Beyond ROAS: The Signal Behind Smarter Ad Spend with AdZeta's CEO

Behind the code.

Behind the code. [singing]

Behind the code. [singing] That's where the [music] future grows.

That's where the [music] future grows. Welcome to Behind the Code presented by

Welcome to Behind the Code presented by Atomic Build. I am Emit. And with me

Atomic Build. I am Emit. And with me today is Abby, the CEO of Adzetta.

today is Abby, the CEO of Adzetta. Really happy to have you here, Abby.

Really happy to have you here, Abby. Thanks for joining us. Instead of giving

Thanks for joining us. Instead of giving your intro, I'd love to hear an intro

your intro, I'd love to hear an intro about yourself and your company.

about yourself and your company. Yeah, fantastic. Thanks for having me,

Yeah, fantastic. Thanks for having me, Amit. Uh, yes. So, I am Abi Agnihotree.

Amit. Uh, yes. So, I am Abi Agnihotree. I'm the CEO of Adseta. right now I mean

I'm the CEO of Adseta. right now I mean I've been I've been in growth marketing

I've been I've been in growth marketing for the last sort of 10 years or so. Uh

for the last sort of 10 years or so. Uh I worked in enterprise growth marketing

I worked in enterprise growth marketing roles. I also worked on the I started

roles. I also worked on the I started from the agency side pivoted to

from the agency side pivoted to enterprise growth marketing. Then I

enterprise growth marketing. Then I launched my own performance marketing

launched my own performance marketing agency. Uh and back in 2023 I launched.

agency. Uh and back in 2023 I launched. So during this time I've managed close

So during this time I've managed close to $150 million in ad spend and I kept

to $150 million in ad spend and I kept on seeing the same problems that

on seeing the same problems that restrict the quality of customers that

restrict the quality of customers that these brands acquire and that's what the

these brands acquire and that's what the premise of the conversation is going to

premise of the conversation is going to be. So excited to be here.

be. So excited to be here. Amazing. Yeah. That's a very topical

Amazing. Yeah. That's a very topical introduction. Described your company

introduction. Described your company really well and what you do and um you

really well and what you do and um you mentioned you were managing about 150

mentioned you were managing about 150 million under paid acquisition and

million under paid acquisition and financial services and other categories.

financial services and other categories. I was thinking about that and I was

I was thinking about that and I was wondering like what did you learn in

wondering like what did you learn in that experience that a lot of marketers

that experience that a lot of marketers in-house might never get to see. So

in-house might never get to see. So during my time running you know these

during my time running you know these large scale paid media acquisition

large scale paid media acquisition campaigns the most standout thing that

campaigns the most standout thing that came out to me is that these ad

came out to me is that these ad platforms are remarkably good at

platforms are remarkably good at optimizing what you give them to. So

optimizing what you give them to. So Meta, Google, Tik Tok these are app like

Meta, Google, Tik Tok these are app like I would add apploving to that piece as

I would add apploving to that piece as well. These are one of the most

well. These are one of the most sophisticated pieces of machine learning

sophisticated pieces of machine learning ever assembled by mankind and they are

ever assembled by mankind and they are really good at doing what they do with a

really good at doing what they do with a massive asterisk. It's dependent and

massive asterisk. It's dependent and it's a direct function of what you feed

it's a direct function of what you feed them as a value. If you feed them

them as a value. If you feed them standard purchase level events, they

standard purchase level events, they will go and find out those people who

will go and find out those people who will you know create that purchase. we

will you know create that purchase. we will maybe offer like you know optimize

will maybe offer like you know optimize for a sign up you know these sort of

for a sign up you know these sort of shallow level events quote unquote the

shallow level events quote unquote the problem is input right so if you are

problem is input right so if you are feeding these shallow events if you go

feeding these shallow events if you go and find out but it will not look into

and find out but it will not look into the long-term picture and the long-term

the long-term picture and the long-term profitability like that's what I saw all

profitability like that's what I saw all the time repeatedly across you know

the time repeatedly across you know across several several years and I also

across several several years and I also see a massive over reliance on creative

see a massive over reliance on creative and right now it's almost getting

and right now it's almost getting demoditized to the to the fact that it's

demoditized to the to the fact that it's becoming a detriment to performance

becoming a detriment to performance because Every B brand with the use of AI

because Every B brand with the use of AI can spin up multiple creatives they can

can spin up multiple creatives they can push them in the platform. The platforms

push them in the platform. The platforms will respond to that. It will obviously

will respond to that. It will obviously be like a vicious flywheel where more

be like a vicious flywheel where more critters get thrown it increases CPMs

critters get thrown it increases CPMs the CPCs and again it's you know it just

the CPCs and again it's you know it just erodess the quality and just you know it

erodess the quality and just you know it it's having a detrimental impact and

it's having a detrimental impact and that's why I hear from all the brands

that's why I hear from all the brands that they're having scalable problems

that they're having scalable problems and that's a function of what and I I

and that's a function of what and I I learned that during my time as well but

learned that during my time as well but now it's getting more aggravated and

now it's getting more aggravated and then lastly I would say the measurement

then lastly I would say the measurement piece where the brands are mostly

piece where the brands are mostly reliant on rorowass as opposed to

reliant on rorowass as opposed to contribution margin or bottom line

contribution margin or bottom line performance uh because rorowass can

performance uh because rorowass can inflate and actually be quite deceptive

inflate and actually be quite deceptive in terms of what you know what you they

in terms of what you know what you they they want you to see. So those are the

they want you to see. So those are the three main learnings and of course since

three main learnings and of course since this in this one decade I was sitting at

this in this one decade I was sitting at almost the midpoint of this was when the

almost the midpoint of this was when the sort of zero interest rate policy was

sort of zero interest rate policy was where people were just throwing money at

where people were just throwing money at the problem. It was easy to scale and

the problem. It was easy to scale and people were just like getting in that

people were just like getting in that sort of wrong habit. Now with the

sort of wrong habit. Now with the non-zor era, people have to be a bit

non-zor era, people have to be a bit more cognizant. Investors require more

more cognizant. Investors require more uh you know attention when it comes to

uh you know attention when it comes to growth numbers and yield economics and

growth numbers and yield economics and those numbers are now coming back to

those numbers are now coming back to bite them. So those wrong habits have

bite them. So those wrong habits have curtailed way too long. But we do see a

curtailed way too long. But we do see a trend whereby people are shifting

trend whereby people are shifting towards profitability and like long-term

towards profitability and like long-term growth as opposed to just grow at all

growth as opposed to just grow at all costs.

