Childhood friends turned Cognitiv co-founders, Jeremy Fain and Dr. Aaron Andalman, went from zebrafish brains and ad-tech trenches to rewiring how brands predict human behavior in an AI-saturated marketplace. Together, they draw a sharp line between predictive and generative AI, show what first-party data can really do, and demystify context and “agents” so marketers can push performance further without losing the plot ... or their customers’ trust.
See omnystudio.com/listener for privacy information.
Listen
Watch
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:00):
Good Company is a production of iHeartRadio.
Speaker 2 (00:03):
There's been a one eighty flip. You know, when we
started this company, there was so much skepticism around the
concept of AI, and we are constantly having to sort
of articulate, you know, what deep learning was and why
it was different than machine learning of the past. And
now every single company out there is saying they're using AI,
(00:23):
so the question becomes, well, how are you using it?
Speaker 3 (00:30):
I'm Michael Casson and this is Good Company. Together we'll
explore the dynamic intersection of media, marketing, entertainment, sports and technology.
I'll be joined by visionaries, pioneers, and yes, even a
couple of disruptors for candid conversations as we break down
how these masters of ingenuity are shaping the future of business,
culture and everything in between. My bet is you'll pick
(00:54):
up a lessen or two along the way. As I
like to say, it's all good, Welcome back to Good Company.
AI has become the soundtrack to every conference panel, pitch deck,
and press release in our business. But underneath that constant chorus,
there's a much more interesting story. What happens when intelligence
(01:16):
is actually built into the fabric of how brands discover audiences,
learn and unlock growth. That's why I'm sitting down with
my friends at Cognitive today. Jeremy Fain, co founder and CEO,
and doctor Aaron Handelman, co founder and chief science officer,
a powerhouse duo with commercial and scientific credibility who has
(01:37):
spent the last decade wiring deep learning into the core
of advertising. Jeremy comes out of the frontlines of digital
media and ad tech, bringing an earnest business acumen and
a masterful fluency in the language of cmos and CFOs.
Aaron comes from the world of neuroscience, where he mapped
how neural circuits process signals in living brains, before turning
(01:59):
that expert tees into the neural systems that now power
Cognitive's decision making for major brands. Together, Aaron and Jeremy
have built a tremendous platform that advances performance and gives
each brand its own way of seeing the world through
custom models, sophisticated context tools, and AI personas that chase
(02:20):
intent instead of clandestine labels and fading IDs. This conversation
will dig into how this resets the standard for data,
reshapes the creative brief, and raises the bar on what
performance should look like Jeremy and erin Welcome to good Company.
Speaker 2 (02:36):
It's great to be here, Thanks for that introduction.
Speaker 3 (02:38):
You are childhood friends now running a deep learning company.
I'm curious both about that journey. But what if each
of you have brought to the table that the other
doesn't have. I mean, you've known each other most of
your life. How has that differentiation kind of helped shape cognitive.
Speaker 4 (02:55):
I would say the interesting thing about our friendship. So
Aaron and our third co founder, Mark and I we
all met in fifth grade in Maryland, Silver Spring, Maryland.
For some reason, we hit it off and we grew
up together until we went off to college, and then
we went off to different careers and different interests in
different studies, but we ended up in this situation where
(03:19):
we each bring the exact thing that you really need
to build an AI company. Aaron had the neuroscience PhD
with a computer science undergrad from Stanford. I had the
business side of advertising and big data through programmatic, and
Mark had been a quintessential software engineer and engineering leader
(03:42):
for his entire career.
Speaker 3 (03:43):
What's so amazing about this is this is like ten
years ago. Yeah, well before the hype around AI and
all of this, So there was some prescient, you know,
thoughts going on.
Speaker 2 (03:57):
I think that happened because I was coming out of Stanford. Stanford,
I mean, the University of Toronto was also a birthplace,
but Stanford was very close to the ground at the
start of this whole AI moment. So deep learning instead
of the neural networks are the core technology that power
it all. And I remember in twenty twelve, I was
(04:20):
studying zeeperfish recording from their brains, but I was getting
excited about artificial neural networks. Like I was auditing some
classes that were being co taught by Andrew Carpathi. He's
actually of now machine learning fame, and there was.
Speaker 5 (04:35):
Just a buzz.
Speaker 2 (04:37):
Google Brain was this early initiative of THEIRS that was
getting started, and so that I knew I wanted to
go do something with neural nets and between now just
call AI. And it was through conversations between Jeremy and
Mark that we were like, hey, I think there's real
potential here, and that's how we kind of got in
on the ground floor.
Speaker 5 (04:57):
But let me clarify something.
Speaker 4 (04:58):
What happened, What really happ happen was Mark and I
wanted to stop what we were doing. I was at
Rubicon Project, he was a sen Edison. We wanted to
start a company together. We came up with this idea
and we're like, all right, let's get Aaron to do this.
And the idea was like a mobile DSP or some
mobile video DSP or something. And Aaron said to us,
that's a dumb idea. I don't want to do that.
(05:20):
I want to do deep learning. And Mark and I
are like, we've never heard of that.
Speaker 5 (05:24):
What is deep learning?
Speaker 4 (05:26):
And then Aaron told us all about this amazing technology
ten years ago and we're like, absolutely, that's amazing. But
they said, but we need a lot of data. And
I was like, well, programmatic media has tons of data,
so this is a great opportunity for us to do something.
Speaker 5 (05:42):
Yeah. I think that thesis proved true.
Speaker 3 (05:44):
Jeremy drilling down a little here. You're very much in
a leadership seat. Was there a particular moment in time
or decision tree that defined how you want to leave
the company?
Speaker 4 (05:57):
I think that and Aaron reminded me of this. Aaron
and I and Mark actually we worked on some little
business ideas while we were in high school together, and
I even signed Aaron's notebook and said earbook and said, hey,
let's let's let's start a business some day together. So
I always wanted to be an entrepreneur, but I have
(06:18):
I have what I would call a medium risk tolerance.
Young entrepreneurs have pretty high risk tolerances. I won a
different way. I sort of meandered my way through digital
marketing and media in different roles, growing each time, sort
of learning.
Speaker 5 (06:33):
Different pieces of business.
Speaker 4 (06:34):
But I would say that once I got to Rubicon
Project and I took over sort of the North American
accounts there, we were building a culture there in real
time because we were about one hundred plus people and
it had grown very quickly.
Speaker 5 (06:49):
This SSP world.
Speaker 4 (06:50):
Had grown and exploded very quickly, and the company needed
somebody to come in and really organize the account system,
how we how we dealt with our clients. And that
is really where I cut my teeth on building organizations
and high performance teams and things like that.
Speaker 5 (07:08):
And I think that was the.
Speaker 4 (07:09):
Point my career at Rubicon over those four years was
the point at which I finally said to myself, Hey,
I think I think I have the skill set and
the maturity to start something and lead something and.
Speaker 3 (07:23):
How funny you use the word maturity there, because we'll
come back to that.
Speaker 5 (07:26):
Yeah, those are a color though.
Speaker 2 (07:28):
So I feel like Jeremy has some like a pretty
strong personality and he's really expressed that in the culture
of Cognitive. So Jeremy likes to have fun. So Cognitive
has like a fun culture.
Speaker 5 (07:41):
We have.
Speaker 2 (07:42):
We have these like we break our company into houses
that are cross teams, and we do competitions and we
all get to know each other.
Speaker 5 (07:48):
And I think.
Speaker 2 (07:49):
Jeremy's leadership style partly cut at Rubicon, but partly it's
just his innate personality. But all the way back to
your first question about like how being childhood friends has
shaped the company, I just wanted to add, like, I
think a huge part of that is the trust that
the three co founders have. It's hard to have trust
with someone you just met at a conference that start
a company with, but we all know each other's parents,
(08:12):
and that trust has just really allowed us to make
better decisions.
Speaker 3 (08:16):
It's interesting I made a decision years ago that I
was never going to go into too business with a friend,
and then I broke that decision and it didn't work out. Twice,
I've made many many, many, many of my closest friends
through business. But that's different. When you go into business
and become friends, that's one thing. When you go into
(08:37):
business with a friend, it's another thing. You guys have
had it work. And you know the rule that I
made when I tried this twice, I made a commitment
with the friend. I said, look, if we're going to
get into business, we have to have even a higher
level of transparency. We must not let anything ste We
have to. If you do something to pisses me off,
(08:58):
I'm going to freaking tell you right now, like I'm
not going to let it faster, and we can't. It's
so interesting, and you guys have obviously found the right mix.
