Latest / The Yield Doctor with James Deaker / What is Kana? Building AdTech with CTO Vivek Vaidya

Transcript

What is Kana? Building AdTech with CTO Vivek Vaidya · The Yield Doctor with James Deaker

0:00
I am James Deika. I am the eel doctor.
0:02
And today's discussion is part of a
0:05
project that I am embarking on to talk
0:07
to companies who are driving change in
0:10
the new AI landscape for advertising.
0:14
Specifically, I'm talking to those who
0:16
are using it to drive solutions for
0:18
publishers and ade providers. And I am
0:21
very transparent about the fact that
0:23
half of the reason I'm doing this is so
0:25
that I have a better take on all the
0:27
companies that are doing interesting
0:29
things in the space. But I also have a
0:31
belief that there are other people out
0:33
there like me who have heard of a
0:35
company, maybe seen it on the Lumcape,
0:38
but don't really know what the company
0:40
does or they maybe at best know what the
0:42
company did five years ago. So all of
0:45
that said, today I'm going to be talking
0:46
to VC Via. He is the CTO of Kana and I
0:50
want to disclose upfront that I have
0:52
worked with Kana in an advisory
0:54
capacity. VC and I first met 20 years
0:58
ago. It's hard to believe and we were
1:00
both working on the same project at WP
1:03
which was later acquired by Microsoft.
1:05
VC then went on to co-found Krux with
1:09
Tom Chavez and it was acquired by
1:12
Salesforce.
1:13
In addition to being the CTO of Kana,
1:16
VVC is the founder and general partner
1:19
at Superset, which is a startup studio.
1:22
VC, thank you so much for taking the
1:24
time to sit down with me today. Thanks
1:27
for having me, James. Looking forward to
1:28
our conversation.
1:30
>> Excellent. So, let's start right at the
1:32
beginning. What is the origin story of
1:34
KA? How did it get started? And how did
1:37
you end up there? I assume that those
1:39
two questions are very closely related.
1:42
Yeah, look, James, uh the registry of
1:44
China is interesting uh and a funny one
1:46
because when as you mentioned uh you
1:49
mentioned Tom Chavez, he and I we've
1:51
been working together for the last 27
1:54
years now. Um actually and he you
1:57
mentioned rap. Tom was the founder and
1:59
CEO of WP um and then Tom and I started
2:02
Krux which was acquired by Salesforce.
2:05
After the acquisition of Krux, I became
2:08
the CTO of Salesforce Marketing Cloud. I
2:11
stayed there for uh two years uh almost.
2:15
But then when I and then I left and then
2:18
Tom and I just were talking about what
2:20
to do next and then we started Superset.
2:22
One of the things we told ourselves as
2:24
we were starting Superset is we are not
2:26
going to build another adtech or
2:28
martekch company. We wanted to learn new
2:30
domains and and and all of that. that
2:33
lasted for uh I think three months or
2:36
maybe four um because we we launched
2:38
Haboo as one of the first companies from
2:41
the studio which was a clean room
2:42
company uh which Tom and I founded uh
2:45
and then we brought on other folks from
2:47
uh Krux to uh run it and that was
2:49
acquired by live ramp but then even even
2:52
after that we were still not ready to
2:54
deep dig deep and and build an adtech or
2:58
martekch again. What changed was uh
3:02
November tw November 29th I think it was
3:05
2022 when Chad GPD came out and uh all
3:09
the things that have happened since then
3:11
right the everything that people are uh
3:15
everything that people used to do is
3:17
being re-evaluated reimagined
3:19
reconceived using AI agentic AI
3:22
generative AI etc and so we said to
3:25
ourselves look what a great opportunity
3:27
to just start from scratch and a clean
3:30
whiteboard. Reimagine all of the ways in
3:32
which problems were earlier problems
3:34
that we had solved with all these
3:36
constraints. Maybe those constraints
3:37
would go go away. And really all of that
3:40
led to the uh formation of Kana. Um we
3:44
we are uh as we like to say an an uh a
3:48
platform for end-to-end
3:50
eentic marketing and we'll get into more
3:53
uh we'll get more into what what that
3:55
means but that's long and short of how
3:58
uh kind of got started and how we got
4:00
here James.