costs. Got it. So you mentioned one thing like

Got it. So you mentioned one thing like shallow versus maybe like a more deep

shallow versus maybe like a more deep cycle and and how how would you describe

cycle and and how how would you describe the two differences like and when you're

the two differences like and when you're meeting with or have a customer coming

meeting with or have a customer coming on board how do you break that cycle? So

on board how do you break that cycle? So let's say they've been doing that now

let's say they've been doing that now for the better part of Zer to today

for the better part of Zer to today let's say for example how do you kind of

let's say for example how do you kind of define to them to break out of that

define to them to break out of that shallow cycle and moving into what you

shallow cycle and moving into what you would say is a better investment in this

would say is a better investment in this channel. So what we do is traditionally

channel. So what we do is traditionally we we ask for what outcome they want to

we we ask for what outcome they want to achieve and reverse engineer from that

achieve and reverse engineer from that rather than trying to impose ourselves

rather than trying to impose ourselves on like whatever like we want them to do

on like whatever like we want them to do or we position ads as so the great a

or we position ads as so the great a common way to look about this is like

common way to look about this is like are the cohorts of customers that you're

are the cohorts of customers that you're acquiring right now making sense for you

acquiring right now making sense for you in the long term is long-term

in the long term is long-term profitability a metric for you and

profitability a metric for you and lastly do you have enough sufficient v

lastly do you have enough sufficient v do you have sufficient variance in your

do you have sufficient variance in your customer cohort to actually benefit from

customer cohort to actually benefit from what we do these three questions are the

what we do these three questions are the three pillars in which the initial

three pillars in which the initial conversations go and if the answers to

conversations go and if the answers to either two of them is even like two of

either two of them is even like two of them out of the three is yes then that's

them out of the three is yes then that's a good way for us to kind of go and like

a good way for us to kind of go and like proceed with some sort of a sort of a

proceed with some sort of a sort of a follow-up. If they're only looking at

follow-up. If they're only looking at sort of the next sale and the next sale

sort of the next sale and the next sale and the next sale and like you know they

and the next sale and like you know they they don't have enough variance in the

they don't have enough variance in the customer portfolio then we need to be

customer portfolio then we need to be cognizant of our resources as well as

cognizant of our resources as well as theirs because in this era where trust

theirs because in this era where trust is so scarce like you need to be upfront

is so scarce like you need to be upfront with the brands what you can deliver and

with the brands what you can deliver and what you cannot and we do that in the

what you cannot and we do that in the very first conversations where we do

very first conversations where we do this analysis and tell them that with

this analysis and tell them that with your data with your compos uh data sort

your data with your compos uh data sort of distribution in your customers this

of distribution in your customers this is what we can do this is what we cannot

is what we can do this is what we cannot do so that we have realistic sort of

do so that we have realistic sort of expectations and we move from Great. So,

expectations and we move from Great. So, do your do your potential clients ever

do your do your potential clients ever disqualify themselves based on those

disqualify themselves based on those like like those three structured

like like those three structured questions that you have from the onset.

questions that you have from the onset. So, before going to any meetings, we

So, before going to any meetings, we already know a lot about their about

already know a lot about their about their customer population based on

their customer population based on publicly available resources, their

publicly available resources, their business model. Like you you can't hide

business model. Like you you can't hide that, right? So, it's usually a question

that, right? So, it's usually a question of like the data distribution. So once

of like the data distribution. So once they're able to share the data and

they're able to share the data and there's a you know not enough evidence

there's a you know not enough evidence that it wants to be it's going to be a

that it wants to be it's going to be a major lift then we are able to be like

major lift then we are able to be like we are we are very sort of transparent

we are we are very sort of transparent in that piece and we tell them that this

in that piece and we tell them that this is not the right fit right now and maybe

is not the right fit right now and maybe you needed more various and they it's

you needed more various and they it's not and that's and then it's a question

not and that's and then it's a question for the brand whether they want to

for the brand whether they want to progress or not but usually when we talk

progress or not but usually when we talk to brands more often than like 85 90% of

to brands more often than like 85 90% of the time we know exactly what problems

the time we know exactly what problems they're facing exactly what their

they're facing exactly what their business model is and if it's not the

business model is and if it's not the case like we are reluctant from our side

case like we are reluctant from our side even taking the conversations and taking

even taking the conversations and taking the calls and we tell them that this is

the calls and we tell them that this is not the right fit for what the stage

not the right fit for what the stage you're in. So it's a two-way street and

you're in. So it's a two-way street and the brands appreciate that.

the brands appreciate that. Right. Yeah. That that makes so there's

Right. Yeah. That that makes so there's quite a bit of pre-work before a

quite a bit of pre-work before a conversation and like because you've

conversation and like because you've been in the space preI era and now AI

been in the space preI era and now AI era. Do you have like a research

era. Do you have like a research methodology you're using through an

methodology you're using through an agent or maybe like an LLM that helps

agent or maybe like an LLM that helps you get to this place? And you don't

you get to this place? And you don't have to share um the proprietary

have to share um the proprietary methodology, but I'd love to hear like

methodology, but I'd love to hear like how you spun that up if if you're using

how you spun that up if if you're using one.

one. So we are using internally in our SAS

So we are using internally in our SAS tool in some of the work which obviously

tool in some of the work which obviously can be identified. A lot of the stuff

can be identified. A lot of the stuff cannot be identified right now because

cannot be identified right now because of the stakes, the high stakes and like

of the stakes, the high stakes and like you know the latency and a few other

you know the latency and a few other things. So we're very cognizant in terms

things. So we're very cognizant in terms of not not just using AI for the sake of

of not not just using AI for the sake of it. We are bootstrapped so we have low

it. We are bootstrapped so we have low investor pressure and I'm the only one

investor pressure and I'm the only one taking the decision decisions based on

taking the decision decisions based on whether it actually adds tangible value.

whether it actually adds tangible value. A lot of people not just in our space

A lot of people not just in our space but just in general they're just using

but just in general they're just using it because of outside pressure a lot of

it because of outside pressure a lot of other things. For me it's the outcome

other things. For me it's the outcome and then the reverse engineer where the

and then the reverse engineer where the AI can actually solve that purpose or

AI can actually solve that purpose or not for us we are using in certain cases

not for us we are using in certain cases where we can like actually churn out the

where we can like actually churn out the mundane processes and pre-qualification.

mundane processes and pre-qualification. Yes. And but in in the case where when

Yes. And but in in the case where when you think about sensitive client

you think about sensitive client information right when you're talking

information right when you're talking about customer data a lot of that can be

about customer data a lot of that can be of course you know we can s anomize a

of course you know we can s anomize a lot of that stuff from agents and uh but

lot of that stuff from agents and uh but a lot of the stuff cannot be identified

a lot of the stuff cannot be identified right now. So we are really particular

right now. So we are really particular about like which pieces the agents are

about like which pieces the agents are touching and which they aren't. So

touching and which they aren't. So there's a conscious flow there to put it

there's a conscious flow there to put it to put it that way.

to put it that way. Right. So like you would say when you

Right. So like you would say when you say like a mundane I guess it's like

say like a mundane I guess it's like routine tasks that you would do through

routine tasks that you would do through a search or a scrape or something like

a search or a scrape or something like that that's okay but then maybe deeper

that that's okay but then maybe deeper level analysis

level analysis customer data

customer data right exactly so handling customer data

right exactly so handling customer data handling customer orders at that level

handling customer orders at that level of scale at that level of scrutiny we

of scale at that level of scrutiny we are we are compliant with sock too and

are we are compliant with sock too and like we take privacy very seriously so a

like we take privacy very seriously so a lot of these things right now are to

lot of these things right now are to that stage where we use AI blindly for

that stage where we use AI blindly for these and same goes for agents as well

these and same goes for agents as well because it's agents are just built on

because it's agents are just built on top of these foundational models. So we

top of these foundational models. So we take that very seriously and all of

take that very seriously and all of those are like you know in a tight cloud

those are like you know in a tight cloud ecosystem but we do like once that sort

ecosystem but we do like once that sort of you know once that data piece and the

of you know once that data piece and the custom piece is taken away a lot of the

custom piece is taken away a lot of the modeling piece can be identified of

modeling piece can be identified of course machine learning has been around

course machine learning has been around for 30 years so we don't rely on a lot

for 30 years so we don't rely on a lot of machine learning and less AI but a

of machine learning and less AI but a lot of that stuff can be identified in

lot of that stuff can be identified in terms of you know the normalization of

terms of you know the normalization of the data and a lot of the other stuff

the data and a lot of the other stuff which

which Got it. Yeah, I think that's a I think

Got it. Yeah, I think that's a I think that's a very good approach especially

that's a very good approach especially when it comes to protecting your

when it comes to protecting your customers data and like any proprietary

customers data and like any proprietary information flowing through your system.

