Latest / The Yield Doctor with James Deaker / How AI Is Reshaping Digital Advertising – A Deep Dive with Alanna Laforet

Transcript

How AI Is Reshaping Digital Advertising – A Deep Dive with Alanna Laforet · The Yield Doctor with James Deaker

0:00
I am James Deika. I am the yield doctor
0:03
and today I'm going to be talking to
0:05
Alana Lafere about where AI is going and
0:09
where it's going to have the most impact
0:11
in digital advertising over the next
0:14
year or two. Alana is one of those
0:17
people who has been everywhere in
0:19
digital media. She did a tour of duty
0:22
through doubleclick, right media, yahoo,
0:25
google, admelt, condai, nest, omnicom,
0:29
and most relevant to this conversation
0:32
is a former head of the IAB tech lab and
0:36
is currently the CEO of Lafere
0:38
Productions, helping a number of
0:40
companies in the space navigate their
0:43
challenges as they look at the need for
0:44
revised strategies around AI. Alana, as
0:48
I read out all those companies, the most
0:51
amazing thing is that you and I, I don't
0:53
think ever worked at the same company at
0:55
the same time. But regardless, welcome
0:58
to the podcast and thank you so much for
1:00
joining me. Thank you so much. And we
1:02
have not worked together. It was just so
1:04
crazy to me. But yes, thank you for
1:06
having me. Excellent.
1:08
um you know, we're going to get into
1:10
speculation of where AI might go, but
1:13
but let's just start with what's
1:15
actually happening on the ground right
1:18
now. Um how would you describe the
1:21
biggest ways that AI is transforming
1:24
digital advertising right now?
1:27
Yeah. Oh goodness, what a topic. So,
1:30
everyone is in learning mode at the
1:33
moment. I think the biggest
1:34
transformation is on the planning side.
1:37
There seems to be a big shift in
1:39
thinking about ways to create media
1:42
plans using AI tools and also in tandem
1:47
consolidating data uh for buying
1:50
purposes. So that's interesting in and
1:52
of itself. Um and then there's the whole
1:55
conversation about just in general
1:58
agentic AI implementation. So everyone
2:01
on LinkedIn is talking about how to talk
2:05
to your respective AI instantiation be
2:09
it cla be it chat GPT everyone's sharing
2:13
prompts to get the most out of an
2:16
advertising uh focused conversation with
2:20
with an AI with their AI of their choice
2:23
which I think is really fascinating um
2:26
and so you know today I everyone's
2:29
testing and learning. There are certain
2:33
gains already, you know, especially on
2:35
the planning media side. Um, on the
2:38
providence side, people are thinking
2:40
about ways to start
2:43
tracking either digital assets, creative
2:48
um, photos if you're a publisher. So, I
2:51
do think there's a lot going on. I just
2:52
think it's also a bit as usual the wild
2:55
west. Nick, one thing that would be
2:57
useful to talk about is what's actually
3:00
happening within companies because I've
3:01
got asked questions repeatedly over the
3:03
last few weeks. Is there a disconnect
3:06
between the amount of talk that's
3:07
occurring and the number of actual
3:10
implementations particularly outside of
3:13
implementations within the biggest war
3:15
gardens? I'm going to, you know, set
3:17
aside Google and Amazon and those sort
3:20
of have the team of data AI on the back
3:22
end. when you think about these you know
3:25
small AI startups are you seeing that
3:29
many are actually getting traction in
3:30
terms of actual implementations or is it
3:33
more still in the experimentation stage
3:36
from my experience these startups are
3:41
still in test mode most of the large
3:42
companies I will say though you know
3:45
publicist has core AI and you know
3:47
they're going to invest quite a bit of
3:49
money I think it's 300 million is what
3:51
was reported um into the instantiation
3:54
of their respective AI solution for
3:57
clients. You know, that right there is a
3:59
big company investing and really
4:02
investing year-over-year, not just, you
4:04
know, testing and learning. So, that's
4:06
really interesting. Um I do think a lot
4:08
of the startups, and I won't name them
4:09
because I don't want to say that they're
4:11
not where they need to be. I don't think
4:14
that's fair, but I I do think that
4:16
everyone is kind of in this marketing
4:18
mode where they're talking up a storm
4:20
about everything that they're doing. And
4:22
I'm and I when I kind of lift the hood
4:25
up and look, a lot of it's still test
4:26
mode on the startup side. I think the
4:29
big companies are are are playing a bit
4:31
deeper because they can um and I do
4:35
think you know the holding companies in
4:37
in particular specifically publicist in
4:40
this example because they've been so
4:41
vocal about it that they are investing
4:43
which I think is fascinating.