You know, I don't want to go down that would
be a more of a sociological conversation, not a not
a business conversation.
Speaker 2 (09:14):
To do with the fact that we're friends from elementary school.
I think that's made a different kind of friend back then.
You know, you're left. I don't know. I think there's
a component of that. I also think we're very different
in our careers and so we balance each other. Like
I bring like this scientific rigor. I want to do
everything with a randomized controlled trial. But Mark and Jeremy
(09:34):
have taught me a lot about Hey, you don't always
have that luxury.
Speaker 3 (09:38):
But aaron interesting in your case. I get Jeremy's transition
from rubicon to and and you know, being in the
business and understanding advertising and early in programmatic and all
of those things that are critical elements in his situation
in his role. But you went from studying you know,
neural circuits and zebra fish to you know, advising brand
(10:00):
and how the realization that neuroscience and advertising belonged in
the same conversation in and of itself is today de
were gair. But when you guys were talking about this.
Speaker 2 (10:12):
It was not Yeah, I mean, it wasn't the path
I expected to go on, either of my But there
were a few different A I wanted to do neural nets,
and I wanted to do them in industry, and so
I was interested in doing something with neural nets, and
this opportunity came along. And honestly, there's so much to
learn about advertising marketing that I didn't have the background for.
(10:32):
So without Jeremy it would never have been a success.
Speaker 3 (10:36):
So can I switch gears to storytelling? Maybe you could
humor our audience with a little back of the napkin test.
If AI data and targeting were words that were banned
AI data and targeting, how would you explain cognitive to
a CMO with those hands.
Speaker 5 (10:56):
I'd be on your back.
Speaker 2 (10:57):
Yeah, I would say, Hey, there's a new technology that
means that computers can understand human content and the world
in a new way. So historically machines were number crunchers,
and I think what's really changed is today computers can
(11:18):
have this huge amount of background knowledge, Like now you
could ask them, hey, tell me all the major events
that happen related to basketball in a year, and it
could list them all out, or tell me where people
who are you know, fans of Fortnite might hang out,
or what content they might consume. That kind of world knowledge.
Computers never had that, and now they do. And that
(11:42):
is a fundamentally disruptive moment because it means that computers
can now have the knowledge that sort of powers marketing intuition,
and that means they can be a new kind of
partner in influencing people and helping them find products they
love or like. And I think cognitive is all about
(12:05):
taking this new technology and figuring out how to use
it to do advertising and marketing better.
Speaker 3 (12:13):
And Jeremy, how would you answer that question.
Speaker 4 (12:15):
I would make I would say more simply, which is
how we started the company, and what we believe we
really do is that we have a new technology. It's
not even new anymore, but ten ten years ago is
new that helps us predict consumer behavior period and I
think that's really what marketing is meant for. Right you
(12:36):
need to understand consumer behavior and figure out when to
insert your message at the right time.
Speaker 3 (12:42):
Good company will be right back after the break. It
strikes me that every pitch today coaims whatever you're pitching
AI powered. What's the quickest way a marketer can have
(13:05):
their excuse the expression bullshit detect her up to tell
whether a product is truly using deep learning versus just
kind of borrowing the language and you know, playing the
jargon game.
Speaker 2 (13:16):
That's a hard question because it kind of depends on
your skill set as the person you know judging them.
Like the way I would do it as someone who
knows a lot of the details about machine learning is
I would ask detailed questions. Okay, tell me where exactly
you're applying what you're calling AI, because I do think,
you know, there's been a one to eighty flip. You know,
when we started this company, there was so much skepticism
(13:40):
around the concept of AI, and we were constantly having
to sort of articulate, you know, what deep learning was
and why it was different than machine learning of the past.
And now, like you said, every single company out there
is saying they're using AI. So the question becomes, well,
how are you using it? Because you can just layer
it on the top. You might have the same fundamental
(14:01):
product and you're just putting a chatbot on top of it.
That's not going to actually improve performance, that's just going
to change the interface. Or you could be thinking deeply
about you know, what you know machines can now do
that they couldn't do before, and figuring out how to
integrate that into your product. My view is you've got
to ask questions that get at really what they mean
(14:23):
when they say it's AI powered.
Speaker 3 (14:25):
One of the things that we've been dealing with is
separating the wheat from the shaft, if you will, for
the cmos, because what I've noticed at this moment is
existential cmos are in this real crunch right now. They're
getting pressure from up above from the CFO or the
CEO or the CTO to be modernizing AI transformation, cost
(14:50):
cutting at the same time. As you know, I think
it's Satia and Adela who gets the most credit of
any executive of this generation for doing We used to
say change the tires while the car is moving. I
think it's better said perform while you transform, because that's
more applicable to what's happening now. And so in that
(15:11):
conversation again we could go down a very black hole,
not a bad hole, just a long journey on this
podcast on that topic. So I want to kind of
switch gears.
Speaker 4 (15:20):
I think there's a simple way of looking at it
which has been hard to understand the last year because
of the generative AI movement and it's all the chatbots
and the foundation models and the lms, and I think
for us, we really try to talk about being a
predictive AI company versus a generative AI company, and we
(15:41):
use generative AI and lllms, but there's inputs to predictive
AI algorithms. And I think a marketer look having a
front end put on and accessibility measures put on lms
and making them easier to use for certain things from
a marketing analysis perspective or planning perspective.
Speaker 3 (16:01):
Hey, have at it.
Speaker 5 (16:02):
That's great.
Speaker 4 (16:03):
There's going to be a million of those companies and
you just try the one that works best for your company.
But predictive AI is big data AI, and it's much
harder to do and you need your own tech to
do that. And so that's really I think you're going
to see more of that differentiation in how people position
themselves in the next couple of years, predictive.
Speaker 5 (16:22):
AI versus generative AI tools.
Speaker 3 (16:26):
Let me switch gears to you know, another one of
the cocktail party words that we hear a lot, first
party data. It's always mattered. It doesn't just now matter,
It's always mattered. And at the same time, you know,
many marketers are asking do we have the right data
and are we using it well? And I guess i'd say,
what's changed in the last couple of years, especially with
(16:48):
LMS that lets marketers you know, get more value out
of that data.
Speaker 2 (16:53):
Yeah, I think, like you said, first party data, I
think has always been valued or at least you know,
has been useful for me years. But I think there
has been a shift. It relates to what I was
saying before which is previously machine learning and add tech
was a lot of number crunching, and so it could
get some value out of your first party data, but
it didn't really understand like let's say you had a
(17:16):
list of people who bought a product. The AI didn't
or the machine learning didn't understand what product you're selling
or have any context for that. It's just a list
of here's the people. And now with foundation models, they
bring all of this understanding to the table, and that's
what allows us and like the others to get more
out of that first party data because you're combining that
(17:38):
first party data, which is you know who bought what,
who clicked on what, with all of this understanding, so
that rather than number crunching, you're kind of understanding that
the AI or that machine learning can really understand that
data in a more fundamental way.
Speaker 3 (17:52):
Both you and I, all three of us get the
chance to speak to cmos frequently. One of the conversations
that I've been interested in is how cmos are genuinely
excited about the opportunity to combine sort of foundation models
with their own data, but also everybody is conscious and
(18:13):
nervous about privacy as we have to be How do
you think brands can tap the value while keeping their
most sensitive information which they have to protect it and
under their control. I think you said it right.
Speaker 2 (18:27):
It's like this, Like you said, the CMO has to
strike that balance, right. They have to protect their data,
but they also want to take advantage of this moment
and get more out of their data.
Speaker 3 (18:36):
There's a line you have.
Speaker 5 (18:37):
To walk there.
Speaker 2 (18:38):
You need to if you're working with a partner, you
need to understand what they.
Speaker 5 (18:41):
Are allowed or not allowed to do with that first
party data.
Speaker 2 (18:44):
For example, we use first party data and combine it
with knowledge from foundation models, but we never share that
data with a foundation model provider like open AI. We
do that in house. You know others might not do that,
which might maybe something you don't want to do. So
it's a question of following the data, where is it
going to go if you share it, and making sure
(19:07):
you're working with partners who respect the importance of that data.