information flowing through your system. One thing just like when we were

One thing just like when we were chatting before and doing some just

chatting before and doing some just research on what you do, you you kind of

research on what you do, you you kind of predicted that signal would be an

predicted that signal would be an overused term um but maybe not well

overused term um but maybe not well understood in adte. Can you just kind of

understood in adte. Can you just kind of go deeper on that and explain what you

go deeper on that and explain what you meant? Yeah, it it it's one of my

meant? Yeah, it it it's one of my favorite topics to discuss at the same

favorite topics to discuss at the same point it edges at at the surface of

point it edges at at the surface of being a pet peeve because everyone talks

being a pet peeve because everyone talks about signal and it's theoretically or

about signal and it's theoretically or like you know if you think about they

like you know if you think about they have a rough notion when they think

have a rough notion when they think about signal or people think about

about signal or people think about signal it's some form of data coming in

signal it's some form of data coming in and data going out roughly on a very

and data going out roughly on a very high level but it can mean a lot of

high level but it can mean a lot of things and all of them have different

things and all of them have different orders of magnitude in terms of the

orders of magnitude in terms of the performance they drive the lift they

performance they drive the lift they drive and it's fine for vendors because

drive and it's fine for vendors because they can sell it but as a buyer if I was

they can sell it but as a buyer if I was a buyer of a of a consumer brand. It

a buyer of a of a consumer brand. It would be a nightmare for me to

would be a nightmare for me to distinguish how they how they are

distinguish how they how they are fundamentally different one from one

fundamentally different one from one another. So it can be broadly classified

another. So it can be broadly classified into five types of signals. You know you

into five types of signals. You know you can talk about creative signals which

can talk about creative signals which are the most overused right now and

are the most overused right now and people are talking about creative

people are talking about creative signals and like you know Andromeda and

signals and like you know Andromeda and meta and a whole host of things where

meta and a whole host of things where how you talk about that retrieval stage

how you talk about that retrieval stage of ad ranking or not ranking but

of ad ranking or not ranking but retrieval stage of the ad ecosystem.

retrieval stage of the ad ecosystem. Then you think about the the measurement

Then you think about the the measurement signal. So the measurement signals would

signal. So the measurement signals would be your typical MTA, MM incrementality.

be your typical MTA, MM incrementality. How you measure your performance based

How you measure your performance based on backward level criteria. So you do

on backward level criteria. So you do it's a reporting problem a measurement

it's a reporting problem a measurement problem. Then there's an attribution

problem. Then there's an attribution signal which is how do you actually make

signal which is how do you actually make sure that the offline the to online and

sure that the offline the to online and maybe throw retail in the mix to some

maybe throw retail in the mix to some degree that coincides with your media

degree that coincides with your media spend and what part of portion of ad

spend and what part of portion of ad spend is actually driving as how what's

spend is actually driving as how what's the attribution piece. So that's the

the attribution piece. So that's the attribution signal. Then you have the

attribution signal. Then you have the audience signal which is the classic

audience signal which is the classic CDPs which I've been doing for 10 years.

CDPs which I've been doing for 10 years. CDBs have been doing that for a good

CDBs have been doing that for a good part of the last decade or so where they

part of the last decade or so where they build some sort of a they take the first

build some sort of a they take the first party data they build some sort of a

party data they build some sort of a customer list and you can pass that as a

customer list and you can pass that as a as an audience layer into the app

as an audience layer into the app platforms and then you can build look

platforms and then you can build look likes off of them based on those

likes off of them based on those snippets or those package of segments

snippets or those package of segments right that's what CDPs are doing the

right that's what CDPs are doing the last one is the bidding signal which is

last one is the bidding signal which is how do you send a forward-looking value

how do you send a forward-looking value to the platform so that it can actually

to the platform so that it can actually operate on that level where it looks

operate on that level where it looks into the future and like predicts which

into the future and like predicts which users are going to be the most valuable

users are going to be the most valuable users. So the first four signals are all

users. So the first four signals are all to some degree backward looking and they

to some degree backward looking and they provide some sort of you know lift which

provide some sort of you know lift which is like I would say single to like low

is like I would say single to like low double digits. When you look at the

double digits. When you look at the bidding signal when you're sending and

bidding signal when you're sending and improving the auction itself, we have

improving the auction itself, we have seen up to 60% increase in ROS which is

seen up to 60% increase in ROS which is a magnitude which is very hard to you

a magnitude which is very hard to you know you know quantify like you know

know you know quantify like you know sorry measure or like you know match

sorry measure or like you know match from when you look at other levels of

from when you look at other levels of especially audience because a lot of the

especially audience because a lot of the vendors are working in that audience

vendors are working in that audience signal space and that that can be very

signal space and that that can be very easily confused with the bidding stream

easily confused with the bidding stream which is fundamentally different in

which is fundamentally different in audiences you don't need to take care of

audiences you don't need to take care of latency you don't need to take care of a

latency you don't need to take care of a whole host of things which come with the

whole host of things which come with the bidding

bidding So that's why the risk reward obviously

So that's why the risk reward obviously is much more the technical complexity is

is much more the technical complexity is more but those are the five main

more but those are the five main categories

categories and you you so I mean that's a that's a

and you you so I mean that's a that's a much broader way to describe signal and

much broader way to describe signal and you mentioned that it's a pet peeve of

you mentioned that it's a pet peeve of yours when I guess described incorrectly

yours when I guess described incorrectly or maybe surface level pushed out there.

or maybe surface level pushed out there. Do you notice that that is a description

Do you notice that that is a description on the customer side or other

on the customer side or other competitors in your field? Like where

competitors in your field? Like where does that come from? I'm I'm looking I'm

does that come from? I'm I'm looking I'm looking it more from if I was in the

looking it more from if I was in the shoes of my buyer because that's how I

shoes of my buyer because that's how I like to think about problems like if I

like to think about problems like if I was the buyer of app zeta myself what

was the buyer of app zeta myself what would I be thinking and if I I'll be

would I be thinking and if I I'll be like wow this is crazy like what sort of

like wow this is crazy like what sort of signal are you doing and this happens to

signal are you doing and this happens to us on a recurring basis in our customer

us on a recurring basis in our customer call in our sales calls right we're

call in our sales calls right we're having calls with prospects it takes a

having calls with prospects it takes a few iterations to get to tell them what

few iterations to get to tell them what exactly we do some of them don't even

exactly we do some of them don't even know it's possible let alone knowing

know it's possible let alone knowing that you this space exists. So this is

that you this space exists. So this is how early the space itself is. You need

how early the space itself is. You need to tell them this can actually be

to tell them this can actually be possible and that's a problem statement

possible and that's a problem statement because you're almost trying to create

because you're almost trying to create awareness and educate the industry but

awareness and educate the industry but also be empathetic that they just don't

also be empathetic that they just don't understand a lot of the times. So comes

understand a lot of the times. So comes with some sort of a learning curve. Of

with some sort of a learning curve. Of course vendors will do the best they

course vendors will do the best they can. We are a vendor. So like of course

can. We are a vendor. So like of course we but at the same point transparency in

we but at the same point transparency in this day and age is super key and we do

this day and age is super key and we do not want to operate in any of the other

not want to operate in any of the other four categories and we are very explicit

four categories and we are very explicit about that that we are not doing any

about that that we are not doing any other other stages. So if you are

other other stages. So if you are looking at signals from that lens,

looking at signals from that lens, measurement, attribution, we are not the

measurement, attribution, we are not the right fit for you and we we tell them up

