and transparent, and saying, it’s not working, not the right tool, or

not the right time. And let’s pick another of the priorities. So when

we experiment with generative AI, we’re also very honest on the

feedback. In fact, it’s really funny. I have this one thing when I

send.

our proof of concepts for testing to the agents, which this is what I

love about this agency. I go into a department head and I say, hey, I

would like a couple of volunteers to help us build a solution where

we’re going to be inserting generative AI. And then I get 15 agents. I

don’t get two, I get like 15, you know. And that is fantastic because

there is the more people that touch it,

Orlando Wood (31:46)

Right.

Andre Vargas (31:52)

the variety of feedback is just invaluable to us. And we’re very

honest. And when I send these guidelines on how to test, I literally

say the very first bullet point says the AI is guilty until proven

innocent. With that mindset, because everybody who works with LLMs in

detail knows that they’re error-prone, they make stuff up. The

buzzword is hallucinations, but…the variety of feedback is just invaluable to us. And we’re very

honest. And when I send these guidelines on how to test, I literally

say the very first bullet point says the AI is guilty until proven

innocent. With that mindset, because everybody who works with LLMs in

detail knows that they’re error-prone, they make stuff up. The

buzzword is hallucinations, but…

Orlando Wood (31:56)

Mm-hmm.

Yeah.

Andre Vargas (32:19)

for data scientists, are errors, just like prediction errors.

Orlando Wood (32:21)

Yeah. You once

explained to me as well as the difference between a bullshitter and a

liar.

Andre Vargas (32:27)

That exactly.

exactly. So, you know, a liar knows he or she is lying because the

goal and objective is to deceive. Right. A bullshitter tells you

stuff, you know, with conviction. May or not know if it’s true or not.

I just believe that. Right. So they’ll just bullshit you. And there’s

like

think there’s articles and books about how LLMs, some people call them

bullshit machines. I think they have a lot more value than people give

them, but less value than it’s promised.

Orlando Wood (32:59)

Well, their

primary job is to give you an answer you’re happy with. So if you walk

away with that information, but you’re happy, the LLM feels like it’s

done its job.

Andre Vargas (33:08)

Yeah, but sometimes you also need the truth. And if you are not

getting the truth, I would not feel comfortable an agent taking what

could be a bullshit answer into a negotiation. Because what is my A-

list client gonna think if, let’s say if we did a script summary that

made three characters up and…

My client who’s a writer, let’s say a writer client talks to our agent

and says, what did you think about my script? And my agent talks about

these three characters that I’ve made up and they don’t appear on a

script. That relationship’s over forever, right? Number one, why

didn’t you actually read my script, right? So when we do these kinds

of…Orlando Wood (33:47)

And

then even worse, why did you pretend to me you had? That’s the

relationship breaker there. You pretended to me, you were false. Yeah.

Andre Vargas (33:51)

Yeah. I can’t do it.

So by

saying the AI is guilty until proven innocent, if we are working on

script summaries, for instance, there is no universe where you have

not read the script. The job is… Yes, exactly.

Orlando Wood (34:07)

That’s one of the places that you and I always agreed on from day one

is that the tool that we

built Koobrik we always say when we calculate the amount of time it

saves and everything else, we always say this is not, we do not

calculate the screen reading time. You must read the screenplay

Andre Vargas (34:22)

Exactly. And this is why we connected with you so well, because of

that alignment of understanding of where does these types of tools add

value to the workflow, right? It’s not, look, I did a summary. It’s

like, no, no, no. What’s the actual job, right? So I appreciate that

very much. And this is why we got connected, I even think, two years

ago. Like, time is flying. But

Orlando Wood (34:43)

Yeah, I know.

Andre Vargas (34:46)

You know, by saying the AI is guilty until proven innocent, I’m also,

how are you gonna know if it’s guilty or not guilty if you haven’t

read the script, right? So that’s number one. Number two, in the

output that we generate, we generated in, you know, in CAA templates,

and we added a couple of things. You know, the EU AI Act has a

requirement that you have to…

Openly if something is generated synthetically you have to say these

outputs was generated with an AI assistant So we add that in the

output, you know, this output has been generated by the know, ca AI

assistant and then the next line says You know Approved by or edited

by and then an agent name if there is no agent name. I would not trust

it Right if there’s an agent name and let’s say Orlando. You’re the

agentand you put in your name and approved and edited by Orlando. And a

year later, I pick up this summary because in some search situation,

you might find this script popping up. Somebody’s looking for aliens,

movies somewhere, blah, blah, comedy. And then the script shows up and

it says Orlando approved it. Then I could give you a call. Hey

Orlando, you read this script and I think there’s an opportunity for

us versus.

Orlando Wood (35:42)

Mm-hmm.