You don't want that data being the next set of
data that goes to train a foundation model, right. Foundation
models trained on public data. First party data is private data.
You've got to make sure it doesn't go into those models.
And the question is how do you do that critically important?
Can we switch gears to context GPT. It's been in
(19:28):
the market for a couple of years, but as a
first of its kind product for someone hearing about it
for the first time, how do you describe what it
actually does for a brand?
Speaker 4 (19:38):
Context GPT is the first contextual targeting tool that allows
you to very easily target at scale an idea or
an audience that you're looking for without IDs. And so
we have two different types of advertised there's these days
(20:00):
that work with us on context GBT. First of all,
it's our fastest growing product. It's sales or skyrocketing because
you use a PROMPT to put in what you're looking for.
We were the first in market with that. But now
you can use this world knowledge that Aaron's been talking
about to surround an audience. So we have again two
(20:22):
types of advertisers. We have sports apparel advertisers who want
to surround the World Cup everything about World Cup they
want to be on, or the Super Bowl or marathons
things like that. Or you have CpG companies like soda
manufacturers who want to be around gen z fun concert
(20:44):
content right or not concert content just gen z right,
they want to surround gen Z, but they want it
to be fun content.
Speaker 5 (20:51):
How do you find that?
Speaker 4 (20:53):
The old way of doing it, you had to know keywords,
you had to understand who are the gen Z music
stars that people are listening to, et cetera. But now
as combining it with foundation models like open AI, which
context GPTs is integrated with, we can fine tune all
of that and deliver topics and segments of URLs based
(21:20):
on the entire content of the page using brand safety
measures like you know, is it is it a negative
review of a product?
Speaker 5 (21:30):
Is it?
Speaker 4 (21:33):
You know, violence and drugs and things like that, And
now you can find every single RL in real time
as it comes online and target it with your ads.
Speaker 2 (21:44):
I mean, I think if it like I think it's
obvious if you use chat GPT that computers just understand
text and images better than ever before. And context GPT
was saying, hey, that's a powerful thing for contextual targeting.
You type in what you want and the content and
LLLM can understand that page and answer whether this page
(22:05):
meets your criteria. So you just type in your criteria
and we'll figure out whether each r L fits that criteria.
And I think another thing we did differently that I
think has been one reason the product has been so
successful is it's radically transparent. You type in your prompt
and we show you every ur L that we think
matches that prompt, and like how relevant it is.
Speaker 5 (22:25):
People were used.
Speaker 2 (22:26):
To like, you pick some taxonomy category like sports and
you have no idea what's in it, And here you
type in a customized prompt and we tell you exactly
what's in it. And I think that that's another reason
it's been popular.
Speaker 3 (22:39):
I love that because what I learned last year a
joke about this frequently. Megan Trainor recorded that great song
It's all about the bass, so she taught us that.
What I learned from Jensen Wang in a conversation was
it's all about the prompt, so you know, yeah, yeah,
bringing the prompt to life. The other big buzz that
(23:00):
came out of early January at CES, which was a
gentic Ai kind of a new agency darling. But I
think there are instances where the term is getting misused wildly.
So what I'd love for you guys to do is,
you know, clear it up. Give me a sentence on
agentic Ai, maybe even two sentences.
Speaker 2 (23:21):
I had a recent addicts Changer piece where I gave
sort of my perspective on where agentic AI is. So
you want the details, go read that. But I think
the one sended summary is sort of right now agentic
AI or there's also this term ad CP. I think
their position to sort of make buying easier at least
over the next year, but I don't think they're really
(23:43):
in a position to make buying smarter or better yet.
And so that's sort of the one sended summary, and
then maybe I could add some color there. So the
term agent, and with respect to machine learning, kind of
comes from the seven these this this whole field of
machine learning called reinforcement learning, and it's actually the idea
(24:06):
of an agent is it's a machine learning tool that
can go out there and optimize something like a KPI.
If we're using advertising lingo, right, We're going to go
optimize a KPI and the agent is going to do
it through trial and error and like figure out how
to maximize this thing by through machine learning. That's how
that's what an agent really means. And today that term
(24:29):
is kind of getting redefined to mean something much more
basic in my opinion, which is today an agent means
asking a chatbot to take some action, Like if you
ask a chatbot to book you a reservation, that's an agent, right,
That's a very different and much simpler kind of agent.
(24:49):
That kind of agent's great for convenience or like, you know,
making something simple. Hey, I want to set up a strategy,
but I don't want to do it myself with twenty clicks.
I just want to tell alem to go set up
a strategy. But that sort of LLM definition of an
agent is not well equipped, at least this year and
probably for the next few years to really optimize an
(25:10):
ad campaign.
Speaker 3 (25:11):
What got me nervous growing up in the entertainment business
and having recent experience with agents, you know, not so good.
Sod about that those of our listeners who who know?
If you know, you know, But it's just funny. I
wish there were another word for agentik.
Speaker 2 (25:28):
I think these agents are cheaper though, yeah, I mean
I don't want, I don't want. Agents are getting better.
Chat Chept was a huge moment last November. Claude of
Anthropic released their new claud code tool from the developer side.
That's another chatchept moment. People are just amazed at how
(25:49):
good that is at producing code for you. Agents are
cool and they're very useful, but and you know, I
had bad experiences with them a year ago, and they're
going to keep getting better, but they're far from being
able to look at billions of rows of impressions and
knowing which one was viewed and which one is clicked.
(26:09):
You know, their their language tools, they think through through language.
They don't they're not really number tools yet.
Speaker 3 (26:15):
So so let me ask a question. You know, we
all understand there's got to be this human plus AI,
you know, collaboration, if for nothing else, kind of to
battle against another often used term recently AI slop, you know,
and and what what there is a rise in concern
I think with good reason around this AI slop. How
(26:39):
do you design tools so that the human stays firmly
in the loop.
Speaker 4 (26:42):
Can I just add something I think on that on
that topic about agents, because I think that's that's the
thing that people seem to miss about the agent to
agent discussions that are happening in the advertising business. They're
talking a lot about Hey, I have a buying agent,
and you're going to have selling agents, and they're going
to talk back and forth and then the buying agent's
(27:04):
going to decide what to do. I think that we
all know in the media business that that seems a
little crazy.
Speaker 5 (27:12):
Every single brand.
Speaker 4 (27:15):
Of seller, let's say publisher, let's just easily call them publishers.
They stand for something in the marketplace. They have to
differentiate themselves. A lot of that has to do with
human creativity and understanding deeply what a brand is and
pushing that into some kind of automated sort of handshake
(27:35):
back and forth. I think is going to be problematic
for a lot of buyers and sellers because of the
ability to manipulate a buying agent in this way. So
I'm interested to see how this moves forward. Besides, something
across commodities in the advertising space.
Speaker 3 (27:53):
Makes good sense. Let's talk about kind of looking ahead
outside of advertising. What area of AAR are you watching closely?
And Aaron, I would throw this your way first, and
I would say, if we were to replay this episode
in twenty four months, is there one prediction about AI
and marketing that you would be comfortable being held to?
(28:15):
So I'm asking you two questions there.
Speaker 2 (28:18):
Yeah, yeah, what I'm paying more attention to is related
to your AI slop question, which is when it came
to the ability of models to sort of reason about
people and their intents.
Speaker 5 (28:34):
A year ago.
Speaker 2 (28:36):
You know, we've been exploring this for a while. There
was a lot of AI slop we've seen over the
last year, sort of a leap forward in terms of
a model's ability to look at a series of behavioral
signals and reason about what the intent of that person
might be from a purchasing perspective or maybe from other perspectives.
And I think that is a meaningful There's still slop there,
(29:00):
but there's potential there too, and that's what we're leading into,
and I'm paying a lot of attention into, like how
reasoning models are going to allow us to make better
performance advertising tools that don't rely on as much first
party data, or at least allow us to get more
(29:21):
out of sort of data list pixel less solutions.
Speaker 3 (29:24):
We're going to hit pause for a moment, but stay
with us after the break. We've got more insights to share. Jeremy,
if I take this away from tech for a moment
(29:45):
and talk about leadership and talent in a volatile market.