right fit for you and we we tell them up front because it's a lot of the vendors

front because it's a lot of the vendors out there will obviously say okay sure

out there will obviously say okay sure we can like you know make ways around

we can like you know make ways around and that's a classic sort of agency

and that's a classic sort of agency mentality because agencies are like so

mentality because agencies are like so broad that if someone comes to them for

broad that if someone comes to them for paid media and someone asks them graphic

paid media and someone asks them graphic design or whatever like sure we do this

design or whatever like sure we do this in our world like so it's a classic that

in our world like so it's a classic that but like for us we need to be laser

but like for us we need to be laser focused on this problem statement

focused on this problem statement because it's very technically

because it's very technically complicated as is and we don't want to

complicated as is and we don't want to be you know spreading ourselves too thin

be you know spreading ourselves too thin by, you know, just increasing the basket

by, you know, just increasing the basket size, so to speak.

size, so to speak. No, I think that's a good approach

No, I think that's a good approach because you're right, a lot of agencies

because you're right, a lot of agencies do the jack of all trades, master of

do the jack of all trades, master of none, but it sounds like you're

none, but it sounds like you're narrowing in on one like set of mastery.

narrowing in on one like set of mastery. When you mention that customers

When you mention that customers sometimes don't believe it's possible,

sometimes don't believe it's possible, do they share their source of disbelief

do they share their source of disbelief or is it something that you've just

or is it something that you've just gathered in your your decade of

gathered in your your decade of experience what the likely source is?

experience what the likely source is? likely sources because most of the times

likely sources because most of the times I think that if this was the case then

I think that if this was the case then why Meta and Google are not doing this

why Meta and Google are not doing this and they almost think of us it's almost

and they almost think of us it's almost like a shift where they think that we

like a shift where they think that we are trying to out bit Meta but it's far

are trying to out bit Meta but it's far from the truth we're not trying to

from the truth we're not trying to outbid anyone like no one can out bit

outbid anyone like no one can out bit Meta or Google in what they're doing but

Meta or Google in what they're doing but in what they do in terms of the bidding

in what they do in terms of the bidding algorithms there's a space in which you

algorithms there's a space in which you can supply this forward-looking signal

can supply this forward-looking signal and they actually augment that and Meta

and they actually augment that and Meta and Google are one of our biggest

and Google are one of our biggest advocates because if advertisers are

advocates because if advertisers are spending on their platform and getting

spending on their platform and getting much more returns. They spend more money

much more returns. They spend more money on the platform. So, it's a win-win for

on the platform. So, it's a win-win for everyone in that ecosystem like all

everyone in that ecosystem like all three, the brand, the meta and us

three, the brand, the meta and us included. So, everyone wins in that sort

included. So, everyone wins in that sort of three-way uh space. Um but it's

of three-way uh space. Um but it's mainly a question of confusion where

mainly a question of confusion where they think that this might be a question

they think that this might be a question of you're coming instead of Meta, but

of you're coming instead of Meta, but that's not the truth. Like, we are

that's not the truth. Like, we are actually complenting their bidding stack

actually complenting their bidding stack by supplying a signal layer and then

by supplying a signal layer and then that does its magic, which is it's

that does its magic, which is it's fantastic at doing anyways.

fantastic at doing anyways. Got it. Yeah. I mean I can see I can see

Got it. Yeah. I mean I can see I can see why there's a assumption there like how

why there's a assumption there like how how you describe it leads to that. Yeah

how you describe it leads to that. Yeah that's one of the questions where I ask

that's one of the questions where I ask like how like why can't Meta and Google

like how like why can't Meta and Google do this themselves. So the the answer to

do this themselves. So the the answer to that is Meta and Google are fantastic at

that is Meta and Google are fantastic at providing you those propensity logics

providing you those propensity logics because they operate on that 7-day

because they operate on that 7-day conversion window and whatever signals

conversion window and whatever signals they receive in that 7-day conversion

they receive in that 7-day conversion window can be great proxies for some

window can be great proxies for some shallow level events but they often are

shallow level events but they often are not really good indicators of long-term

not really good indicators of long-term customer lifetime value because a lot of

customer lifetime value because a lot of these indicators will be maturing after

these indicators will be maturing after 7 days. So anything you send to after to

7 days. So anything you send to after to the platform after 7 days is is just a

the platform after 7 days is is just a blind spot. platforms have already

blind spot. platforms have already learned and they have gone towards that

learned and they have gone towards that sort of user which is the easiest to

sort of user which is the easiest to acquire because that's the signal you're

acquire because that's the signal you're giving them. You're telling them to

giving them. You're telling them to acquire users on checkout value. For

acquire users on checkout value. For example, if you have a check out process

example, if you have a check out process or you want it to optimize towards that

or you want it to optimize towards that AOV, then platforms will optimize

AOV, then platforms will optimize towards that and it will go and find

towards that and it will go and find that user at the minimum cost possible.

that user at the minimum cost possible. And that's what it does brilliantly by

And that's what it does brilliantly by the way. But it what it does not do is

the way. But it what it does not do is it does not take into account whether

it does not take into account whether that one check out value is a one time

that one check out value is a one time buyer or that check out value is a 12

buyer or that check out value is a 12 month subscriber.

month subscriber. So the value can vary from $60 to $600,

So the value can vary from $60 to $600, but at the time of the checkout, it's

but at the time of the checkout, it's only $60. So the forward-looking

only $60. So the forward-looking indicator is not possible unless you

indicator is not possible unless you actually feed in predictions. So it's

actually feed in predictions. So it's great at what it does, but again it goes

great at what it does, but again it goes back to the fact that what input you're

back to the fact that what input you're feeding it.

feeding it. So yeah, it's I guess it's the bad in

So yeah, it's I guess it's the bad in bad out kind of thing that could could

bad out kind of thing that could could un unfold. And you mentioned prediction.