4:45
So you you and I have been in the space
4:47
a long time as we just made a reference
4:49
to uh and every few years you see a wave
4:52
of innovation around a new technology.
4:55
Uh obviously this right now we're
4:57
talking AI. I saw one report that listed
5:00
something like 77 different ad tech
5:02
startups focused on AI. um what is good
5:05
for the investor community potentially
5:08
and it may be good for the long-term
5:10
innovation but in the short term we may
5:13
see fragmentation within the industry uh
5:17
just at a time when if we weren't having
5:19
AI cons you know innovation one might
5:22
argue that the industry is ripe for
5:24
consolidation so with with all these
5:27
startups are we potentially going to see
5:29
confusion in the market again until we
5:32
start to see winners emerge again. Did
5:34
we ever not see confusion in the
5:37
market? Um, I was thinking about this as
5:40
I was driving today in LA. Uh, the term
5:45
stack, we've used the term ad tech stack
5:48
for a very long time. Um, because you
5:51
and I have built and been in the
5:52
industry for quite quite a bit. Um, I
5:56
think that term is changing because
5:58
we're not layering anymore long term. I
6:02
think, you know, this is a time where we
6:04
need to start learning about the new
6:05
technologies that are coming our way and
6:08
thinking about how to implement them in
6:10
the right ways. And so our inclination
6:14
as an advertising community is to make a
6:17
layer cake, right? Just keep layering on
6:20
solutions, be it data, be it bidding
6:23
strategies, you know, we have header
6:24
bidding, we have first look in the ad
6:27
server, we have blah blah blah. It's
6:28
like this. And then on those layers can
6:31
be different implementations of things
6:34
that needs to change that is causing so
6:36
much havoc in our world because we've
6:40
just kept piling on and what we have now
6:44
is a great industry but it needs to be
6:46
optimized and refreshed and so
6:49
technology like AI you know there's
6:52
other things too but like AI you know AI
6:55
we can use it for multiple different
6:57
things we can you know Joseky when he
7:00
was at Media Math, he had this big thing
7:02
where he talked about bringing in your
7:05
own algorithm. That was we talked about
7:08
it at Right Media. We were, you know,
7:09
back in the day and everybody kind of
7:12
went down that path and bang that drum.
7:16
But now everyone has their own algorithm
7:18
because you can use these open-source
7:22
algos and pay for the paid version if
7:24
you want to to start combining the
7:27
prompt AI software with the execution AI
7:31
software like a replet um which I use to
7:34
create software programs and that's
7:37
really cool because you become that that
7:40
solutions provider where before you had
7:43
to rely on that layer cake to give you
7:46
what you need um as a either an
7:49
advertiser, agency, publisher, whatever.
7:52
So, I do think that that is going to be
7:53
broken apart, which will which will
7:55
cause a bit of chaos, right? Because
7:58
there'll be more solutions coming into
8:00
the market that are more homegrown
8:02
versus um you know, the traditional
8:06
corporate insertion, if you will. So, so
8:09
let me just play that back to you. I
8:11
didn't necessarily think we were going
8:12
to go there, but it's an interesting
8:13
concept. And so, so let me so when you
8:16
say the layers, you know, we talk about
8:17
an SSP layer, a DSP layer, the creative
8:21
optimization, the analytics, etc. And
8:23
within every one of those layers,
8:25
there's, you know, a couple of clear
8:27
players you have to work with, and there
8:28
might be 10 or 20 different vendors. And
8:30
so that's why everybody ends up having
8:33
to work with everybody. What I think
8:36
you're getting to, and I've actually
8:37
heard others allude to this, is this
8:39
idea that with the right type of AI
8:42
implementation, we may end up closer to
8:45
an end toend stack, but with a series of
8:47
plugins around specific agents that you
8:50
can use in specific circumstances. And
8:52
you just mentioned the algorithm. Is
8:54
that sort of the the change in the whole
8:56
landscape that you're talking about?