I mean, we're in a volatile market both from the
talent perspective and as well the competitive landscape. But Aaron
said it at the beginning at cognitive, how do you
as a leader keep that culture, collaborative, fun enjoyable in
a category that is intense and fast moving and highly competitive.
(30:10):
I'm just curious what are the kind of secrets you
can share with our audience about how you, as a leader,
you know, navigate this interesting time in the market.
Speaker 5 (30:21):
For sure.
Speaker 4 (30:23):
Yeah, I think that for us we're on the upside
of AI and transformation from an internal perspective, not just
from our tool set for marketers, but from our tooling
as a company. And I think that's really giving us
an advantage. So Amazon just laid off sixteen thousand more people.
Speaker 3 (30:44):
Saw that as and then it's on top of fourteen
thousand in December to look at that and say thirty
thousand people within the space of a month, which those
are real numbers.
Speaker 4 (30:55):
Yeah, and they are very transparent about it being AI
efficiency for us, we're using it on the other side,
instead of needing three hundred, five hundred thousand people, we
need one hundred and fifty to two hundred people to
do what three hundred, four hundred, five hundred people we're
going to do. I think two three years ago, and
I think that is a big advantage for us and
(31:18):
companies that have come up at this time versus companies
that are trying to retool large organizations in a different way.
Speaker 3 (31:27):
So if you don't start with a yoke around your neck,
you've got an ability to build for the current, not
for rebuild from the past. Let me let me, let
me talk about that. In terms of collaboration, you work
with partners like Magnet, Index, Marketplace, open Ai, others. Where
do you see the greatest opportunity to simplify the path
(31:47):
from sort of hey I have an idea to actually
activating it.
Speaker 4 (31:52):
Well, I think we've seen an acceleration there. We work
in advertising technology. It's probably the most competitive and the
fastest evolving industry. I think out there period what Index
did last year, Magnite and open ai, open x and
all these other guys, Pubmatic are trying to copy, and
then somebody will come up with a new idea and
(32:13):
everyone else we'll pile on that. Because they're trying to
gain an advantage on a constant basis around something that
is becoming they're trying to stop from becoming commoditized. So
we work very quickly and efficiently with our partners, especially
where we know that AI needs this extra thing right
(32:37):
and with Index we're integrated and co located in their
server system. They've been working hard on containers, which is
great for some companies, but for us, we want to
be able to run full neural network models and we
can't use a container. We have to have our own
servers and our own systems and our own datas databases
(32:58):
all together in one place so that we can in
real time predict this consumer behavior that we're talking about.
We worked very hard with Index about that, working with
Magnete on that, and we're helping them understand what that
next generation of modeling is going to require. And I
think they've been really responsive to that.
Speaker 3 (33:21):
Well, look, you know, none of us go through this alone,
none of us get there alone. So the idea, Jeremy,
and I've seen this with you over the years, and
I've watched that, and I've watched your ability to be
a collaborator, and you know, I see it in both
how you operate but also how you internally but how
you operate externally with partners and the collaboration. So I
(33:43):
commend you on that, guys. I come to my favorite part,
my lightning round. So I'm going to throw some questions
at you.
Speaker 2 (33:50):
We'll do our best.
Speaker 3 (33:51):
Aaron, I'll start with you. What's your greatest professional fear?
Speaker 2 (33:55):
Oh, professional fear As an academic, I don't want to
be wrong, you know. I feel like you're wrong once
and your trust is forever hurt. So I like to
be careful in my wording because I think I fear
undermining my credibility.
Speaker 5 (34:12):
That's it. That's it.
Speaker 3 (34:12):
That's a good one. Jeremy is there. This is good
because we've touched on so many in this conversation. Is
there one industry buzzword you wish would disappear forever?
Speaker 5 (34:23):
Last click attribution?
Speaker 3 (34:26):
Good? Okay, If you had twenty four hours, this is
a jump ball.
Speaker 5 (34:30):
Either one of you grab it.
Speaker 3 (34:31):
If you had twenty four hours with no restrictions, what's
one adventure you would embark on.
Speaker 2 (34:36):
Oh, I would like to summit a peak, you know,
I'd like, I'm gonna get out of town. I want
to get to the top of some mountain. Not Everest
too hard, something more reasonable. But I really enjoy getting
to the top of a mountain and looking out. And
if I had twenty four hours, I would go do that.
Speaker 3 (34:54):
This is for both of you. Was there a particular
mentor early in your career? And you know, like I
heard at the beginning of our conversation, I think, Aaron,
you said it, and it's so important in relationships. You said,
we knew each other's parents. I love that because you
understood your Mark and Jeremy through a different lens. Okay,
(35:15):
you knew their parents. That was very telling to me.
I hope our audience heard that as well. But did
you have a mentor either one of you in your
life that early that gave you a bit of advice that.
Speaker 5 (35:27):
You know?
Speaker 2 (35:28):
For me, it was actually my PhD advisor, Michael Fee.
He was a physicist by training and he and he
spent a lot of one on one time with me
just teaching me a science, the scientific method, how to think,
how to go back to fundamentals. He just taught me
(35:48):
if you go back to physics, if you go back
to the scientific method, you will make progress. You know,
the goal of you know, academias sort of pushed the
envelope of knowledge, and he really taught me how to
do that in Jeremy.
Speaker 4 (36:00):
For me, my first real mentor ended up being a
woman who I worked with. She was a partner on
a project I was on at Digitas. That was sort
of my first job in marketing and interactive marketing and advertising.
Speaker 3 (36:16):
But by the way, Jeremy, I think that's where we
first met.
Speaker 4 (36:19):
Actually, I mean, it was a long time ago, and
it was a great time to be at Digitas. I
can't that was a great That was a great company
for me to really grow up in marketing on before
I went to business school. But we were on this
big project for Morgan Stanley. We were building this huge
interactive dashboard for them back in you know, early two
(36:40):
thousands when this was the newest, newest thing, and I
had been brought on to be an assistant development manager.
Speaker 5 (36:47):
This was back when I was still coding.
Speaker 4 (36:50):
And the manager got fired from the project because we
missed a bunch of deadlines and it just didn't work out.
And I was thrust into this manager role when I
was like twenty five four years old, twenty three years old,
and I didn't know anything about management. And I think
everybody who goes into management with no management training does
things exactly the way they would do things, and that
(37:11):
is not a good way to manage a large group
of people.
Speaker 5 (37:14):
It was a hard it was a hard learning.
Speaker 4 (37:18):
Experience for me, but my boss, my partner on that project,
sat me down and really taught me a lot of
very fundamental things about management, one of them being, Hey,
you have to talk to everybody individually. You have to
understand them individually and what motivates them individually. And that
was the beginning of my sort of career as a manager.
And it was She was by far the most helpful
(37:40):
to me in that time when I had no idea,
and I always look back on that as being a huge,
a huge.
Speaker 5 (37:48):
Turning point for me.
Speaker 4 (37:49):
It could have gone really badly, and she really helped me. Guys,
you've really helped me today. You've clarified a couple of
things in this world of jargon that I think brought
clarity for me, and I think it will as well
for our audience. I want to thank you Aaron and
Jeremy for taking the time, Thank you for joining me
(38:11):
on Good Company.
Speaker 5 (38:12):
Thanks for asking us.
Speaker 3 (38:14):
Yeah, thank you.
Speaker 2 (38:15):
I enjoyed the conversation.
Speaker 3 (38:21):
I'm Michael Casson, thanks for listening to Good Company.
Speaker 1 (38:25):
Good Company is brought to you by Three C Ventures
and iHeart Podcasts Special thanks to Alexis Borgerer Pudeo, our
executive producer and head of Content and Talent, and to
Carl Ketle, executive producer.
Speaker 5 (38:37):
At iHeart Podcasts.
Speaker 1 (38:39):
Episodes are produced and edited by Mary Doo. Thanks for
joining us.
Speaker 5 (38:44):
We'll see you next time.
Good Company is a production of iHeartRadio.
Speaker 2 (00:03):
There's been a one eighty flip. You know, when we
started this company, there was so much skepticism around the
concept of AI, and we are constantly having to sort
of articulate, you know, what deep learning was and why
it was different than machine learning of the past. And
now every single company out there is saying they're using AI,
(00:23):
so the question becomes, well, how are you using it?
