Latest / AI DTC WTF / From Prompt Engineer to AI Architect: Florian Litterst's Full Agency Playbook

Transcript

From Prompt Engineer to AI Architect: Florian Litterst's Full Agency Playbook · AI DTC WTF

0:00
If you work in D2C and you use AI and
0:05
you're wondering what the
0:06
F is going on every week,
0:08
this is your podcast, the AI D2C WTF
0:11
podcast, your home for
0:13
tactical tips, strategies, and
0:14
ideas that you can implement right now in
0:17
your AI workflows to make
0:18
your brand or agency more money.
0:22
All right, welcome to another episode of
0:24
AI D2C WTF. Glad to have
0:28
Florian Litter's German
0:29
legend on the show today to talk about
0:31
all the things that he is
0:32
doing with AI and in his business
0:35
and agency. Florian, how's it going, by
0:38
the way? It's good to have
0:39
you. This must be the most
0:40
complicated last name you've ever said,
0:43
right? The German, my last name. No,
0:46
sounds very complicated.
0:48
Absolutely not. Yesterday, I had Marin,
0:52
who is from Croatia, and he
0:55
is a very well-known Croatian
0:57
marketer, and his last name is Isch Dwan
1:00
Tick. I had obviously been
1:05
mispronouncing this for years,
1:07
and I probably just did again. So it was
1:10
tough, you guys.
1:11
Thanks for having me, guys.
1:13
Glad being here. Yeah, I'm Florian. I'll
1:16
just do this very quickly. I run an
1:19
agency here in Germany
1:20
doing paid social since 2012-ish. So
1:24
quite a long time. We know each other
1:25
since, I don't know,
1:27
2000, what? I don't know, 15, maybe? So
1:31
long. Long, long, long time ago. And
1:33
yeah, playing around
1:35
with lots of AI stuff in the agency to
1:38
find new ways to navigate
1:40
through this AI shift, basically,
1:42
for our clients and for ourselves. Yeah.
1:46
So Florian, like, what are
1:47
you, how are you using AI now?
1:49
What are the things that you're doing
1:50
that you've seen the most impact on? For
1:52
us, a big breakthrough
1:54
actually started last year. We did a team
1:57
workshop, and we discussed
2:00
about how can we make use of AI?
2:02
What's the best way? What's the best
2:03
strategic way of using AI? And
2:06
we stepped back a bit, and we
2:08
thought about how can we think of AI, of
2:12
using AI in different
2:13
roles, in different departments,
2:15
and different use cases. And that brought
2:17
us to, okay, we have creative. In our
2:20
agency, when you run
2:21
paid social as a brand or as an agency,
2:24
it doesn't matter. You do need creative
2:25
strategy. You do need
2:27
surely some kind of copywriting. You do
2:29
need design and video editing. That's the
2:31
biggest part you do.
2:33
And analytics and reports. And in all of
2:39
those roles, basically, we developed
2:41
workflows, tools. We're
2:42
using multiple tools for that, that help
2:44
us with strategy for
2:46
strategy. And quite lately, two weeks
2:48
ago, maybe, we implemented some AI agents
2:51
that helped us to find new ideas,
2:54
basically, for creative
2:55
strategy done in cloud. We have workflows
2:59
for image generation, and
3:01
we have workflows for video
3:02
generation. And when it comes to
3:04
creatives, I do believe the best case
3:08
right now is like a hybrid
3:10
case where you use real content and AI
3:13
content and mix this
3:15
together. That's basically a quick
3:17
rundown of what we do in each department.
3:20
Cool. So there's, I think
3:22
there's a lot of interesting
3:23
stuff there. But what I also heard a lot
3:26
of agency owners say is
3:28
that it's when they start
3:29
to see kind of real value out of it. And
3:41
it's, it's easier to grasp. When you guys
3:44
had this team meeting,
3:45
and you're kind of decided how to go into
3:47
this, what was the first kind of things
3:48
you started doing? And
3:49
what kind of tools did you
3:51
start playing around with?
3:52
I mean, that was back in June last year.
3:56
So in AI world, that's quite
3:58
a long time ago, I would say.
4:00
So we honestly, we started mostly using
4:02
ChatGPT, using deep research for creative
4:05
strategy, finding new
4:06
ideas, new angles, new hooks. But we knew
4:11
where in which direction we need to go,
4:13
because we know which
4:14
type of content performs on, on ads. So
4:17
that's, that's where I
4:18
started being like an architect for,
4:20
for prompt. So I was more like a prompt
4:22
engineer and engineered
4:23
different prompts to help us,
4:27
yeah, make use of ChatGPT to find new
4:29
angles on the strategy side.