8:58
Yeah, I mean, I think the plug-in idea
9:00
is correct. And I think what's important
9:03
to to realize is people can make their
9:06
own plugins a lot easier now because of
9:10
the speed of development, right? Like
9:12
you can use these AI development tools.
9:14
And I I use Replet as my example because
9:16
I've been using it, but there are
9:17
others. Um, as well as the prompts,
9:21
right? And asking questions and doing
9:23
your research. And you can create your
9:25
own IP off of that using your own
9:29
imagination. you can use those tools to
9:31
create IP a lot faster than before and
9:34
that opens up a whole world that we
9:36
didn't have necessarily before in adtech
9:39
right so plugins will be more um robust
9:42
than they were before yeah um one of the
9:45
areas well one of the questions I get is
9:47
where will we see the most impact and
9:49
the least impact and I've taken the
9:52
position that in the short to
9:53
medium-term we'll see the least impact
9:56
around the delivery optimization and
10:00
audiences because what we used to call
10:02
machine learning I think is a proxy for
10:05
the fact of a lot of intellectual sort
10:07
of heavy lifting went on in those areas
10:09
and to me there's less immediate payback
10:12
and that there's more payback in areas
10:15
like ad operations maybe on the
10:18
analytics side if you can do sort of no
10:20
code type an analytics are you broadly
10:23
in agreement with that or how where let
10:24
me put it where do you think the most
10:26
impact's going to be when open AI opens
10:29
up their uh prompt windows for
10:32
advertising, that's going to be number
10:34
one, right? Their dwell time is what, 13
10:36
minutes a user, that's insane, right?
10:39
That you know that let's flip it a
10:42
little bit potentially. You know, the
10:44
planning tools that are being created
10:45
using AI, creative optimization, I think
10:47
those you'll have you'll have gains
10:49
there, right? because you're you're
10:51
thinking about things in a holistic way
10:52
and as long as you're modeling correctly
10:55
and using the right data to model
10:56
correctly um you will have optimization
10:59
gains on the the yield side and on the
11:01
planning side. But flip it like so let's
11:04
say these solutions open up advertising.
11:07
We've been looking for dwell time to go
11:10
up. We've been looking for users to be
11:13
specific. uh we'll be looking for users
11:16
who are doing something interesting and
11:17
and and looking for intent. these prompt
11:22
software be it cloud be it chat GPT they
11:25
have intent they have demographics if
11:28
they have you know login I think they
11:30
have I think CHP has like 20 million
11:32
paid users right now um I think they
11:35
have like aund I don't even know 20
11:38
million per month um it's you know I
11:42
think that's where from an advertising
11:44
perspective where are the audiences
11:46
going and what value can we as an
11:49
industry bring to those audiences. And I
11:52
think, you know, that's a really
11:53
fascinating conversation to have. That's
11:56
again not where I expected that you
11:58
would go. So, let's drill a little bit
12:00
on that because in in that world that
12:02
paints the AI as almost a publisher the
12:06
way that, you know, Google started out
12:08
with a very clean interface. It's just a
12:10
search box. And of course, there were
12:12
tons of ways that they became an
12:13
advertiser downstream. you uh you're
12:18
basically taking the position that as an
12:20
aggregator of time and attention the
12:22
natural extension is actually as a
12:25
publisher around the AI. Yeah. Yeah. I
12:28
mean like 13 minute session times is
12:30
amazing right? No publisher has that
12:32
maybe Facebook may maybe right but not
12:35
in the same
12:36
way. It's interesting and you'll have
12:38
intent coming in because you're asking
12:40
asking questions.
12:43
Yes. All right. Uh, again, I hadn't
12:46
hadn't thought we were going to go that,
12:48
but it is an interesting I I've got to
12:50
assume though that the at least in the
12:53
short term, the advertising value may be
12:57
less than subscription value. Although
13:00
again, interesting trade-offs there in
13:02
terms of especially for the people who
13:03
aren't prepared to pay the 20 or $200 a
13:06
month for the the more advanced version.