4:01
>> Perfect. Thank you. Uh so maybe maybe
4:04
this next question leads you into what
4:06
is agentic marketing. But I'm going to
4:08
ask a similar question for every company
4:10
I'm going to talk to which is explain
4:12
what your company does as if you're
4:14
explaining it to somebody who's been in
4:16
the industry for a number of years but
4:19
other than the name has no idea what you
4:20
do. What what is the problem that you're
4:22
trying to solve?
4:23
>> Yeah. And so one of the things that if
4:25
you think about what uh whether it's
4:29
adtech or martekch um doesn't matter
4:33
both uh domains which are rel they're
4:36
related domains they have um a plethora
4:40
of tools products SAS platforms that are
4:44
used by the companies who are in
4:47
publishers for example or media
4:49
companies they use or brands they use a
4:52
plethora of these tools to run their
4:54
business, right? And the challenge that
4:57
they have is these tools don't always
5:00
talk to each other cleanly. That's
5:02
problem number one. Problem number two
5:04
is that they were built as SAS platforms
5:07
and they always came with a uh a
5:12
services engagement that involved
5:14
implementation or configuration of the
5:16
SAS uh SAS platform to meet your
5:19
business. What that meant was that a a
5:23
business could not run cleanly in those
5:28
in those SL SAS platforms. It always had
5:30
to be mapped, configured, adapted, take
5:33
pick your word, right? And so operators
5:36
were constantly in this mode of download
5:40
data from one platform, download data
5:42
from another platform in the form of CSV
5:44
files, put them into a Google
5:46
spreadsheet or an Excel spreadsheet, do
5:48
joins, do reporting, get insights, and
5:51
then activate or go feed data back into
5:53
these systems. So a lot of dashboards
5:56
got built that told people what's so.
6:00
But there was no system that connected
6:03
all these dots and told the operator so
6:06
what. That is the problem that Kana is
6:09
trying to solve for both media companies
6:12
as well as marketers. It's the so what
6:14
problem on top of all of the different
6:17
data systems, data platforms, SAS
6:20
platforms, etc. that they use to run
6:22
their business.
6:25
So let me uh ask a question. I'm
6:28
struggling to work out how to ask it
6:30
precisely, but given that many companies
6:34
now have access to the same underlying
6:36
agents or models, how do you
6:41
differentiate at KA? Like what's, you
6:43
know, cuz there are probably a dozen
6:45
other companies who see the same
6:47
opportunity that you see. They have
6:49
access to the same underlying model. How
6:51
do you differentiate?
6:53
>> Yeah, it's a great question, James, and
6:54
it's it's a very relevant one. And uh
6:58
and you're right, all companies,
7:00
everybody trying to build now, whether
7:01
it's the one person building in their
7:03
garage to the uh 5,000 people behemoth,
7:07
they have access to the same models.
7:10
Anthropic, an entropic and open AI,
7:12
Google, take your pick, the open source,
7:14
open weight ones, doesn't matter. They
7:16
have access to the same models. So how
7:18
do you differentiate? You def you
7:20
differentiate by first developing a deep
7:23
understanding of the workflows and the
7:26
domain in which you're operating. Right?
7:28
So uh what what that means is not just
7:31
saying oh an ad server serves ads that's
7:33
not domain expertise. What what you need
7:35
to understand is every company that uses
7:39
an ad server or uses an order management
7:42
system or uses a uh CRM system
7:46
configures it differently. configures it
7:49
to work within the within the rules and
7:51
confines and constraints of their
7:53
business. You do not want to replicate
7:57
that mess in the agentic platform that
8:00
you're building. Right? So that's the
8:02
first way in which you differentiate is
8:04
we have built a canonical uh model that
8:08
data model that spans the entirety of
8:11
adtech and martekch. Why are we able to
8:13
do that? Because we spent the last 25
8:16
years building in it. Now the data model
8:18
itself is not enough. On top of that,
8:21
what you have to build is a semantic
8:23
layer that that is that goes deep into
8:27
the customer's implementation of these
8:30
systems, the ad server, the MS, the CRM,
8:33
etc. and is able to reverse engineer
8:36
their business and map it to the
8:39
canonical model. And the reason why this
8:41
is important is you do not want the
8:44
operator, your customer to pay that tax
8:47
of mapping their system into yours.