un unfold. And you mentioned prediction. So like I think you have a clear

So like I think you have a clear distinction between a prediction and a

distinction between a prediction and a signal. Can can you just describe the

signal. Can can you just describe the difference like how you interpret those

difference like how you interpret those two different words in your space? So

two different words in your space? So the predict so they're two fundament so

the predict so they're two fundament so you're absolutely spot on that

you're absolutely spot on that prediction and signals are fundamentally

prediction and signals are fundamentally different things because the you can

different things because the you can have a perfectly perfectly trained model

have a perfectly perfectly trained model and it when you predict something the

and it when you predict something the accuracy of the model is determined on

accuracy of the model is determined on like the root mean square error or like

like the root mean square error or like the may and the makes which are the

the may and the makes which are the standard statistical metrics and what

standard statistical metrics and what that does is it tells you what you're

that does is it tells you what you're predicting what's the actual what's the

predicting what's the actual what's the discrepancy in that right so a model can

discrepancy in that right so a model can be the most accurate piece or it can be

be the most accurate piece or it can be the most accurate thing ever assembled

the most accurate thing ever assembled and you can you can have a very robust

and you can you can have a very robust model but it can have no impact

model but it can have no impact whatsoever on the option mechanics

whatsoever on the option mechanics because of the way you're sending it. So

because of the way you're sending it. So if you are sending predictive values

if you are sending predictive values first of all you cannot send them raw

first of all you cannot send them raw you cannot send them raw because they're

you cannot send them raw because they're probabilistic in nature and these

probabilistic in nature and these systems by design are only designed for

systems by design are only designed for deterministic outcomes which are not

deterministic outcomes which are not probabilistic which have already

probabilistic which have already happened which are does not come with a

happened which are does not come with a probability they have they come with a

probability they have they come with a time stamp and when you send them

time stamp and when you send them without any sort of calibration or vi

without any sort of calibration or vi without taking care of the auction

without taking care of the auction mechanics uh it can be considered noise

mechanics uh it can be considered noise to them if you're set in a very tightly

to them if you're set in a very tightly clustered or if you're sending them like

clustered or if you're sending them like very widely clustered We'll go into

very widely clustered We'll go into technical details, but like long story

technical details, but like long story short, there are two pillars here and

short, there are two pillars here and they don't talk interchangeably and

they don't talk interchangeably and they're two separate ends of the

they're two separate ends of the spectrum where model effectiveness is

spectrum where model effectiveness is not correlated or it's not a direct

not correlated or it's not a direct function of what the signal design might

function of what the signal design might look like. So the effectiveness of the

look like. So the effectiveness of the auction is a different thing and the

auction is a different thing and the prediction itself is a different thing

prediction itself is a different thing and they both need to be talked closely

and they both need to be talked closely together and that's where the

together and that's where the calibration comes in.

calibration comes in. Got it. So, so like so there you've seen

Got it. So, so like so there you've seen in your experience very strong technical

in your experience very strong technical models but they perform operationally

models but they perform operationally weak when deployed like basic. Okay, got

weak when deployed like basic. Okay, got it. So, exactly. So, is is that is that

it. So, exactly. So, is is that is that one of the core lifts that you have to

one of the core lifts that you have to undergo when someone comes onto your

undergo when someone comes onto your platform?

platform? 100%.

100%. 100%. Yes. The most technically those

100%. Yes. The most technically those two pieces the calibration and the

two pieces the calibration and the orchestration understanding the oxygen

orchestration understanding the oxygen mechanics is are the two biggest pieces

mechanics is are the two biggest pieces of why this is so complex and why

of why this is so complex and why in-house teams more often than not get

in-house teams more often than not get this wrong as well because everyone

this wrong as well because everyone overindexes of your model accuracy. All

overindexes of your model accuracy. All the time spent by data science and

the time spent by data science and machine learning teams goes in how

machine learning teams goes in how accurate we can make the model not on

accurate we can make the model not on how that downstream effect goes because

how that downstream effect goes because these are two different things. You can

these are two different things. You can invest all all your time in that but it

invest all all your time in that but it can have a 95% appricate model.

can have a 95% appricate model. Fantastic. Does it talk to the platform

Fantastic. Does it talk to the platform in which it has an impact on your

in which it has an impact on your campaigns? If yes, brilliant. If not,

campaigns? If yes, brilliant. If not, then all that model and all that compute

then all that model and all that compute and manpower has gone for nothing

and manpower has gone for nothing because you're not building a model for

because you're not building a model for reporting. You're building a model for

reporting. You're building a model for bidding. And these are two fundamentally

bidding. And these are two fundamentally different things. Because when you think

different things. Because when you think about LTV, everyone thinks about LTV

about LTV, everyone thinks about LTV like 95% of the conversations which go

like 95% of the conversations which go talk about LTV on a very loose sense.

talk about LTV on a very loose sense. They all talk about historical LTV. They

They all talk about historical LTV. They all look at backward looking LTV which

all look at backward looking LTV which is used for reporting which can be used

is used for reporting which can be used for any sort of you know forecasting you

for any sort of you know forecasting you know that sort of thing. it has no

know that sort of thing. it has no correlation with the the forward-looking

correlation with the the forward-looking nature of how you're optimizing bids and

nature of how you're optimizing bids and customer acquisition

customer acquisition and that's where the PLTB space comes in

and that's where the PLTB space comes in and LTB and PLTV are distinct categories

and LTB and PLTV are distinct categories where they need to be treated very

where they need to be treated very drastically differently

drastically differently and do do you do your clients get that

and do do you do your clients get that like instinctively or is it something

like instinctively or is it something still a lot of education and you're

still a lot of education and you're pouring a lot of time? Yeah,

pouring a lot of time? Yeah, it's education but once you tell them

it's education but once you tell them they get it right away. It's a it's a

they get it right away. It's a it's a very quick sort of you know aha sort of

very quick sort of you know aha sort of a moment where okay this makes sense and

a moment where okay this makes sense and it's not necessarily a bad thing because

it's not necessarily a bad thing because the way everything is structured around

the way everything is structured around you know LTB you know it's if you go

you know LTB you know it's if you go unless you're actually looking at from a

unless you're actually looking at from a very deep level you would know that it's

very deep level you would know that it's all bad for looking and it only solves a

all bad for looking and it only solves a certain select cohort of purposes

certain select cohort of purposes and unless you go into the deep ends of

and unless you go into the deep ends of it and solve the purpose and again it

it and solve the purpose and again it leads back to what I was saying you have

leads back to what I was saying you have to leave that from the outcome if your

to leave that from the outcome if your outcome is improving your customer

outcome is improving your customer policy and app for the shoe quality and

policy and app for the shoe quality and improving the cohorts of customer

improving the cohorts of customer acquire. You cannot use platform looking

acquire. You cannot use platform looking as your only metric because it will not

as your only metric because it will not impact the bing decisions. So again lead

impact the bing decisions. So again lead with the outcome and reverse engineer

with the outcome and reverse engineer from there and then see the best tool

from there and then see the best tool out there and then you will be in a

out there and then you will be in a better position. That's what the brands

better position. That's what the brands you work with usually are.

you work with usually are. I'm not a buyer in your space but that

I'm not a buyer in your space but that was a convincing string of sentences to

was a convincing string of sentences to like to to go in and use your your

like to to go in and use your your product and work with you. So I can see

product and work with you. So I can see why you're like I I assume your aha

why you're like I I assume your aha moments with customers happen quite

moments with customers happen quite quickly.

quickly. Yeah. Uh in usually in that first

Yeah. Uh in usually in that first conversation a lot of these sort of

conversation a lot of these sort of quote unquote misconceptions uh for a

quote unquote misconceptions uh for a lack of better word they come into play

lack of better word they come into play and it's but again that's the I we sort

and it's but again that's the I we sort of thrive in that ecosystem because it's

of thrive in that ecosystem because it's a it's a challenge we accept uh it comes

a it's a challenge we accept uh it comes with a risk because we have potentially

with a risk because we have potentially longer sales cycle longer decision

longer sales cycle longer decision cycles. But I would rather be in that

cycles. But I would rather be in that position than being working at the

position than being working at the cutting edge of stuff rather than being

cutting edge of stuff rather than being multiple years late or decades late for

multiple years late or decades late for that matter and working in something

that matter and working in something which is very mundane. Right? Because

which is very mundane. Right? Because this category is super new only a

this category is super new only a handful of players exist on the planet

handful of players exist on the planet and that's for a reason because it's

and that's for a reason because it's very you need a lot of moving pieces to

very you need a lot of moving pieces to get this right.