4:30
And then we played around with
4:31
lots of image generation tools. I mean,
4:34
back then the first big breakthrough, I
4:37
think, was the image
4:38
generation of OpenAI, which was not so
4:41
good from today's perspective. Then
4:45
Google introduced Nano
4:46
Banana and from there took off basically
4:48
an image generation and what
4:49
we can do in image generation.
4:51
And with video, it's still kind of
4:55
okay-ish, I would say. I
4:57
honestly don't think there is
4:59
already a big breakthrough in video
5:01
generation. There are new models like Sea
5:03
Dance, which came out two
5:05
weeks ago, I think, which is quite good
5:07
and quite strong. But video is
5:10
like, it's hybrid. So we have
5:13
real content with, made with content
5:15
creators, and then we mix
5:16
and match this with AI content,
5:18
which helps us basically to produce stuff
5:20
that we couldn't produce with content
5:22
creators or that they
5:23
basically forgot to produce. And this
5:26
solves a big problem. Yeah.
5:28
So let's, let's get into specifics on
5:30
this. So of the things that you've built
5:32
thus far right now with
5:33
your agency, what are the things that are
5:35
the most high leverage
5:36
and, and specifically like
5:38
exactly how, how did you set those up?
5:40
You talked about agents. Is that maybe
5:41
where you want to start?
5:42
I would say, yeah, I would say. So how
5:44
is, how are you building them with Cloud?
5:45
Like how specifically
5:46
are you setting them up? Like with, you
5:49
know, what do they look like? What kind
5:50
of tools are you using?
5:52
I would say it's, it's the agents, um,
5:55
which are quite new because they are more
5:57
on autopilot right now.
5:59
Up until then, it was like, we need to
6:01
prompt something, we
6:02
need to wait, and we need to
6:04
guide it a bit more. And now it's way
6:06
more on autopilot. What we
6:08
did is we created a database
6:10
in the backend using Airtable. And in
6:12
Airtable, we have like different
6:14
informations about our
6:15
clients, about the brand, about the
6:17
products. And based on this...
6:18
And what kinds of stuff are you giving
6:19
them in Airtable? You're
6:21
giving it like the, the name,
6:23
the vertical, the website, you're giving
6:26
it like the persona. What
6:30
else are you putting in there?
6:31
Yeah. Persona. And it depends on the
6:33
brand, but sometimes you
6:34
need to go down on product
6:36
levels. Sometimes you can just say on the
6:37
brand level, depends on
6:38
the brand and each case.
6:41
And based on this, we've created, um,
6:44
yeah, two different
6:45
agents. One agent is, that's,
6:49
I would say that's unique to the EU
6:50
because in the EU, in Meta's ads library,
6:53
you have a transparency
6:55
guideline, which means in the EU, you can
6:58
see how much reach an
6:59
advertiser had on one ad or the
7:02
other ad. That's unique in the EU
7:03
compared to the US. And
7:05
based on this, my idea was, um,
7:08
why can't we create an agent that always,
7:11
um, analyzes the ad
7:12
library and checks for
7:14
ads with the highest reach from up to
7:18
five competitors,
7:19
analyzes those, and then pushes
7:22
back angle ideas that we might be
7:23
missing. So we always like monitoring
7:26
competitor brands from our
7:28
clients. Their ads library, we always see
7:31
which ads, uh, or we see
7:33
based on the transparency, uh,
7:34
rules and ads library, which of those ads
7:37
get the highest reach. And
7:39
I would say in most cases,
7:41
not in all cases, but in most cases, the
7:43
highest reach means the best
7:45
performance, not in all cases,
7:46
but in most, uh, and based on this, we,
7:49
this, this information, so
7:50
the ads, uh, scraped using
7:53
APFI, which is like a tool for APIs,
7:56
basically. Um, we're scraping
7:59
data out of the ads library.
8:00
We are also, um, uploading, um, the
8:03
images into air table, and we
8:05
are also pulling the videos.
8:06
The videos is then, the videos then
8:08
pushed to whisper from open
8:10
AI, which is transcribing the
8:12
video. And then we have everything that's
8:14
set basically in the
8:15
video in air table. And then
8:17
Claude is going for everything and is
8:18
ranking. Okay. This is a good
8:19
angle. This is a good angle.