Speaker 3 (00:30):
I'm Michael Casson and this is Good Company. Together we'll
explore the dynamic intersection of media, marketing, entertainment, sports and technology.
I'll be joined by visionaries, pioneers, and yes, even a
couple of disruptors for candid conversations as we break down
how these masters of ingenuity are shaping the future of business,
culture and everything in between. My bet is you'll pick
(00:54):
up a lessen or two along the way. As I
like to say, it's all good, Welcome back to Good Company.
AI has become the soundtrack to every conference panel, pitch deck,
and press release in our business. But underneath that constant chorus,
there's a much more interesting story. What happens when intelligence
(01:16):
is actually built into the fabric of how brands discover audiences,
learn and unlock growth. That's why I'm sitting down with
my friends at Cognitive today. Jeremy Fain, co founder and CEO,
and doctor Aaron Handelman, co founder and chief science officer,
a powerhouse duo with commercial and scientific credibility who has
(01:37):
spent the last decade wiring deep learning into the core
of advertising. Jeremy comes out of the frontlines of digital
media and ad tech, bringing an earnest business acumen and
a masterful fluency in the language of cmos and CFOs.
Aaron comes from the world of neuroscience, where he mapped
how neural circuits process signals in living brains, before turning
(01:59):
that expert tees into the neural systems that now power
Cognitive's decision making for major brands. Together, Aaron and Jeremy
have built a tremendous platform that advances performance and gives
each brand its own way of seeing the world through
custom models, sophisticated context tools, and AI personas that chase
(02:20):
intent instead of clandestine labels and fading IDs. This conversation
will dig into how this resets the standard for data,
reshapes the creative brief, and raises the bar on what
performance should look like Jeremy and erin Welcome to good Company.
Speaker 2 (02:36):
It's great to be here, Thanks for that introduction.
Speaker 3 (02:38):
You are childhood friends now running a deep learning company.
I'm curious both about that journey. But what if each
of you have brought to the table that the other
doesn't have. I mean, you've known each other most of
your life. How has that differentiation kind of helped shape cognitive.
Speaker 4 (02:55):
I would say the interesting thing about our friendship. So
Aaron and our third co founder, Mark and I we
all met in fifth grade in Maryland, Silver Spring, Maryland.
For some reason, we hit it off and we grew
up together until we went off to college, and then
we went off to different careers and different interests in
different studies, but we ended up in this situation where
(03:19):
we each bring the exact thing that you really need
to build an AI company. Aaron had the neuroscience PhD
with a computer science undergrad from Stanford. I had the
business side of advertising and big data through programmatic, and
Mark had been a quintessential software engineer and engineering leader
(03:42):
for his entire career.
Speaker 3 (03:43):
What's so amazing about this is this is like ten
years ago. Yeah, well before the hype around AI and
all of this, So there was some prescient, you know,
thoughts going on.
Speaker 2 (03:57):
I think that happened because I was coming out of Stanford. Stanford,
I mean, the University of Toronto was also a birthplace,
but Stanford was very close to the ground at the
start of this whole AI moment. So deep learning instead
of the neural networks are the core technology that power
it all. And I remember in twenty twelve, I was
(04:20):
studying zeeperfish recording from their brains, but I was getting
excited about artificial neural networks. Like I was auditing some
classes that were being co taught by Andrew Carpathi. He's
actually of now machine learning fame, and there was.
Speaker 5 (04:35):
Just a buzz.
Speaker 2 (04:37):
Google Brain was this early initiative of THEIRS that was
getting started, and so that I knew I wanted to
go do something with neural nets and between now just
call AI. And it was through conversations between Jeremy and
Mark that we were like, hey, I think there's real
potential here, and that's how we kind of got in
on the ground floor.
Speaker 5 (04:57):
But let me clarify something.
Speaker 4 (04:58):
What happened, What really happ happen was Mark and I
wanted to stop what we were doing. I was at
Rubicon Project, he was a sen Edison. We wanted to
start a company together. We came up with this idea
and we're like, all right, let's get Aaron to do this.
And the idea was like a mobile DSP or some
mobile video DSP or something. And Aaron said to us,
that's a dumb idea. I don't want to do that.
(05:20):
I want to do deep learning. And Mark and I
are like, we've never heard of that.
Speaker 5 (05:24):
What is deep learning?
Speaker 4 (05:26):
And then Aaron told us all about this amazing technology
ten years ago and we're like, absolutely, that's amazing. But
they said, but we need a lot of data. And
I was like, well, programmatic media has tons of data,
so this is a great opportunity for us to do something.
Speaker 5 (05:42):
Yeah. I think that thesis proved true.
Speaker 3 (05:44):
Jeremy drilling down a little here. You're very much in
a leadership seat. Was there a particular moment in time
or decision tree that defined how you want to leave
the company?
Speaker 4 (05:57):
I think that and Aaron reminded me of this. Aaron
and I and Mark actually we worked on some little
business ideas while we were in high school together, and
I even signed Aaron's notebook and said earbook and said, hey,
let's let's let's start a business some day together. So
I always wanted to be an entrepreneur, but I have
(06:18):
I have what I would call a medium risk tolerance.
Young entrepreneurs have pretty high risk tolerances. I won a
different way. I sort of meandered my way through digital
marketing and media in different roles, growing each time, sort
of learning.
Speaker 5 (06:33):
Different pieces of business.
Speaker 4 (06:34):
But I would say that once I got to Rubicon
Project and I took over sort of the North American
accounts there, we were building a culture there in real
time because we were about one hundred plus people and
it had grown very quickly.
Speaker 5 (06:49):
This SSP world.
Speaker 4 (06:50):
Had grown and exploded very quickly, and the company needed
somebody to come in and really organize the account system,
how we how we dealt with our clients. And that
is really where I cut my teeth on building organizations
and high performance teams and things like that.
Speaker 5 (07:08):
And I think that was the.
Speaker 4 (07:09):
Point my career at Rubicon over those four years was
the point at which I finally said to myself, Hey,
I think I think I have the skill set and
the maturity to start something and lead something and.
Speaker 3 (07:23):
How funny you use the word maturity there, because we'll
come back to that.
Speaker 5 (07:26):
Yeah, those are a color though.
Speaker 2 (07:28):
So I feel like Jeremy has some like a pretty
strong personality and he's really expressed that in the culture
of Cognitive. So Jeremy likes to have fun. So Cognitive
has like a fun culture.
Speaker 5 (07:41):
We have.
Speaker 2 (07:42):
We have these like we break our company into houses
that are cross teams, and we do competitions and we
all get to know each other.
Speaker 5 (07:48):
And I think.
Speaker 2 (07:49):
Jeremy's leadership style partly cut at Rubicon, but partly it's
just his innate personality. But all the way back to
your first question about like how being childhood friends has
shaped the company, I just wanted to add, like, I
think a huge part of that is the trust that
the three co founders have. It's hard to have trust
with someone you just met at a conference that start
a company with, but we all know each other's parents,
(08:12):
and that trust has just really allowed us to make
better decisions.
Speaker 3 (08:16):
It's interesting I made a decision years ago that I
was never going to go into too business with a friend,
and then I broke that decision and it didn't work out. Twice,
I've made many many, many, many of my closest friends
through business. But that's different. When you go into business
and become friends, that's one thing. When you go into
(08:37):
business with a friend, it's another thing. You guys have
had it work. And you know the rule that I
made when I tried this twice, I made a commitment
with the friend. I said, look, if we're going to
get into business, we have to have even a higher
level of transparency. We must not let anything ste We
have to. If you do something to pisses me off,
(08:58):
I'm going to freaking tell you right now, like I'm
not going to let it faster, and we can't. It's
so interesting, and you guys have obviously found the right mix.
You know, I don't want to go down that would
be a more of a sociological conversation, not a not
a business conversation.
Speaker 2 (09:14):
To do with the fact that we're friends from elementary school.
I think that's made a different kind of friend back then.
You know, you're left. I don't know. I think there's
a component of that. I also think we're very different
in our careers and so we balance each other. Like
I bring like this scientific rigor. I want to do
everything with a randomized controlled trial. But Mark and Jeremy
(09:34):
have taught me a lot about Hey, you don't always
have that luxury.