8:50
Right? So that's the first way in which
8:52
we differentiate.
8:54
Well, I I was going to say then if
8:56
because there's so much even to unpack
8:58
and you've got I'll let you come to the
9:00
other ways you differentiate but given
9:02
that you have a data and a layer up
9:04
front and given that every customer
9:06
business is different what does the
9:09
process look like of an implementation
9:11
when you've got something a structure
9:13
and to your point the customer may have
9:15
a different MS or ad server and you need
9:18
to sort of adjust in the old days
9:21
pre-agentic a consulting firm would come
9:24
in and that looks like a beautiful 6
9:26
month 24 month engagement of doing all
9:29
the ETL and other things that I know you
9:31
know so well. So what does that
9:33
engagement look like now?
9:35
>> Yeah, it's it's a perfect it's a perfect
9:36
question and it kind of leads to my
9:38
second differentiatorial as well, but
9:40
I'll answer your question first, which
9:41
is that uh the cheeky answer is well, we
9:45
use AI, right? And um and the the the
9:49
non-cheeky answer or the more serious
9:51
answer is we've actually configured
9:53
these modules that that do use AI to do
9:57
that mapping. That is what the semantic
9:59
layer that I was talking about involves
10:01
right think about what the consultants
10:03
used to do six that those six-monthlong
10:05
projects they would go in and interview
10:07
people. Hey how does it work? out of
10:08
them and then they would inspect the
10:10
systems and then you know we would get
10:12
data engineers in there who would write
10:14
the SQL to do the transformations etc
10:16
etc etc all of that now can be done
10:19
using AI what you need though is you
10:22
need to you need to seed the AI you need
10:24
to tell the agents hey here is the
10:27
domain you're operating in here are the
10:29
types of things that that customers will
10:31
configure when it comes to an ad server
10:33
or an MS etc and and then you let the
10:37
agents do their work to figure out h how
10:40
the customer's business maps to your uh
10:44
your model. Now the important thing over
10:46
here is you will not get it right 100%
10:49
of the time for 100% of the customer's
10:51
business. So how do you allow the
10:54
customer to inject their expertise,
10:59
their intelligence, their specific
11:01
workflows into this whole process and
11:04
that's what we call that's where I come
11:05
to my second differentiator which is
11:07
build with we have designed our platform
11:10
not just the canonical data model and
11:12
the semantic layer but the the way in
11:14
which our platforms applications and
11:16
agents get configured through this mode
11:19
of operation we call build with Remember
11:22
when we used to uh in the last 20 years
11:24
or so there was always this conundrum
11:27
that enterprises had to go through which
11:28
was build or buy right and both had both
11:32
had pros and pros and cons. We think
11:34
there is a third way and we call it
11:36
build with which gives you the customer
11:40
all of the benefits of SAS
11:43
plus the benefits of configuring the SAS
11:46
platform exactly to operate in in the
11:49
way you want without taking the
11:53
maintenance burden yourself. Kana does
11:56
all of that for you. So that's our
11:57
second differentiator James and we we
11:59
find that it's it's uh working really
12:02
well with uh uh our current customer
12:05
implementations where even after you've
12:08
done all the setup etc. customers always
12:10
want to say well but there is this one
12:12
report or one workflow that is so unique
12:15
to us that we just want to implement
12:17
ourselves. Well how do you allow the
12:19
customers to do that without without
12:22
staffing an engineering team from your
12:25
company do that for them? that that's
12:27
not a good model. So we've created this
12:29
whole uh motion called buildwith which
12:32
is a combination of uh software and
12:36
agents that allow customers to to to def
12:39
to to create their own extensions if you
12:41
will to the product using conversational
12:44
AI. So that's our second maybe third
12:46
differentiator. James
12:47
>> got it. Uh you are a relatively young
12:51
company and I'm sure you're trying to
12:52
put wins on the board. I think you just
12:54
announced uh that you've been working
12:56
with a partner and I see the press
12:58
release around that.