get this right. Right. A just that's a very just that's

Right. A just that's a very just that's a great way to go into my my next

a great way to go into my my next question which is just about your team

question which is just about your team like what kind of team do you need to

like what kind of team do you need to build a company like this cuz you're

build a company like this cuz you're sitting across growth, machine learning,

sitting across growth, machine learning, infrastructure, marketing like there's

infrastructure, marketing like there's so many disciplines you're touching. How

so many disciplines you're touching. How do you just like look at that when

do you just like look at that when you're building out your team?

you're building out your team? So three main pillars stand out. One is

So three main pillars stand out. One is data engineering. So someone with a very

data engineering. So someone with a very strong infrastructure background who can

strong infrastructure background who can who has knowledge about the

who has knowledge about the architectural decisions, what plumbing

architectural decisions, what plumbing decisions to make, what's the best thing

decisions to make, what's the best thing for um to to use for a certain use case

for um to to use for a certain use case based on cloud environments all that

based on cloud environments all that sort of stuff. Then you look at

sort of stuff. Then you look at production quality machine learning

production quality machine learning experience where people have actually

experience where people have actually built production models not just

built production models not just theoretical models but actually in

theoretical models but actually in quality at at scale and then lastly

quality at at scale and then lastly people who understand app platform

people who understand app platform mechanics as well because that's super

mechanics as well because that's super critical when it comes to platform

critical when it comes to platform mechanics. I've personally seen myself

mechanics. I've personally seen myself overseeing a lot of this stuff because I

overseeing a lot of this stuff because I come from growth marketing but I've

come from growth marketing but I've overseen that data science component. So

overseen that data science component. So I can talk from a products for both of

I can talk from a products for both of these pieces the data science and the

these pieces the data science and the the and the app platform mechanics but

the and the app platform mechanics but we need specialists in the these roles

we need specialists in the these roles and we do have specialists in these

and we do have specialists in these roles. So for production policy ML we

roles. So for production policy ML we have a stats PhD from UC Berkeley he

have a stats PhD from UC Berkeley he actually worked in this stuff at Google

actually worked in this stuff at Google in terms of these building these

in terms of these building these pipelines and he was a data scientist uh

pipelines and he was a data scientist uh there. So he knows exactly how to build

there. So he knows exactly how to build these systems and then for infra as well

these systems and then for infra as well we have someone who actually works in

we have someone who actually works in that sort of the service cappy plumbing

that sort of the service cappy plumbing infrared and connects the dots when it

infrared and connects the dots when it comes to these two. So these three are

comes to these two. So these three are critical most teams would have two out

critical most teams would have two out of the three and as much as I hate to

of the three and as much as I hate to say it it's not enough cuz like you need

say it it's not enough cuz like you need to have all three right to get this at

to have all three right to get this at least at a bare minimum. If you have

least at a bare minimum. If you have people who understand the architecture

people who understand the architecture really really well then you can have you

really really well then you can have you know people using them using AI to kind

know people using them using AI to kind of you know be as disruptive as possible

of you know be as disruptive as possible they can move at a velocity of 10x but

they can move at a velocity of 10x but if the architectural and decision-m is

if the architectural and decision-m is not right then AI won't help you so

not right then AI won't help you so that's one thing which I'll look for

that's one thing which I'll look for hiring is like how much they can like

hiring is like how much they can like spin on their like how much they can

spin on their like how much they can think on their own from like a critical

think on their own from like a critical thinking force principles architectural

thinking force principles architectural mindset because then AI can be a

mindset because then AI can be a multiplier a force multiplier but if not

multiplier a force multiplier but if not AI will only be garbage in garbage out

AI will only be garbage in garbage out like you mentioned earlier. So you need

like you mentioned earlier. So you need to have that sound reasoning and these

to have that sound reasoning and these three pillars are the ones which we use

three pillars are the ones which we use for hiring.

for hiring. And do do you create like a environment

And do do you create like a environment where you can measure technical judgment

where you can measure technical judgment through your hiring process? Is it like

through your hiring process? Is it like possible to do that or is it just based

possible to do that or is it just based on their like his historical work or

on their like his historical work or public work that they posted?

public work that they posted? So great question. There's no like meta

So great question. There's no like meta style into coding interviews because

style into coding interviews because right now the dynamic has changed and

right now the dynamic has changed and like how would you know if the person is

like how would you know if the person is not using AI for that piece either? And

not using AI for that piece either? And to be fair though, it's not the right

to be fair though, it's not the right judgment either because I would not I

judgment either because I would not I would not not want them to use AI in a

would not not want them to use AI in a way if that makes sense. Like I I would

way if that makes sense. Like I I would encourage them, right? So if I was

encourage them, right? So if I was judging them just on the M of that only

judging them just on the M of that only that logic only, it would be unfair

that logic only, it would be unfair because I'm like snatching the the one

because I'm like snatching the the one some of the best tools out there from

some of the best tools out there from them. However, what we do search for and

them. However, what we do search for and look for is how much they can think like

look for is how much they can think like a product owner, think like an

a product owner, think like an architectural decision maker and how

architectural decision maker and how much they can actually, you know, not

much they can actually, you know, not just work themselves but inspire a team.

just work themselves but inspire a team. And those are some sort of first

And those are some sort of first principles preI and post AI that has

principles preI and post AI that has changed for me because previously I used

changed for me because previously I used to hire data scientists who had very

to hire data scientists who had very strong vertical exposure into what they

strong vertical exposure into what they were working on. So if I need to work

were working on. So if I need to work for direct to consumer or any sort of

for direct to consumer or any sort of vertical specifically I will have for

vertical specifically I will have for that particular vertical experience. If

that particular vertical experience. If they have it, great. They check check

they have it, great. They check check the box and they're now it's actually

the box and they're now it's actually flipped. I need people who can actually

flipped. I need people who can actually who have more horizontal experience and

who have more horizontal experience and they can actually cascade out to

they can actually cascade out to different verticals and apply the same

different verticals and apply the same level of critical mindset and first

level of critical mindset and first principal thinking across the verticals

principal thinking across the verticals and not just being limited to one

and not just being limited to one vertical and that can happen if you

vertical and that can happen if you think like a product owner if you have

think like a product owner if you have good architectural understanding and

good architectural understanding and then of course with the talk through the

then of course with the talk through the use of AI tools now you're like a TX

use of AI tools now you're like a TX team. So that's one of the things our

team. So that's one of the things our team is able to move at at a velocity of

team is able to move at at a velocity of a 40% if I'm being kind

a 40% if I'm being kind which is massive.

which is massive. Yeah. Yeah. No kidding. Yeah. So and

Yeah. Yeah. No kidding. Yeah. So and that must help especially being

that must help especially being bootstrapped you're keeping your team

bootstrapped you're keeping your team quite lean and you're investing in like

quite lean and you're investing in like all the right elements to move fast with

all the right elements to move fast with like judgment layer which is what

like judgment layer which is what everyone's citing now is the most

everyone's citing now is the most important thing right over all AI models

important thing right over all AI models is the actual judgment to deploy them

is the actual judgment to deploy them and and on that like this is just I

and and on that like this is just I guess for broader set of people out

guess for broader set of people out there just founders outside of adtech

there just founders outside of adtech what do you think they can learn between

what do you think they can learn between like uh building an intelligent model

like uh building an intelligent model and then actually having it act in the

and then actually having it act in the real world

real world yeah I mean that activation is the

yeah I mean that activation is the product for us. So I keep on thinking

product for us. So I keep on thinking about that and I was really particular

about that and I was really particular about when we were building at Zeta and

about when we were building at Zeta and that it does not just live in a

that it does not just live in a dashboard. So what I like to say again

dashboard. So what I like to say again this is me being biased but I feel like

this is me being biased but I feel like data used to be a mode 10 years ago.