Andre Vargas (35:57)

not knowing if anybody validated and then, you know, do you trust it?

Right? So that’s the thing. Right. So, so we have to tie in these

things. So there’s always a human in the loop, especially with LLMs,

anybody listening to the podcast, if you’re using LLMs and you are not

having a human validating, you know, the error rates are pretty high.

You know, if you have an employee that gave you errors,

Orlando Wood (36:02)

100%.

Andre Vargas (36:20)

20, 30 % of the time you would be firing them, right? So just be super

careful in terms of how do you manage the validation of the outputs of

these LLMs is really important. So for us, it’s kind of like the human

in the loop has always been centered. This is what we called it code

assigned. This is what we’re always really involved with the agents so

that the LLMs or machine learning or Intel never deviates from what

the agents need to do for their clients, right?

That’s really, really important.

Orlando Wood (36:48)

how do you guys, you know, with all of the uncertainty, how do you

guys help artists navigate this new world? mean, it seems that not

only are there massive shifts happening in terms of where client,

where, consumers and, and, and audiences are watching their, their

things where,

you know, the shifts in technology, but shifts in technology across

the board. mean, VO3 just came out that has the ability to not just

remake a person, but also dialogue, also diegetic sound. You know,

you’ve got the ability to clone voices. You’ve got the ability to

clone writing styles. It seems like technology suddenly is this all

fronts attack. You know, what’s your guy’s role in protecting artists

in that?

space from these sorts of threats.Andre Vargas (37:36)

Yeah, so by staying really close with these big tech companies and

understanding their roadmaps and their plans, I mentioned earlier, you

know, the work, one of our groups inside the CA company, our strategic

development group is extremely focused in specifically what you’re

just asking about. By being really connected with this whole tech

ecosystem of what’s coming,

able to feedback back into the agency and to the agents for them to

communicate to managers, to lawyers, to artists themselves, hey this

is coming, this is what it means for you, if you want to opt out

here’s a link. So there is a lot of communication that our group

collaborates with this strategic development group in terms of

bringing that information of what’s going on on the outside very

quickly inside the agency so that the agents can communicate it out to

their clients.

and the people that are around our clients as well. So it’s a lot of

information sharing when it comes to what’s happening specifically on

generative AI.

We’ve also, you know, our clients see our outputs from Intel. You

know, a lot of the reports that we provide, they see them. I’ve heard

them talk about our data work, like in podcasts or somewhere else. And

I’m like, a client literally talks about the data work we’ve done with

such passion and like sometimes even better than I would.

Orlando Wood (38:53)

Yeah.

It’s powerful,

it’s an incredible tool for them to have in their arsenal. It must

feel very comforting for them to know that that work’s being done.

Andre Vargas (39:02)

It’s wonderful.

think when they see it, they’re blown away. Like, I didn’t know these

things. Like, I remember doing a, we were doing a presentation about a

touring client to the manager.

And this manager, you know, if you saw him in a dark alley, he would

run like he was like six foot seven, huge guy from, from the

Netherlands. So even speaking with like a Netherlandic accent and

showed him, you know, how we optimize tour routes and ticket sales and

venue selections and what are their audiences and what are their

affinities and so forth. And he literally just hit the table like, Oh

my God, I knew somebody once would figure this thing out. This is

awesome.Orlando Wood (39:45)

Yeah.

Andre Vargas (39:45)

So, you know, and there are many times where agents basically tell us,

hey, you know, I have a production company, client of mine, they’re

in, they want to understand, you know, about their projects. So we

bring them into the office. We put Intel on a big screen and they just

go into the screen and start strategy. They talk to each other. All we

do is just click around and they interact. You know, there’s like

three producer buddies that have a production company. They start

looking at the data. Can it show?

Orlando Wood (40:03)

Yeah.

Andre Vargas (40:12)

me this and can you show me that and that interaction is also inputs

for us you know this is how clients think about it you know ⁓ the

changes that we could make so ⁓

Orlando Wood (40:16)

Mm. Yeah.

I imagine that’s a huge reason

that you guys have probably the least amount of churn in terms of

clients than any other agency. I’d imagine that if you’re going like,

this is the level on which they understand everything I do on a daily

basis and how it works, you’re not going to leave that place, are you?

Andre Vargas (40:38)

Well, I can’t take credit for that. can tell you CAA is a magnificent

platform where we’re just a little part of that, right? So in our

strategies, it’s interesting. Like I created what would be the similar

of like a startup.

Orlando Wood (40:40)

⁓ please, Andre, take credit for it.

Andre Vargas (40:53)

kind of like an investment deck. And I have a slide that has what’s

the, know, where that we play in the value chain of CAA, right? So one

of them is actually that stickiness where it’s not only stickiness for

clients, but also for agents, right? So if we are able to provide the

services,

to our agents that then serve our clients in the best way possible,

that retention mechanism is just amplified, right? So the platform

itself is humongous. The data just adds to that in addition to, right?