12:59
>> Yeah, we just you want to talk about
13:01
that?
13:02
>> Sure. Sure. Yeah, it's uh so we just uh
13:04
we've been working with time uh for a
13:06
few months and uh there are two products
13:10
that we've been working with them on
13:12
with more to come uh in the future. and
13:14
the the two that we're working with
13:16
their uh ad ops ad sales organization
13:20
both in the pre-sale uh portion as well
13:23
as the postsale version. So one product
13:25
is media proposal generator which helps
13:28
sales teams turn briefs from agencies
13:31
and advertisers into proposals. Um the
13:35
second product is we call campaign
13:37
orchestrator which helps adops teams
13:40
post uh post sales once campaigns are
13:42
running how do you keep track of
13:44
everything that's going on? How do you
13:45
optimize for delivery? How do you
13:47
minimize um or eliminate make goods? Um
13:50
how do you minimize underd delivery etc
13:53
etc. So that's the campaign orchestrated
13:55
bit. The both pro product both products
13:58
have their own challenges when it comes
14:00
to uh specific specific customers cuz
14:03
every publisher every media company has
14:05
their own rules around how they want to
14:07
manage aes how they manage inventory how
14:09
they manage their product catalog rate
14:11
cards etc etc. So that's that's really
14:13
the work that we've been doing with time
14:15
and they're a very happy successful
14:17
customer.
14:18
>> And tell me about the breadth of
14:20
solutions that you're trying to bring to
14:22
market because you know you've talked
14:24
about the fact that there's this
14:25
opportunity to bring AI and you've
14:27
talked about the problem statement of
14:29
all these different ad servers etc. And
14:32
now you've given a couple of very
14:33
specific examples of the things that
14:34
you're doing for time. Is that where
14:36
you're focusing on or is the aspiration
14:38
larger?
14:39
>> It's much larger James. So we we have a
14:43
uh first
14:45
one of the things that really excited us
14:46
about building in this space again is
14:49
that because because of AI uh being an a
14:54
big accelerant in the software
14:56
engineering process we could we could
14:58
have a much bigger vision than uh than
15:02
before. So to give you some context, the
15:05
entirety of Krux, everything we built at
15:08
Krux is two agents and applications in
15:12
the Kana platform. Two, right? And and
15:15
we have uh we have seven applications
15:17
now, both that span both publishers and
15:20
marketers. Um and now of course when you
15:24
have multiple products or applications,
15:26
the question always is, well, where do
15:28
you start? And uh we've architected the
15:31
and designed the applications to work in
15:33
a way where depending on the customer's
15:37
specific pain point, the thing that's
15:40
keeping them up at night the most, the
15:42
most urgent pain point, we can start
15:44
over there. So actually at a time we
15:47
started with with campaign uh co-
15:50
orchestrator. Um although their bigger
15:53
challenge was in media proposal
15:55
generation because campaign orchestrator
15:58
was an easier implementation to get
16:00
started. One of the things that is is
16:03
hard these days is nobody trusts AI and
16:06
the events of the last 2 3 weeks or
16:09
month or so have not made that any
16:11
easier right so you have to first build
16:14
a lot of trust with your users and
16:16
that's another way in which we
16:17
differentiate James is we have we have
16:19
this whole chassis that has governance
16:21
and guardrails and and L&M orchestration
16:23
and all of that built in that users can
16:26
use to uh get comfortable and confident
16:30
that the AI is actually giving them
16:32
answers that they can trust. They can
16:35
provide real-time feedback. That
16:36
feedback gets uh um gets incorporated
16:39
into their into the intelligence lead
16:41
that we have for them um etc. But to to
16:45
answer directly answer your question,
16:46
the ambition is much much bigger. It
16:48
spans both publishers and marketers and
16:50
we have customers uh in in both in both
16:53
categories right now. But uh but there
16:56
is a uh deployment path if you will. We
16:59
start over here then we go to the next
17:01
step then we go to the next step.