Speaker 3 (09:38):
But aaron interesting in your case. I get Jeremy's transition
from rubicon to and and you know, being in the
business and understanding advertising and early in programmatic and all
of those things that are critical elements in his situation
in his role. But you went from studying you know,
neural circuits and zebra fish to you know, advising brand
(10:00):
and how the realization that neuroscience and advertising belonged in
the same conversation in and of itself is today de
were gair. But when you guys were talking about this.
Speaker 2 (10:12):
It was not Yeah, I mean, it wasn't the path
I expected to go on, either of my But there
were a few different A I wanted to do neural nets,
and I wanted to do them in industry, and so
I was interested in doing something with neural nets, and
this opportunity came along. And honestly, there's so much to
learn about advertising marketing that I didn't have the background for.
(10:32):
So without Jeremy it would never have been a success.
Speaker 3 (10:36):
So can I switch gears to storytelling? Maybe you could
humor our audience with a little back of the napkin test.
If AI data and targeting were words that were banned
AI data and targeting, how would you explain cognitive to
a CMO with those hands.
Speaker 5 (10:56):
I'd be on your back.
Speaker 2 (10:57):
Yeah, I would say, Hey, there's a new technology that
means that computers can understand human content and the world
in a new way. So historically machines were number crunchers,
and I think what's really changed is today computers can
(11:18):
have this huge amount of background knowledge, Like now you
could ask them, hey, tell me all the major events
that happen related to basketball in a year, and it
could list them all out, or tell me where people
who are you know, fans of Fortnite might hang out,
or what content they might consume. That kind of world knowledge.
Computers never had that, and now they do. And that
(11:42):
is a fundamentally disruptive moment because it means that computers
can now have the knowledge that sort of powers marketing intuition,
and that means they can be a new kind of
partner in influencing people and helping them find products they
love or like. And I think cognitive is all about
(12:05):
taking this new technology and figuring out how to use
it to do advertising and marketing better.
Speaker 3 (12:13):
And Jeremy, how would you answer that question.
Speaker 4 (12:15):
I would make I would say more simply, which is
how we started the company, and what we believe we
really do is that we have a new technology. It's
not even new anymore, but ten ten years ago is
new that helps us predict consumer behavior period and I
think that's really what marketing is meant for. Right you
(12:36):
need to understand consumer behavior and figure out when to
insert your message at the right time.
Speaker 3 (12:42):
Good company will be right back after the break. It
strikes me that every pitch today coaims whatever you're pitching
AI powered. What's the quickest way a marketer can have
(13:05):
their excuse the expression bullshit detect her up to tell
whether a product is truly using deep learning versus just
kind of borrowing the language and you know, playing the
jargon game.
Speaker 2 (13:16):
That's a hard question because it kind of depends on
your skill set as the person you know judging them.
Like the way I would do it as someone who
knows a lot of the details about machine learning is
I would ask detailed questions. Okay, tell me where exactly
you're applying what you're calling AI, because I do think,
you know, there's been a one to eighty flip. You know,
when we started this company, there was so much skepticism
(13:40):
around the concept of AI, and we were constantly having
to sort of articulate, you know, what deep learning was
and why it was different than machine learning of the past.
And now, like you said, every single company out there
is saying they're using AI. So the question becomes, well,
how are you using it? Because you can just layer
it on the top. You might have the same fundamental
(14:01):
product and you're just putting a chatbot on top of it.
That's not going to actually improve performance, that's just going
to change the interface. Or you could be thinking deeply
about you know, what you know machines can now do
that they couldn't do before, and figuring out how to
integrate that into your product. My view is you've got
to ask questions that get at really what they mean
(14:23):
when they say it's AI powered.
Speaker 3 (14:25):
One of the things that we've been dealing with is
separating the wheat from the shaft, if you will, for
the cmos, because what I've noticed at this moment is
existential cmos are in this real crunch right now. They're
getting pressure from up above from the CFO or the
CEO or the CTO to be modernizing AI transformation, cost
(14:50):
cutting at the same time. As you know, I think
it's Satia and Adela who gets the most credit of
any executive of this generation for doing We used to
say change the tires while the car is moving. I
think it's better said perform while you transform, because that's
more applicable to what's happening now. And so in that
(15:11):
conversation again we could go down a very black hole,
not a bad hole, just a long journey on this
podcast on that topic. So I want to kind of
switch gears.
Speaker 4 (15:20):
I think there's a simple way of looking at it
which has been hard to understand the last year because
of the generative AI movement and it's all the chatbots
and the foundation models and the lms, and I think
for us, we really try to talk about being a
predictive AI company versus a generative AI company, and we
(15:41):
use generative AI and lllms, but there's inputs to predictive
AI algorithms. And I think a marketer look having a
front end put on and accessibility measures put on lms
and making them easier to use for certain things from
a marketing analysis perspective or planning perspective.
Speaker 3 (16:01):
Hey, have at it.
Speaker 5 (16:02):
That's great.
Speaker 4 (16:03):
There's going to be a million of those companies and
you just try the one that works best for your company.
But predictive AI is big data AI, and it's much
harder to do and you need your own tech to
do that. And so that's really I think you're going
to see more of that differentiation in how people position
themselves in the next couple of years, predictive.
Speaker 5 (16:22):
AI versus generative AI tools.
Speaker 3 (16:26):
Let me switch gears to you know, another one of
the cocktail party words that we hear a lot, first
party data. It's always mattered. It doesn't just now matter,
It's always mattered. And at the same time, you know,
many marketers are asking do we have the right data
and are we using it well? And I guess i'd say,
what's changed in the last couple of years, especially with
(16:48):
LMS that lets marketers you know, get more value out
of that data.
Speaker 2 (16:53):
Yeah, I think, like you said, first party data, I
think has always been valued or at least you know,
has been useful for me years. But I think there
has been a shift. It relates to what I was
saying before which is previously machine learning and add tech
was a lot of number crunching, and so it could
get some value out of your first party data, but
it didn't really understand like let's say you had a
(17:16):
list of people who bought a product. The AI didn't
or the machine learning didn't understand what product you're selling
or have any context for that. It's just a list
of here's the people. And now with foundation models, they
bring all of this understanding to the table, and that's
what allows us and like the others to get more
out of that first party data because you're combining that
(17:38):
first party data, which is you know who bought what,
who clicked on what, with all of this understanding, so
that rather than number crunching, you're kind of understanding that
the AI or that machine learning can really understand that
data in a more fundamental way.
Speaker 3 (17:52):
Both you and I, all three of us get the
chance to speak to cmos frequently. One of the conversations
that I've been interested in is how cmos are genuinely
excited about the opportunity to combine sort of foundation models
with their own data, but also everybody is conscious and
(18:13):
nervous about privacy as we have to be How do
you think brands can tap the value while keeping their
most sensitive information which they have to protect it and
under their control. I think you said it right.
Speaker 2 (18:27):
It's like this, Like you said, the CMO has to
strike that balance, right. They have to protect their data,
but they also want to take advantage of this moment
and get more out of their data.
Speaker 3 (18:36):
There's a line you have.
Speaker 5 (18:37):
To walk there.
Speaker 2 (18:38):
You need to if you're working with a partner, you
need to understand what they.
Speaker 5 (18:41):
Are allowed or not allowed to do with that first
party data.
Speaker 2 (18:44):
For example, we use first party data and combine it
with knowledge from foundation models, but we never share that
data with a foundation model provider like open AI. We
do that in house. You know others might not do that,
which might maybe something you don't want to do. So
it's a question of following the data, where is it
going to go if you share it, and making sure
(19:07):
you're working with partners who respect the importance of that data.
You don't want that data being the next set of
data that goes to train a foundation model, right. Foundation
models trained on public data. First party data is private data.
You've got to make sure it doesn't go into those models.
And the question is how do you do that critically important?
Can we switch gears to context GPT. It's been in
(19:28):
the market for a couple of years, but as a
first of its kind product for someone hearing about it
for the first time, how do you describe what it
actually does for a brand?
Speaker 4 (19:38):
Context GPT is the first contextual targeting tool that allows
you to very easily target at scale an idea or
an audience that you're looking for without IDs. And so
we have two different types of advertised there's these days
(20:00):
that work with us on context GBT. First of all,
it's our fastest growing product. It's sales or skyrocketing because
you use a PROMPT to put in what you're looking for.