So obviously, thank you for the compliment, but I think it’s

definitely powerful when we’ve heard agents tell us when they’ve won a

project because they brought the data. We started calling itto our agents that then serve our clients in the best way possible,

that retention mechanism is just amplified, right? So the platform

itself is humongous. The data just adds to that in addition to, right?

So obviously, thank you for the compliment, but I think it’s

definitely powerful when we’ve heard agents tell us when they’ve won a

project because they brought the data. We started calling it

that one of our specialties is turning a no into a yes, right? So the

initial response is like, I’m not sure we can do this or we’re not

gonna do it. And then we bring the data and then it’s like, yeah,

we’re gonna do it, right? So turning nos to yeses, getting a better

understanding of the value of our clients work just lifts everybody

up, right? And that’s very rewarding in itself. So you saying that,

that’s really wonderful too. Thank you.

Orlando Wood (42:00)

Well, pleasure. I say I only say it because it’s true. But yeah, I

mean, I think that, you know, it seems to me that you guys are quite

safe, actually, in this new world of democratized craft making there’s

so much about that that’s that’s, you know, sad, but interesting. And,

you know, and we’re certainly seeing.

The massive shift to online distribution in YouTube so many years ago,

now becoming a very real way that people are consuming their media.

YouTube was just unseated. Disney is the largest television

distributor in the world. I think the similar thing is happening here

with the craft to tell stories, but it seems to me that you guys are

sort of quite safe. The more artists that there are out there, the

more storytellers that there are out there. If somebody…

create something truly phenomenal out of their bedroom in Wisconsin,

you guys are still the place that those artists will come for

representation. And it seems to me that it’s only gonna be more

important what you do with all of this technological change on the

creator side.

Andre Vargas (43:00)

For sure, think opportunities only get amplified with more ways for

artists to showcase their art and tell their stories. think we just

need to make sure that artists are always going to have to make sure

that they monetize their work and they are well represented and not

taken advantage of.

they’re not necessarily data scientists, they’re not lawyers

themselves, they’re not agents themselves. And I think the role CAA

plays is to make sure to take care of them in a complicated business

world when they are basically artists, right? So I see our job in the

data team kind of like supporting that as well, telling them, they

might get some stats around video views or engagement and they might

get 90 days worth of data, but how does that relate to everybody else?

indexing across similar other audiences. And we can enrich a lot of

that information so that they feel like well taken care of and

themselves come up with new ideas and new opportunities to tell bigger

and better stories out there as well. I’m looking forward to see what

the next looks like. So imagine before there wereindexing across similar other audiences. And we can enrich a lot of

that information so that they feel like well taken care of and

themselves come up with new ideas and new opportunities to tell bigger

and better stories out there as well. I’m looking forward to see what

the next looks like. So imagine before there were

creators today, they were called, you know, influencers and they were

like in social media influencer space and we have a whole digital team

focusing on them. What’s Darman going to be doing next, right? So all

kinds of really interesting things that are that

net new only 10 years ago, less than 10 years ago. And now with this

technology, what’s next going to show up? Are we going to be looking

at more of the vertical type of mini things that are happening in

Latam today? There’s telenovelas in the shape of a mobile phone. So

more mediums, more opportunity, think our clients will push us to

where they need us to be.

Orlando Wood (44:39)

Yeah.

Andre Vargas (44:47)

So that’s where we benefit from their inventions and their creativity

and we are there to support them.

Orlando Wood (44:53)

Yeah,

I just have one more question that I’ve been asking everybody, which

is I had a great boss named Dan Wieden who was a giant in the

advertising world. And he had a quote where he said, you’re useless to

me unless you fucked up big at least twice. And so can you remember

and Would you be willing to admit a time where you messed up and what

you learned from it?

Andre Vargas (45:11)

Well, if I could use something that really helped me in my career, I

would have to rewind time a little bit further back than CAA because

that’s really helped me tremendously. remember my passion for data has

been going on for, I don’t know, 25 years, right? And I was in

management consulting in a company called PA Consulting Group.

Orlando Wood (45:21)

Well, fair,

Andre Vargas (45:37)

They are actually the ones that brought me from Denmark to the US in

2001. And I remember receiving an RFP for a company who wanted to do

data strategy, roadmap, planning. They wanted to build a data team

inside their company. And coming from Denmark and reading a lot of theyou know, responses that I saw from competition and stuff like that.

And I’m like, wow, Americans really not ever tell you the answer. They

just kind of like tell you enough so that you call them back. Right.