17:03
>> Tell me what the future looks like not
17:05
just for Khan but for Kana but for the
17:08
industry. I know uh Tom Shaveis who
17:11
we've referenced has a a pet saying
17:14
which I actually love which is we're not
17:16
going to pave over the cart path which I
17:18
interpret as we're not going to just
17:20
make the the old workflows faster. So,
17:24
one um concern that a buyer might have
17:27
or I might have is, "Oh my goodness,
17:29
I've got to get another application.
17:31
I've already got six that I'm working
17:33
with." Is there a world where uh we end
17:37
up with most of a organization being
17:41
able to work in less systems than they
17:44
do now? And that's just from the
17:45
application layer. What else changes
17:48
ideally in the industry and in your best
17:50
customers going forward?
17:52
>> Yeah. So I think I think it's two
17:53
things. One, we absolutely
17:56
don't want to pay the car path. Right
17:59
now the the flip side of that is the
18:03
point that you mentioned which is well
18:05
are you going to come in and just give
18:07
me a completely new way of doing things.
18:08
That's that I do things this way. So the
18:11
first order of business is just to help
18:13
them do what they do better more
18:15
efficiently. Um
18:18
uh work with them to identify outcomes
18:20
etc etc. and then and then work with
18:23
them to uh to to lead them down a path
18:28
which imagines new possibilities, new
18:30
ways of working, new ways of doing
18:32
business, new products that can be
18:34
introduced, new workflows that can be
18:36
built. Maybe you eliminate, oh, you were
18:38
first going to step A and doing X and
18:41
then going to step B and doing Y. Maybe
18:43
you don't need to do that. You can
18:45
directly go to step C because AI does
18:47
all of that for you. So a lot of that
18:49
depends on the customer's appetite. Um a
18:52
huge part of this James is also uh
18:56
change management which I'm sure you're
18:58
uh encountering as well in your
18:59
consulting work which is uh
19:03
people are scared about about the impact
19:06
that AI is going to have on their jobs
19:07
and and livelihoods even sometimes. So
19:10
helping them navigate that, helping them
19:12
uh implement that change in a
19:15
constructive manner is also part of what
19:17
we do as at Ghana. Yeah. I I think I
19:20
gave a speech on stage in Austin uh a
19:24
number of months ago where it was
19:25
primarily on change management and data
19:28
governance and my guidance was go slow
19:32
and make sure that you understand what
19:34
your systems are doing and make sure
19:36
that you have an audit log. And since
19:39
you said data governance, I assume that
19:41
there's an audit log built into all the
19:42
stuff that you're building.
19:44
>> 100%. There's like audit logs up to
19:46
bazoo, James. It's in in there. That's
19:48
the thing, right? Like these days it's
19:50
it's I don't want to say software
19:52
engineering has become easier. It has,
19:54
but but you have to take all of that
19:57
that advance that advancement with
19:59
caution because you can really boil the
20:02
ocean and you have to you have to know
20:03
how to boil the ocean. So yes, we have
20:05
audit logs. We have a we have a detailed
20:08
record of every agent execution, what
20:10
permissions it executed as, what data it
20:13
had access to, what tools ran, the whole
20:15
nine yards in our in our systems that we
20:18
can make and we need that actually cuz
20:21
depending on the industries you're
20:22
working in, regulators might might need
20:25
access to that and customers have to be
20:27
able to provide it. As you've gone
20:30
through this journey re-engaging with
20:34
the martekch landscape and trying to uh
20:37
deploy agents into it, what is surprised
20:40
you the most?