data used to be a mode 10 years ago. Data is not a mode anymore. I mean sure

Data is not a mode anymore. I mean sure we're looking at different lenses. I

we're looking at different lenses. I mean sure if you're looking at

mean sure if you're looking at foundational models data is still the

foundational models data is still the mode. The quality of the cleanliness of

mode. The quality of the cleanliness of data is all super important. I'm just

data is all super important. I'm just saying from a from a machine learning

saying from a from a machine learning and data science standpoint and from a

and data science standpoint and from a brand's perspective as well like you

brand's perspective as well like you know having more data is not the moat.

know having more data is not the moat. How you use that data and how you

How you use that data and how you activate them is what you need to be

activate them is what you need to be good at. And when you think of you know

good at. And when you think of you know other founders and not just in general

other founders and not just in general like you know data scientists and people

like you know data scientists and people who are in this space almost think about

who are in this space almost think about it from an activation lens how this data

it from an activation lens how this data can actually quantify and move to

can actually quantify and move to business outcomes because a lot of the

business outcomes because a lot of the data scientists out there they're

data scientists out there they're fantastic in their own technical right

fantastic in their own technical right but they do not have enough of a sort of

but they do not have enough of a sort of a business hat on or a product hat on

a business hat on or a product hat on that how will this tie back into some

that how will this tie back into some sort of a tangible lift that we can

sort of a tangible lift that we can provide. That can be rorowass, that can

provide. That can be rorowass, that can be contribution margin, that can be

be contribution margin, that can be anything, any number that needs to tie

anything, any number that needs to tie back and then you reverse engineer from

back and then you reverse engineer from there and see how you can actually go

there and see how you can actually go and build that system which allows that

and build that system which allows that to you you to kind of you know mimic

to you you to kind of you know mimic that lift. So having that sort of

that lift. So having that sort of product add-on is one of the most

product add-on is one of the most important things because now with the

important things because now with the roles converging the product the PM

roles converging the product the PM roles the technical PM roles the coding

roles the technical PM roles the coding roles they all are converging into one

roles they all are converging into one thing and the best thing you can do

thing and the best thing you can do right now is learn to think like a

right now is learn to think like a product owner and once you do that I

product owner and once you do that I think you'll be you'll be in a much

think you'll be you'll be in a much better decision a much better position

better decision a much better position rather and you'll be able to add

rather and you'll be able to add significant value to any ecosystem

significant value to any ecosystem because now when people talk about lean

because now when people talk about lean teams you can be like a three person

teams you can be like a three person team of a product coder And then

team of a product coder And then obviously with AI you can be like a

obviously with AI you can be like a force multiplier. So I think that's what

force multiplier. So I think that's what I would say is the most important thing

I would say is the most important thing at at least at least that's what we look

at at least at least that's what we look for in terms of hiring and that's what

for in terms of hiring and that's what we think is arguably one of the most

we think is arguably one of the most resonant theme with people who are

resonant theme with people who are hiring at good levels and also

hiring at good levels and also succeeding in their roles.

succeeding in their roles. If someone doesn't have that skill just

If someone doesn't have that skill just yet but they're technically sound in one

yet but they're technically sound in one particular area. Would you say like your

particular area. Would you say like your model of customer outcome working

model of customer outcome working backwards is like a quick way for them

backwards is like a quick way for them to accelerate their thinking as a

to accelerate their thinking as a product owner?

product owner? Yeah, it's a cheat sheet. Yeah, it's a

Yeah, it's a cheat sheet. Yeah, it's a cheat code. Yeah, of course. Because

cheat code. Yeah, of course. Because that's I think that's the easy way and I

that's I think that's the easy way and I always I always tell my technical team

always I always tell my technical team especially the ones who are super

especially the ones who are super technical that you need to start with

technical that you need to start with the why. Like why are you doing or why

the why. Like why are you doing or why are you taking a certain decision at

are you taking a certain decision at this particular time and what value does

this particular time and what value does it add? And once you start with that

it add? And once you start with that your decisions here in the technical

your decisions here in the technical space will keep on changing because then

space will keep on changing because then you're thinking not just from that like

you're thinking not just from that like in silo but you're thinking and

in silo but you're thinking and contextualizing that from a business

contextualizing that from a business perspective and once you do that over

perspective and once you do that over passage of time it's more reps it takes

passage of time it's more reps it takes a bit more manual effort in the

a bit more manual effort in the beginning but it will come naturally

beginning but it will come naturally afterwards and once that happens you are

afterwards and once that happens you are a much stronger engineer in this day and

a much stronger engineer in this day and age because you already are embibing

age because you already are embibing that product mindset in you

that product mindset in you subconsciously and before you know it

subconsciously and before you know it it's right there. So that's why the

it's right there. So that's why the reverse thinking really works.

reverse thinking really works. Uhhuh.

Uhhuh. And starting with the Y is my second

And starting with the Y is my second piece to add to that.

piece to add to that. Yeah. Yeah. No, I think that's a very

Yeah. Yeah. No, I think that's a very sound methodology for people converging

sound methodology for people converging from if you're a PM now going into the

from if you're a PM now going into the technical side or if you're an engineer

technical side or if you're an engineer going into the other side. I mean

going into the other side. I mean ultimately everyone is converging

ultimately everyone is converging towards a business outcome internally

towards a business outcome internally and externally for your clients. So I

and externally for your clients. So I think you have it right and you're

think you have it right and you're building a lean team around that. I'm

building a lean team around that. I'm going to ask you just some rapid

going to ask you just some rapid questions like outside of of your

questions like outside of of your business just more about you and how you

business just more about you and how you approach your day-to-day. The first one

approach your day-to-day. The first one I wanted to ask you was like what

I wanted to ask you was like what domains capture your attention outside

domains capture your attention outside of adtech

of adtech healthcare. Uh there's a lot of

healthcare. Uh there's a lot of incentive happening in healthcare right

incentive happening in healthcare right now which I feel can be disrupted with

now which I feel can be disrupted with the use of AI. You know I'm looking at a

the use of AI. You know I'm looking at a lot of research right now happening in

lot of research right now happening in in you know cancer research and these

in you know cancer research and these sort of areas where AI can massively

sort of areas where AI can massively disrupt that areas. And of course it

disrupt that areas. And of course it comes with massive compliance issues.

comes with massive compliance issues. But that's one thing which I think can

But that's one thing which I think can massively beneficial not just the space

massively beneficial not just the space itself but humanity in general and this

itself but humanity in general and this should be used as a multiplier in that

should be used as a multiplier in that space where we are talking about

space where we are talking about dramatically different outcomes not just

dramatically different outcomes not just and and that's where I think the

and and that's where I think the paradigm of AI almost being used as like

paradigm of AI almost being used as like some sort of a negative converation when

some sort of a negative converation when it comes to data centers and all that

it comes to data centers and all that stuff it can spin a very positive effect

stuff it can spin a very positive effect and that's I think I personally am very

and that's I think I personally am very interested in that piece and looking

interested in that piece and looking forward to that. So that's something

forward to that. So that's something that you track like adjacent to

that you track like adjacent to track as No, I don't track that often. I

track as No, I don't track that often. I would love to have more time to track

would love to have more time to track it.

it. Yeah. Yeah.