And I said, how about I actually give them the answer in my response

to their RFP. And I took a chance and I wrote literally, what you need

to do is this, this, this. I gave them the answer. Right. Because

Why are they asking for a consulting firm? It might be because they

can’t do it themselves. So me giving them the answer, it’s great. So

we got the job. The CIO of this company was blown away. He says, the

first time I get an RFP response that actually tells me what you’re

going to do. All the other ones seem to just be vague. And I don’t

know how to pick, but yours stood out literally because you told me

the answer.

And I was like, well, yeah, let’s get this done. So I win the job.

Couldn’t be happier. I have goosebumps telling you this story. And I

show up on the Monday. And on the Tuesday, he tells me, hey, I’m going

to change jobs. I got an opportunity at a different company. This was

the sponsor of the project, right? Well, Andre was very, very excited

about this job. And

Inside I said, ooh, we need to pause this project because I don’t know

who the next CIO is, what’s the sponsorship situation. But I was

blinded by the success of have sold this project in such a unique way

that I just said, no, no, we’re just going to get going. And very

painful project because there was no sponsorship. There was no CIO for

a while. Access to the people I needed wasn’t really great.

Orlando Wood (47:09)

Interesting.

Andre Vargas (47:27)

The team that we built to do this delivery were working almost like

chickens without a head and eight weeks after that, I don’t even think

anybody read the output of that project, right? So what I learned from

that is the importance of the sponsorship from the business side that

drives whatever you’re trying to do. If they’re lost, you’re basically

just floating around.

Orlando Wood (47:36)

⁓ no.

Mm-hmm.

Andre Vargas (47:50)

You know, yes, did I invoice for sure. Did they pay? I guess there was

a job code and it was in their AP system. So we got paid. But the fact

that we just meandered, ⁓ I would consider that a huge screw up. But

what I learned from there was, okay, don’t be stubborn, number one,

just because you’re so proud of the project that you are in, but the

importance.You know, yes, did I invoice for sure. Did they pay? I guess there was

a job code and it was in their AP system. So we got paid. But the fact

that we just meandered, ⁓ I would consider that a huge screw up. But

what I learned from there was, okay, don’t be stubborn, number one,

just because you’re so proud of the project that you are in, but the

importance.

Orlando Wood (47:50)

Yeah.

Yeah.

Yeah.

Andre Vargas (48:15)

of that business sponsorship. And that is really what I did also at

CAA. The first thing, if I were gonna do data, I better have the CFO

on my side, the president of the company, the CEO of the company, the

department heads, the agency board, the managing directors. I have

them all join a data steering committee every month. They all know

what we’re doing. Whenever there is something, I go to them and I say,

hey, you know, we need resources or hey, there’s a conflict between

these two departments. What should be the priority?

Orlando Wood (48:32)

Yeah, it’s smart.

Andre Vargas (48:41)

That sponsorship is what made Intel a success, but I think I learned

to build sponsorship because of that screw up.

Orlando Wood (48:48)

Yeah, I think it also feeds into one of the talents I’ve noticed about

you. You’re very good at explaining, you know, what’s going on. You’re

also incredibly personable as well. And I think that that is kind of

what, you know, what we’ve really appreciated. And I think that, you

know, from my perspective, think artists should be very confident in

you and how you, you know, react to things. I’ve been…

You know, one the reasons I’ve been so excited to talk to you, I think

that, you know, your your ethics in the face of everything that’s

going on right now, I think really sets you apart. And I certainly

have have been very impressed. ⁓ But.

Andre Vargas (49:21)

Yeah.

I think

a big difference too is not being a data tech company, I’m not out

there selling dashboards or selling the latest LLM wrapper. I’m here

to serve an agency that serves incredible talent. So that Northstar

that we started with literally, we serve this A-list superstar client

that’s my boss as a Northstar, really settles the whole team. And we

really leave thea big difference too is not being a data tech company, I’m not out

there selling dashboards or selling the latest LLM wrapper. I’m here

to serve an agency that serves incredible talent. So that Northstar

that we started with literally, we serve this A-list superstar client

that’s my boss as a Northstar, really settles the whole team. And we

really leave the

that we take care of each other, good things happen. It’s basically

the strategy of the company is the strategy of a data team. And our

team has been together now eight years. It’s the same group and it’s

only been growing as the competency and capability has grown in the

company. So yeah, it’s a blast to be honest, to be here.

Orlando Wood (50:06)

Well on that note, I won’t take up any more of your time. I so

appreciate you sitting down with us and it’s been fascinating for me

and I just really enjoy talking to you.

Andre Vargas (50:14)

Same here, and let’s keep doing more good stuff together. All right.

My pleasure.

Orlando Wood (50:18)

Absolutely, that sounds great. Thanks so much, Andre. Have a good one.