We were the first in market with that. But now
you can use this world knowledge that Aaron's been talking
about to surround an audience. So we have again two
(20:22):
types of advertisers. We have sports apparel advertisers who want
to surround the World Cup everything about World Cup they
want to be on, or the Super Bowl or marathons
things like that. Or you have CpG companies like soda
manufacturers who want to be around gen z fun concert
(20:44):
content right or not concert content just gen z right,
they want to surround gen Z, but they want it
to be fun content.
Speaker 5 (20:51):
How do you find that?
Speaker 4 (20:53):
The old way of doing it, you had to know keywords,
you had to understand who are the gen Z music
stars that people are listening to, et cetera. But now
as combining it with foundation models like open AI, which
context GPTs is integrated with, we can fine tune all
of that and deliver topics and segments of URLs based
(21:20):
on the entire content of the page using brand safety
measures like you know, is it is it a negative
review of a product?
Speaker 5 (21:30):
Is it?
Speaker 4 (21:33):
You know, violence and drugs and things like that, And
now you can find every single RL in real time
as it comes online and target it with your ads.
Speaker 2 (21:44):
I mean, I think if it like I think it's
obvious if you use chat GPT that computers just understand
text and images better than ever before. And context GPT
was saying, hey, that's a powerful thing for contextual targeting.
You type in what you want and the content and
LLLM can understand that page and answer whether this page
(22:05):
meets your criteria. So you just type in your criteria
and we'll figure out whether each r L fits that criteria.
And I think another thing we did differently that I
think has been one reason the product has been so
successful is it's radically transparent. You type in your prompt
and we show you every ur L that we think
matches that prompt, and like how relevant it is.
Speaker 5 (22:25):
People were used.
Speaker 2 (22:26):
To like, you pick some taxonomy category like sports and
you have no idea what's in it, And here you
type in a customized prompt and we tell you exactly
what's in it. And I think that that's another reason
it's been popular.
Speaker 3 (22:39):
I love that because what I learned last year a
joke about this frequently. Megan Trainor recorded that great song
It's all about the bass, so she taught us that.
What I learned from Jensen Wang in a conversation was
it's all about the prompt, so you know, yeah, yeah,
bringing the prompt to life. The other big buzz that
(23:00):
came out of early January at CES, which was a
gentic Ai kind of a new agency darling. But I
think there are instances where the term is getting misused wildly.
So what I'd love for you guys to do is,
you know, clear it up. Give me a sentence on
agentic Ai, maybe even two sentences.
Speaker 2 (23:21):
I had a recent addicts Changer piece where I gave
sort of my perspective on where agentic AI is. So
you want the details, go read that. But I think
the one sended summary is sort of right now agentic
AI or there's also this term ad CP. I think
their position to sort of make buying easier at least
over the next year, but I don't think they're really
(23:43):
in a position to make buying smarter or better yet.
And so that's sort of the one sended summary, and
then maybe I could add some color there. So the
term agent, and with respect to machine learning, kind of
comes from the seven these this this whole field of
machine learning called reinforcement learning, and it's actually the idea
(24:06):
of an agent is it's a machine learning tool that
can go out there and optimize something like a KPI.
If we're using advertising lingo, right, We're going to go
optimize a KPI and the agent is going to do
it through trial and error and like figure out how
to maximize this thing by through machine learning. That's how
that's what an agent really means. And today that term
(24:29):
is kind of getting redefined to mean something much more
basic in my opinion, which is today an agent means
asking a chatbot to take some action, Like if you
ask a chatbot to book you a reservation, that's an agent, right,
That's a very different and much simpler kind of agent.
(24:49):
That kind of agent's great for convenience or like, you know,
making something simple. Hey, I want to set up a strategy,
but I don't want to do it myself with twenty clicks.
I just want to tell alem to go set up
a strategy. But that sort of LLM definition of an
agent is not well equipped, at least this year and
probably for the next few years to really optimize an
(25:10):
ad campaign.
Speaker 3 (25:11):
What got me nervous growing up in the entertainment business
and having recent experience with agents, you know, not so good.
Sod about that those of our listeners who who know?
If you know, you know, But it's just funny. I
wish there were another word for agentik.
Speaker 2 (25:28):
I think these agents are cheaper though, yeah, I mean
I don't want, I don't want. Agents are getting better.
Chat Chept was a huge moment last November. Claude of
Anthropic released their new claud code tool from the developer side.
That's another chatchept moment. People are just amazed at how
(25:49):
good that is at producing code for you. Agents are
cool and they're very useful, but and you know, I
had bad experiences with them a year ago, and they're
going to keep getting better, but they're far from being
able to look at billions of rows of impressions and
knowing which one was viewed and which one is clicked.
(26:09):
You know, their their language tools, they think through through language.
They don't they're not really number tools yet.
Speaker 3 (26:15):
So so let me ask a question. You know, we
all understand there's got to be this human plus AI,
you know, collaboration, if for nothing else, kind of to
battle against another often used term recently AI slop, you know,
and and what what there is a rise in concern
I think with good reason around this AI slop. How
(26:39):
do you design tools so that the human stays firmly
in the loop.
Speaker 4 (26:42):
Can I just add something I think on that on
that topic about agents, because I think that's that's the
thing that people seem to miss about the agent to
agent discussions that are happening in the advertising business. They're
talking a lot about Hey, I have a buying agent,
and you're going to have selling agents, and they're going
to talk back and forth and then the buying agent's
(27:04):
going to decide what to do. I think that we
all know in the media business that that seems a
little crazy.
Speaker 5 (27:12):
Every single brand.
Speaker 4 (27:15):
Of seller, let's say publisher, let's just easily call them publishers.
They stand for something in the marketplace. They have to
differentiate themselves. A lot of that has to do with
human creativity and understanding deeply what a brand is and
pushing that into some kind of automated sort of handshake
(27:35):
back and forth. I think is going to be problematic
for a lot of buyers and sellers because of the
ability to manipulate a buying agent in this way. So
I'm interested to see how this moves forward. Besides, something
across commodities in the advertising space.
Speaker 3 (27:53):
Makes good sense. Let's talk about kind of looking ahead
outside of advertising. What area of AAR are you watching closely?
And Aaron, I would throw this your way first, and
I would say, if we were to replay this episode
in twenty four months, is there one prediction about AI
and marketing that you would be comfortable being held to?
(28:15):
So I'm asking you two questions there.
Speaker 2 (28:18):
Yeah, yeah, what I'm paying more attention to is related
to your AI slop question, which is when it came
to the ability of models to sort of reason about
people and their intents.
Speaker 5 (28:34):
A year ago.
Speaker 2 (28:36):
You know, we've been exploring this for a while. There
was a lot of AI slop we've seen over the
last year, sort of a leap forward in terms of
a model's ability to look at a series of behavioral
signals and reason about what the intent of that person
might be from a purchasing perspective or maybe from other perspectives.
And I think that is a meaningful There's still slop there,
(29:00):
but there's potential there too, and that's what we're leading into,
and I'm paying a lot of attention into, like how
reasoning models are going to allow us to make better
performance advertising tools that don't rely on as much first
party data, or at least allow us to get more
(29:21):
out of sort of data list pixel less solutions.
Speaker 3 (29:24):
We're going to hit pause for a moment, but stay
with us after the break. We've got more insights to share. Jeremy,
if I take this away from tech for a moment
(29:45):
and talk about leadership and talent in a volatile market.
I mean, we're in a volatile market both from the
talent perspective and as well the competitive landscape. But Aaron
said it at the beginning at cognitive, how do you
as a leader keep that culture, collaborative, fun enjoyable in
a category that is intense and fast moving and highly competitive.
(30:10):
I'm just curious what are the kind of secrets you
can share with our audience about how you, as a leader,
you know, navigate this interesting time in the market.
Speaker 5 (30:21):
For sure.
Speaker 4 (30:23):
Yeah, I think that for us we're on the upside
of AI and transformation from an internal perspective, not just
from our tool set for marketers, but from our tooling
as a company. And I think that's really giving us
an advantage. So Amazon just laid off sixteen thousand more people.