13:08
Might they be prepared to do that for an
13:10
advertising enhanced experience? Well,
13:13
it's also the demographics. I mean, to
13:15
play devil's advocate on my own comment,
13:16
the demographics are 18 to 34 for the
13:19
most part, right? The cool thing is it
13:21
looks like women females and males are
13:24
split identified as female and male or
13:26
split 50/50, which is interesting in and
13:28
of itself. So, there's that. But yeah,
13:31
you know, there's there's other
13:33
conversation points around that, but I
13:34
think it's interesting um who who's
13:37
who's going to benefit the most when we
13:40
think small players versus big players.
13:43
That could be either small publishers
13:44
versus big publishers, small advertisers
13:47
versus big advertisers. I mean, you're
13:49
taking the position that, you know,
13:51
you're you're using AI yourself to to
13:54
develop, you know, programs, to develop
13:57
product. uh you're pro you're not alone
13:59
in that. And so are we going to see the
14:02
democratization of a lot of the industry
14:06
or as we've seen so often um there's
14:10
some
14:10
democratization until the big world
14:13
gardens get involved and then they take
14:15
over again and maybe it's a new world
14:16
garden. Where where does this go? I
14:19
think
14:20
democratization will not happen. I think
14:23
what will happen and this isn't I'll use
14:26
my own experience the minutia the
14:30
tedious
14:31
tasks get easier right you can create
14:34
workflows in a better way you can create
14:36
agents to help you do whatever task you
14:39
want to accomplish be it optimization or
14:41
what have you but you're teaching it if
14:43
it's something that you want to create
14:45
um you're doing research research
14:47
becomes easier as long as you check your
14:49
work because part of the issue is
14:51
hallucinations Right? You don't want to
14:54
just rely solely on, you know, the
14:57
answers you're getting because God knows
14:59
if it's correct. Um, which I've been
15:02
burned before. I mean, I won't name the
15:04
software product or the AI installation
15:07
that my we were using um to test
15:10
creative, but I had asked my husband
15:13
actually asked it to sharpen a photo of
15:15
our daughter and it put an axe into
15:19
her. I'm like, no, no.
15:22
That's not what we meant. But thank you
15:25
for playing, you know. So, you know,
15:27
there is a ways to go. I do think what
15:31
is going to be front and center in my
15:35
opinion is the concept yet again of
15:39
imagination. These tools open you up
15:42
where you can start focusing on
15:44
innovation as opposed to the minutia
15:48
building the rails that support your
15:50
idea.
15:54
ulating on that and again because this
15:56
is the the yield doctor podcast focused
15:58
on pricing and demand. uh if we if we
16:02
think about who the the sort of winners
16:04
and losers are when it comes to
16:07
advertising AI overall net positive or
16:11
net negative for I guess I'm going to
16:14
say the current digital publishers
16:16
because I could have just said digital
16:17
publishers in general leaving it open to
16:19
to new incumbents but for the existing
16:21
digital publishers positive or negative
16:26
h I would say I would say
16:30
um dangerous at the moment for
16:33
publishers because it depends. So you
16:35
were talking about yield, right? And the
16:38
publisher side, they're providing the
16:42
supply, right? And the data, which I
16:45
always differentiate in my head because
16:47
I'm old school.
16:48
Um there's a danger there of yet again
16:52
thinking about supply as a
16:55
multiplier and you optimizing to create
16:59
more supply. And that's where I think we
17:03
get into the danger area again. It goes
17:05
back to like the old school notion of
17:07
refresh tags, right? like there needs to
17:10
be when a publisher is using any sort of
17:13
algorithmic
17:15
optimizations there needs to be a
17:17
thought towards quality and I think it's
17:20
quickly could be
17:22
that we yet again optimized
17:25
towards mass consumption and mass
17:28
impression volume versus quality if that
17:32
makes sense. It it does. But but let's
17:35
go a little further because there was a
17:37
story I want to say last year and it was
17:40
Sports Illustrated I want to say got
17:42
caught out publishing stories that were
17:45
essentially AI generated and they were
17:47
crediting it as if it were a real
17:49
person. Is that the type of example that
17:51
you're talking about in terms of
17:53
creation of content that may or may not
17:55
be real or are you going in a different
17:56
direction? That's one that's one that's
17:58
one aspect and I think by the way in the
18:00
sports illustrated example if they got
18:02
in trouble because they did to your
18:04
point they didn't slug it properly. If a
18:07
publisher wants to create AI generated
18:09
we just had decrypt and I was publisher
18:10
of decrypt. We did have an AI column and
18:13
we slugged it appropriately and it was
18:15
fine. Everyone knew that it was
18:17
generated by algorithm and it was okay.