Yeah. Yeah. But yeah, there's lots happening in our

But yeah, there's lots happening in our space itself. But like that's one thing

space itself. But like that's one thing which I'll be once once I have some more

which I'll be once once I have some more time, I'll be more interested in like

time, I'll be more interested in like hearing and listening to podcasts and

hearing and listening to podcasts and stuff. Of course, in terms of AI, I

stuff. Of course, in terms of AI, I always look at you know the cutting edge

always look at you know the cutting edge stuff when it comes to foundational

stuff when it comes to foundational models. What new models are coming in.

models. What new models are coming in. Now it's to talk about harnesses, right?

Now it's to talk about harnesses, right? I mean how you actually have harnesses

I mean how you actually have harnesses being angled properly in order to make

being angled properly in order to make sure that your model does not drift and

sure that your model does not drift and it does not loosen because that's one of

it does not loosen because that's one of the biggest concerns right now. So

the biggest concerns right now. So that's that category is becoming more

that's that category is becoming more and more prevalent now which was not the

and more prevalent now which was not the case but it should have been the case

case but it should have been the case much earlier. So that's like these few

much earlier. So that's like these few things are very interesting to me.

things are very interesting to me. Got it. Yeah. I mean those are it's it's

Got it. Yeah. I mean those are it's it's sometimes it's hard to drift away from

sometimes it's hard to drift away from your core work and you naturally I think

your core work and you naturally I think go into places that contribute to what

go into places that contribute to what you're doing dayto-day. So that makes a

you're doing dayto-day. So that makes a lot of sense. That actually leads me to

lot of sense. That actually leads me to my my next question which is what

my my next question which is what actually takes you out of founder mode

actually takes you out of founder mode where you can switch your mind off like

where you can switch your mind off like is there is there

is there is there yeah I like to read and play tennis uh

yeah I like to read and play tennis uh two opposite ends of the spectrum but

two opposite ends of the spectrum but like you know one is more sedentary one

like you know one is more sedentary one is more active uh but like working out

is more active uh but like working out you know playing tennis uh just high

you know playing tennis uh just high intensity stuff is like a great way to

intensity stuff is like a great way to kind of you know keep your endorphins

kind of you know keep your endorphins flowing keep your you know energy right

flowing keep your you know energy right up there I pay a lot of importance on

up there I pay a lot of importance on good sleep as well because that's super

good sleep as well because that's super critical for productivity and I I can

critical for productivity and I I can feel that if I've not slept well. So

feel that if I've not slept well. So these things are so prepared and these

these things are so prepared and these are non-negotiables for me. I can have a

are non-negotiables for me. I can have a long day at work but I'll still work out

long day at work but I'll still work out at whatever time of the night. So these

at whatever time of the night. So these are things which are really important

are things which are really important and these are for every follower. I

and these are for every follower. I highly recommend because they can have a

highly recommend because they can have a massive difference in what you do during

massive difference in what you do during your day job.

your day job. Yeah. No, I agree. And as a tennis

Yeah. No, I agree. And as a tennis player, I have to ask you and throw in a

player, I have to ask you and throw in a book recommendation. Have you read the

book recommendation. Have you read the inner game of tennis?

inner game of tennis? No.

No. Okay. Um you got to add it to your list.

Okay. Um you got to add it to your list. I think you'll enjoy it. Yeah. Yeah.

I think you'll enjoy it. Yeah. Yeah. It's a fantastic one about just mindset

It's a fantastic one about just mindset about playing sports and running

about playing sports and running businesses. Um, I'll send you a link.

businesses. Um, I'll send you a link. There's a podcast from Jovovich and you

There's a podcast from Jovovich and you know his new documentary is coming.

know his new documentary is coming. There was an announced documentary

There was an announced documentary earlier.

earlier. So there's a lot of good content coming

So there's a lot of good content coming but I haven't read the book. I'll

but I haven't read the book. I'll definitely read that.

definitely read that. Okay. I I'll I'll share the link with

Okay. I I'll I'll share the link with you. It's a really good one. I think

you. It's a really good one. I think you'll enjoy it. Um, and then lastly, is

you'll enjoy it. Um, and then lastly, is there a founder or a business leader

there a founder or a business leader that you borrow from or emulate and like

that you borrow from or emulate and like what is something that you take from

what is something that you take from them that really resonates with you in

them that really resonates with you in your day-to-day? There's a lot of things

your day-to-day? There's a lot of things which I liked about Ryan Jesse in the

which I liked about Ryan Jesse in the way he operates his business and the way

way he operates his business and the way he thinks about problems and the way he

he thinks about problems and the way he thinks about hiring recruiting in

thinks about hiring recruiting in general. I think he's the one

general. I think he's the one who in my opinion is like I can relate

who in my opinion is like I can relate the most to more empathetic more

the most to more empathetic more personal level and also the fact that

personal level and also the fact that more unfiltered right a lot of these

more unfiltered right a lot of these founders out there are like you know

founders out there are like you know super filtered and super like you know

super filtered and super like you know particular like you know in terms of

particular like you know in terms of oversharing he is very open in that

oversharing he is very open in that piece and he has that sort of product

piece and he has that sort of product lens he comes from that background he

lens he comes from that background he has like culture as well like industrial

has like culture as well like industrial industrial engineering and stuff so I

industrial engineering and stuff so I really resonate with like a lot of the

really resonate with like a lot of the things that he talks about in his

things that he talks about in his podcast and the way he thinks about

podcast and the way he thinks about problems the way he shares

problems the way he shares I'm more in the lens of oversharing than

I'm more in the lens of oversharing than under sharing because transparency right

under sharing because transparency right now is like, you know, it's it's

now is like, you know, it's it's critical and it's key. So, I resonate

critical and it's key. So, I resonate the most with the way he talks in the

the most with the way he talks in the way he presents himself.

way he presents himself. Yeah. Yeah. He he would and he's been

Yeah. Yeah. He he would and he's been heavily scrutinized in different areas,

heavily scrutinized in different areas, but he's always taken it on and

but he's always taken it on and explained thoroughly why he took a

explained thoroughly why he took a position. So, um I can see why that that

position. So, um I can see why that that like especially also in your space,

like especially also in your space, you're building a small team. You must

you're building a small team. You must be it must be a very trans transparent

be it must be a very trans transparent and high like canandoandor must be

and high like canandoandor must be valued by you inside your small team. So

valued by you inside your small team. So that makes a lot of sense. Yeah. Yeah.

that makes a lot of sense. Yeah. Yeah. So so that's it for today. I mean I'm

So so that's it for today. I mean I'm going to this was great. I really

going to this was great. I really appreciated your time and hearing your

appreciated your time and hearing your insights. I learned a lot like

insights. I learned a lot like genuinely. I'm not an expert by no means

genuinely. I'm not an expert by no means in the space. So it was really nice to

in the space. So it was really nice to hear you just briefly explain so many

hear you just briefly explain so many things in a 30 minute cycle. So I

things in a 30 minute cycle. So I appreciate that.

appreciate that. Behind the cold.

Behind the cold. That's [music] where the future grows.

That's [music] where the future grows. Behind the cold. [music] Behind the co.