Speaker 3 (30:44):
Saw that as and then it's on top of fourteen
thousand in December to look at that and say thirty
thousand people within the space of a month, which those
are real numbers.
Speaker 4 (30:55):
Yeah, and they are very transparent about it being AI
efficiency for us, we're using it on the other side,
instead of needing three hundred, five hundred thousand people, we
need one hundred and fifty to two hundred people to
do what three hundred, four hundred, five hundred people we're
going to do. I think two three years ago, and
I think that is a big advantage for us and
(31:18):
companies that have come up at this time versus companies
that are trying to retool large organizations in a different way.
Speaker 3 (31:27):
So if you don't start with a yoke around your neck,
you've got an ability to build for the current, not
for rebuild from the past. Let me let me, let
me talk about that. In terms of collaboration, you work
with partners like Magnet, Index, Marketplace, open Ai, others. Where
do you see the greatest opportunity to simplify the path
(31:47):
from sort of hey I have an idea to actually
activating it.
Speaker 4 (31:52):
Well, I think we've seen an acceleration there. We work
in advertising technology. It's probably the most competitive and the
fastest evolving industry. I think out there period what Index
did last year, Magnite and open ai, open x and
all these other guys, Pubmatic are trying to copy, and
then somebody will come up with a new idea and
(32:13):
everyone else we'll pile on that. Because they're trying to
gain an advantage on a constant basis around something that
is becoming they're trying to stop from becoming commoditized. So
we work very quickly and efficiently with our partners, especially
where we know that AI needs this extra thing right
(32:37):
and with Index we're integrated and co located in their
server system. They've been working hard on containers, which is
great for some companies, but for us, we want to
be able to run full neural network models and we
can't use a container. We have to have our own
servers and our own systems and our own datas databases
(32:58):
all together in one place so that we can in
real time predict this consumer behavior that we're talking about.
We worked very hard with Index about that, working with
Magnete on that, and we're helping them understand what that
next generation of modeling is going to require. And I
think they've been really responsive to that.
Speaker 3 (33:21):
Well, look, you know, none of us go through this alone,
none of us get there alone. So the idea, Jeremy,
and I've seen this with you over the years, and
I've watched that, and I've watched your ability to be
a collaborator, and you know, I see it in both
how you operate but also how you internally but how
you operate externally with partners and the collaboration. So I
(33:43):
commend you on that, guys. I come to my favorite part,
my lightning round. So I'm going to throw some questions
at you.
Speaker 2 (33:50):
We'll do our best.
Speaker 3 (33:51):
Aaron, I'll start with you. What's your greatest professional fear?
Speaker 2 (33:55):
Oh, professional fear As an academic, I don't want to
be wrong, you know. I feel like you're wrong once
and your trust is forever hurt. So I like to
be careful in my wording because I think I fear
undermining my credibility.
Speaker 5 (34:12):
That's it. That's it.
Speaker 3 (34:12):
That's a good one. Jeremy is there. This is good
because we've touched on so many in this conversation. Is
there one industry buzzword you wish would disappear forever?
Speaker 5 (34:23):
Last click attribution?
Speaker 3 (34:26):
Good? Okay, If you had twenty four hours, this is
a jump ball.
Speaker 5 (34:30):
Either one of you grab it.
Speaker 3 (34:31):
If you had twenty four hours with no restrictions, what's
one adventure you would embark on.
Speaker 2 (34:36):
Oh, I would like to summit a peak, you know,
I'd like, I'm gonna get out of town. I want
to get to the top of some mountain. Not Everest
too hard, something more reasonable. But I really enjoy getting
to the top of a mountain and looking out. And
if I had twenty four hours, I would go do that.
Speaker 3 (34:54):
This is for both of you. Was there a particular
mentor early in your career? And you know, like I
heard at the beginning of our conversation, I think, Aaron,
you said it, and it's so important in relationships. You said,
we knew each other's parents. I love that because you
understood your Mark and Jeremy through a different lens. Okay,
(35:15):
you knew their parents. That was very telling to me.
I hope our audience heard that as well. But did
you have a mentor either one of you in your
life that early that gave you a bit of advice that.
Speaker 5 (35:27):
You know?
Speaker 2 (35:28):
For me, it was actually my PhD advisor, Michael Fee.
He was a physicist by training and he and he
spent a lot of one on one time with me
just teaching me a science, the scientific method, how to think,
how to go back to fundamentals. He just taught me
(35:48):
if you go back to physics, if you go back
to the scientific method, you will make progress. You know,
the goal of you know, academias sort of pushed the
envelope of knowledge, and he really taught me how to
do that in Jeremy.
Speaker 4 (36:00):
For me, my first real mentor ended up being a
woman who I worked with. She was a partner on
a project I was on at Digitas. That was sort
of my first job in marketing and interactive marketing and advertising.
Speaker 3 (36:16):
But by the way, Jeremy, I think that's where we
first met.
Speaker 4 (36:19):
Actually, I mean, it was a long time ago, and
it was a great time to be at Digitas. I
can't that was a great That was a great company
for me to really grow up in marketing on before
I went to business school. But we were on this
big project for Morgan Stanley. We were building this huge
interactive dashboard for them back in you know, early two
(36:40):
thousands when this was the newest, newest thing, and I
had been brought on to be an assistant development manager.
Speaker 5 (36:47):
This was back when I was still coding.
Speaker 4 (36:50):
And the manager got fired from the project because we
missed a bunch of deadlines and it just didn't work out.
And I was thrust into this manager role when I
was like twenty five four years old, twenty three years old,
and I didn't know anything about management. And I think
everybody who goes into management with no management training does
things exactly the way they would do things, and that
(37:11):
is not a good way to manage a large group
of people.
Speaker 5 (37:14):
It was a hard it was a hard learning.
Speaker 4 (37:18):
Experience for me, but my boss, my partner on that project,
sat me down and really taught me a lot of
very fundamental things about management, one of them being, Hey,
you have to talk to everybody individually. You have to
understand them individually and what motivates them individually. And that
was the beginning of my sort of career as a manager.
And it was She was by far the most helpful
(37:40):
to me in that time when I had no idea,
and I always look back on that as being a huge,
a huge.
Speaker 5 (37:48):
Turning point for me.
Speaker 4 (37:49):
It could have gone really badly, and she really helped me. Guys,
you've really helped me today. You've clarified a couple of
things in this world of jargon that I think brought
clarity for me, and I think it will as well
for our audience. I want to thank you Aaron and
Jeremy for taking the time, Thank you for joining me
(38:11):
on Good Company.
Speaker 5 (38:12):
Thanks for asking us.
Speaker 3 (38:14):
Yeah, thank you.
Speaker 2 (38:15):
I enjoyed the conversation.
Speaker 3 (38:21):
I'm Michael Casson, thanks for listening to Good Company.
Speaker 1 (38:25):
Good Company is brought to you by Three C Ventures
and iHeart Podcasts Special thanks to Alexis Borgerer Pudeo, our
executive producer and head of Content and Talent, and to
Carl Ketle, executive producer.
Speaker 5 (38:37):
At iHeart Podcasts.
Speaker 1 (38:39):
Episodes are produced and edited by Mary Doo. Thanks for
joining us.
Speaker 5 (38:44):
We'll see you next time.
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
Kingdom of Fraud
It’s the unlikeliest of criminal partnerships: a devout polygamist from an insular Utah sect joining forces with a shadowy Armenian tycoon from LA. The result - a billion dollar fraud conspiracy. In Kingdom of Fraud, investigative reporter Michele McPhee traces the origins of the extraordinary alliance between Jacob Kingston and Levon Termendzhyan. Together, the two men trigger the largest tax investigation in American history and weave around themselves a web of dirty cops, influential political relationships and transnational money laundering. All this is set against the backdrop of Jacob Kingston’s clan – The Order. A powerful and secretive polygamist organization in Salt Lake City. To whom Jacob is desperate to prove his worth. Kingdom of Fraud is produced by Novel for iHeart Podcasts. For more from Novel, visit https://novel.audio/. You can listen to new episodes of Kingdom of Fraud completely ad-free and 1 week early with an iHeart True Crime+ subscription, available exclusively on Apple Podcasts. Open your Apple Podcasts app, search for “iHeart True Crime+, and subscribe today!