18:19
Right. But journalistic rules dictate
18:21
you have to slug it. Yeah. Right. So,
18:25
you know, as long as it's clearly
18:27
marked, I think that's totally fine. Um,
18:32
you don't want to fool or defraud your
18:35
user and your advertiser. So, that's
18:38
number one. But even one layer down, you
18:42
know, there's this whole notion of of,
18:44
you know, generated traffic and and and
18:47
server and publisher impression farms
18:49
and all that lovely stuff that we've
18:51
talked about forever. You know, you also
18:54
could use algorithms and AI to do that,
18:57
right? So, you don't want to like start
18:58
generating impressions that are just
19:00
utter useless nonsense as well.
19:05
It's not talked about a lot, but I have
19:07
to imagine again in in the short term,
19:10
we're going to see more of that because
19:12
this is making it easier to generate,
19:15
you know, content or users that look
19:18
like they're they're human and and
19:20
they're not. And so it's the arms race
19:23
between the the fraudsters and the, you
19:25
know, sort of the gatekeepers in terms
19:27
of who really is human. And I've got to
19:29
imagine that world's getting more
19:31
difficult in the short term. Well, it's
19:32
we've always had bots, right? As we all
19:34
know, now we have armies and agents of
19:36
AI bots that could be consuming content
19:39
and driving up impression value. And
19:41
that just needs to be talked about and
19:44
added to all the, you know, the
19:45
companies that watch fraud like TAG.
19:49
um and just have an industry industry
19:50
conversation about it.
19:53
uh what are you most excited about when
19:55
we have because I know that you're
19:56
excited to have this conversation as we
19:58
got into there are so it's such a
20:01
multiaceted conversation which I think
20:03
at least I'll speak for myself why why
20:05
I've stayed in this industry is there
20:07
are so many interesting problems uh this
20:10
lay adds a new layer onto it what are
20:12
you most excited about for the next year
20:15
I think this conversation illustrates
20:18
why I'm so excited we you know we kind
20:20
of went all over the place and we we you
20:24
know I think I surprised you, you
20:25
surprised me with questions and it's
20:27
good because I think we're all to your
20:29
point we're learning again right um I
20:33
don't purport to be an AI expert I
20:35
purport to be an AI
20:38
uh in training um scholar if you will
20:43
right I mean I think it's fun to learn
20:46
new things it's really fun to think
20:48
about ways to improve our industry
20:51
um you know maybe fulfilling the things
20:54
that we've wanted to do for so long like
20:57
an algorithm in a box right your own ad
21:00
decisioning in an easy way you know but
21:02
there's barriers to entry too so I think
21:04
like one of the things that I'm also
21:05
curious about is lowering the cost of
21:08
all this right because I think I think
21:11
open ad pays like five billion a year to
21:14
keep everything running that's a lot so
21:18
you know we're used to spending a lot in
21:19
adte we're used to tech debt. Um, but
21:23
you know, I think that the pro the
21:25
opportunity is to a be innovative and
21:28
and b lower that tech debt, too. Alana,
21:31
I've enjoyed this immensely. If people
21:34
want to find you, uh, what's the best
21:37
way either in real life on social media
21:40
or in and we haven't even touched on the
21:42
number of nonprofits that you're
21:43
involved in, but what's the what's the
21:46
best way for people to to track you? Oh
21:49
my goodness. Um, I you can reach out to
21:52
me on on LinkedIn. Uh, you can reach out
21:54
to me. Uh, it's Alana
21:58
[Music]
22:00
stillmagazine.xyz. Feel free to reach
22:02
out. I'd love to continue the
22:03
conversation with anybody who wants to
22:04
chat about these things. I think it's
22:06
really fascinating. And yes, I also run
22:09
a farm. So, if you're interested in
22:10
farming, how can I talk about that, too?
22:13
I I've I want to do some podcast just on
22:17
the other interesting things that people
22:18
are doing with their lives, but we I
22:20
don't have space for that right now. But
22:21
good luck with the farm and the farmers
22:24
market and uh have a good rest of your
22:26
day. Thank you. Great chatting with you.
22:29
[Music]