20:41
>> Great question. Uh the thing that
20:44
surprised me the most is the same
20:46
problems that we were solving uh 20 25
20:51
years ago are still there, right? Like
20:54
people are still struggling with how to
20:56
manage rate cards efficiently. people
20:57
are still struggling with how to do
20:59
inventory management, avail forecasting
21:01
properly, uh, and all of that. That's on
21:03
the publisher side. On the marketing
21:05
side, the problems we solved at Krux,
21:08
um, are still problems, you know, it's
21:11
just it just they're slightly different
21:12
flavors and whatnot, which is both, uh,
21:16
exciting and encouraging, but also um,
21:19
you know, it like where we really want,
21:22
this is this is to the point about not
21:23
paving the car path. We really want to
21:25
go beyond that now and I hope that we
21:28
can lead our customers on that journey
21:30
with this all of these new toys that are
21:32
available to us. Now in a similar vein
21:35
of sort of what surprised you if I'd ask
21:38
you where would you take a contrarian
21:40
position? So what is something that you
21:44
believe that is different than the
21:46
majority of people in either ad tech or
21:49
or AI development?
21:51
Yeah, it's a so again I don't know
21:53
whether it's too so contrarian but I
21:57
don't believe that AI is going to take
22:01
over like full autonomy. I don't
22:03
advocate for that. In fact, we like to
22:06
say the A in AI stands for augmented. So
22:11
it's augmented intelligence, not
22:12
artificial intelligence. Um yes, use the
22:16
tools, but don't forget human judgment.
22:19
And it's it's something that's been
22:21
brought to the for recently. Uh all of
22:24
this talk about AI slop meat proxying
22:27
and all of these things that people are
22:29
throwing about where you know uh you ask
22:32
somebody for something they just put in
22:33
the AI and give you the answer. No,
22:35
that's not what we want. We actually
22:36
want humans to exercise judgment. The
22:38
same applies for software and agents and
22:41
applica agentic applications as well. So
22:43
our view is that we that augmented
22:46
intelligence is is going to it's going
22:49
to be the norm and we've designed our
22:51
platform and all of our go to market and
22:54
all of our implementations around that.
22:57
So again my goal in doing these sessions
22:59
is to educate myself about other
23:01
companies. So I've tried to ask you
23:03
things that I think others would want to
23:05
know uh as we come to the end of our
23:08
time. What other things about Kana and
23:11
where you are now, where you're going,
23:13
do you think are of general interest to
23:16
people in the space that I haven't that
23:18
we haven't got into yet?
23:20
>> Yeah, I think look my my my
23:23
advice or or answer uh when I'm asked
23:26
these types of questions always
23:29
just imagine everything is possible now,
23:33
right? So don't don't hold any of your
23:35
ideas back. Um, and you will and but
23:39
look for the right partner. Look for the
23:41
partner who can build with you uh as you
23:44
as you and depending on what problem
23:46
you're trying to solve. Kana may not be
23:48
the right partner. Of course, we're very
23:49
upfront about that, but but um challenge
23:53
your partners. Challenge your partners
23:56
to to explain to you how they're going
23:58
to help you sense,
24:01
decide, and act. any AI platform that
24:04
you uh that you work with needs to be
24:07
able to do all of those three things.
24:09
Help you sense make sense out of
24:10
information decide hey here's what you
24:12
need to do and then act make it so make
24:15
those decisions effective in whatever
24:16
system they need to be effective in
24:18
>> Vic thank you for taking the time to
24:20
talk today as we come to the end where
24:23
can people learn more about you and
24:26
about kana
24:27
>> so kana our website kana.ai AI. I'm
24:30
easily found on LinkedIn. Reach out to
24:32
me. I'm also at Vive at.ai. Uh I love to
24:35
interact with people who have similar
24:37
ideas and or want to talk about all of
24:39
this stuff. So please reach out. Thank
24:41
you for everybody in the audience who
24:43
stayed to the very end. I am James
24:45
Deika. I am the yield doctor. Thanks for
24:48
joining us.