8:21
That's my idea, uh, for you guys that you
8:23
create this or that angle
8:24
that you may be missing this or
8:26
that, or this is good by done by whatever
8:29
type of brand that's quite helpful.
8:31
So you're having, so the, the agent or
8:34
the agent goes through and you're
8:36
continuing to feed it all
8:37
this information. And other than the
8:39
tools that you mentioned,
8:40
you're using Claude as the
8:41
one that's plugging in air table, APFI
8:44
using whisper. Are there
8:46
other things, other tools in
8:48
there that are helping like pull it
8:49
together? And if so, what are those?
8:51
It's, it's pretty much that it's a
8:53
Claude, which is like the architect and
8:55
which is building my ideas.
8:57
Um, air table, um, APFI and whisper.
9:01
Yeah, that's it for that case. We also
9:03
have another case where we,
9:05
so this is like the, um, the ad library
9:08
agent and we, we have kind of the same
9:11
idea. We had kind of
9:12
the same idea with customer reviews and
9:15
we are not pulling customer
9:16
reviews from our clients. We are
9:19
pulling the reviews from their
9:21
competitors, which means we always see
9:24
the top three positive and top three
9:27
negative comments from the competitors,
9:29
from our clients, which is
9:31
kind of the same idea. We are,
9:33
we always get ideas for like, okay, these
9:35
are angles you're missing.
9:37
These are angles that are
9:39
maybe like a anti pitch angle, whatever,
9:42
because they are lacking
9:44
in quality here or there.
9:45
You could go into that direction as an
9:47
angle with, with, uh, for
9:49
your client and, uh, basically,
9:52
yeah, uh, get this information out of
9:55
Trustpilot, TrustShops and
9:56
Amazon. Also done with APFI
9:58
and Claude and air table in the
9:59
background. Just a follow up. Um, um, a
10:04
lot of people I talked to
10:04
today use Superbase versus, uh, air
10:06
table. Um, I know they both do the same
10:09
things. Uh, but what I'm
10:11
interested in trying to understand is,
10:13
um, um, what are you doing then with all
10:16
the learnings and the data
10:17
clock gives you based on what it's
10:20
analyzing? How, how do
10:21
you, how do you use that?
10:23
So right now it's right now it's
10:25
basically pushing, uh, the idea. So it's
10:28
pulling this data every 30
10:29
days for each of our clients, not daily
10:32
every 30 days, because you
10:33
need new data that needs to be
10:35
analyzed. And that's why we decided to go
10:37
with every 30 days. And then
10:38
it's pushing like a message
10:41
into the Slack channel. Basically we have
10:43
a Slack channel for each client, uh, in
10:45
our agency and there,
10:47
our agent is pushing a message with, I
10:50
analyzed whatever 50 ads
10:51
here, the top ads. I think this
10:53
is good. This is good. So I asked Claude
10:55
to score those ads or to score the
10:58
reviews. Uh, and based
11:00
on the scoring, it's basically sending us
11:02
the ideas. And then we have
11:04
a set of ideas and a set of
11:06
ads that we might be, uh, producing. Are
11:09
you then kind of looping back what is
11:12
created in terms of how
11:14
it delivers? I mean, is it like a
11:16
feedback loop of, was it actually a good
11:18
idea? This did this ad
11:20
perform or is that maybe kind of maybe
11:22
the next level of what
11:23
you guys are thinking on?
11:24
This is not done by AI yet. Um, this is
11:28
done by our creative
11:28
strategist. Um, and I think a creative
11:31
strategist, a good creative strategist is
11:32
always looking what worked
11:35
in the past and is always
11:35
looking into like what might work in the
11:38
future. And this is
11:39
basically just at the end, it's just
11:41
in another angle that we're testing, but
11:44
the idea is coming from AI
11:45
and we save lots of time, uh,
11:48
in research here. Yeah. So you have, so
11:51
you have the, this agent set up. So the
11:53
other stuff you have
11:54
in terms of, uh, image generation and
11:57
video generation, talk
11:58
through those and how those are
11:59
set up and, um, yeah. And what kinds of
12:02
things they're, they're
12:03
doing and what they're built on.
12:05
Well, we played so much with different
12:09
models and different tools
12:09
out there. And we came to the
12:13
conclusion that, okay, the most important
12:15
thing is having good raw
12:17
material, a good kind of
12:19
imagery from each product to basically
12:22
train a model or to basically use an
12:26
image as, as a reference
12:28
image, for example. That's why we
12:30
started, um, a month ago, uh,
12:33
doing photo shootings, AI ready
12:35
photo shootings in a studio, which
12:37
basically means we get the product and
12:39
we, um, create images of a
12:42
product in each angle with perfect light.
12:45
Um, so we can use this image, this
12:48
product image to train
12:50
models or to use it as a reference image.
12:52
That's where we are right
12:54
now. Um, and then the tool we're
12:57
using or the platform we're using is
12:58
called, uh, kreia.ai. That's what we use,
13:02
uh, kreia.ai with K,
13:05
not C, not C, kreia.ai. AI. And in, in
13:08
that, uh, that's like Higgs field, for
13:11
example, it's pretty
13:11
similar to Higgs field. Basically it's a,
13:14
it's a wrapper. It's a kind
13:15
of a wrapper, which, uh, um,
13:18
is plugged into, uh, every model through
13:20
the API. And in CREA, we can
13:23
basically create images with
13:25
nano banana, nano banana 2, uh, whatever,
13:28
every, every version of nano banana,
13:30
every version of, uh,
13:32
the latest video models like C dance, for
13:35
example. And then in CREA, we built,
13:38
or we're building workflows in a way that
13:41
we say, okay, we do know
13:42
which type of imagery works as a
13:45
performance ad. So for example, product
13:48
enhance is classic one. Uh,
13:50
each product shot in a hand from
13:52
like a eco perspective is usually a top
13:55
performing image ad. Um, and
13:57
we created a workflow just to
13:58
create multiple images of a product in
14:02
different hands, basically
14:03
different settings, different
14:04
hands. So that's a workflow for us where
14:06
we basically upload the
14:08
imagery from different
14:09
products. It's always the same workflow.
14:11
And then it creates
14:11
different images that are from our,
14:14
that we know perform as an ad. So it's
14:17
like the workflow is built on,
14:21
built, uh, built in a way that we created
14:24
each, uh, different
14:24
workflows for each use case where we
14:26
know, okay, this type of imagery works.
14:29
Um, and, uh, yeah, that's
14:31
for image and for video. It's
14:32
basically my experience with video is
14:35
like so-and-so sometimes it's
14:37
quite good. Sometimes it's, um,
14:39
slop, AI slop. And, uh, for our
14:42
customers, um, I'm trying to not create
14:44
AI slop. Um, that's why we started,
14:48
uh, in a way that we try to, um, try to
14:51
widen in a way, our B-roll library for
14:54
video content in a way
14:55
that we create content that we haven't
14:58
created before with videos.
14:59
And this is done kind of like
15:02
an art director, I would, uh, art
15:04
director, I would say. So we create
15:06
different, um, different
15:09
images based on the images, uh, the
15:11
photos we've created, and then those
15:13
images, we animate those.
15:15
And then based on this, it's like a
15:16
sketchbook, I don't know. Right. So we
15:19
have different images,
15:20
different images getting animated, and
15:21
then we have five different B-rolls and
15:23
these five different
15:24
B-rolls are then fed into the video team
15:27
and mix and matched
15:29
with our existing content.
15:31
That's, uh, what we do there. So, and so
15:33
basically a client signs
15:34
with you and they come and say,
15:36
these are our products and you're, you
15:37
look at the top products and then you
15:38
make sure to load those
15:39
into the like static workflows that you
15:41
have, which is a product holding in hand.
15:42
And they're the things
15:43
that you have seen that have worked
15:45
before. And then with the video,
15:47
basically taking like
15:49
animated gifs essentially. And then
15:51
you're building those into the
15:53
preexisting video assets that you
15:55
have is like the hook. Is that
15:56
essentially kind of what it is? Exactly.
15:58
Exactly. Yeah. Exactly. Yeah.
16:00
And, and you're going through, so when
16:04
you're building these out and
16:07
you're, you know, starting to
16:09
build them new creatives and new creative
16:12
direction and stuff like is, is there,
16:14
there's the research
16:15
phase that's taking place that we talked
16:17
about before with the
16:18
agent and that's then spitting
16:20
out as well, some ideas for just the
16:22
creative strategist or does
16:23
the client also see those or
16:26
like what's, you know, cause the
16:28
innovation on not finding new personas
16:30
and stuff, I assume maybe you're
16:31
telling the client as well, but so maybe
16:32
some of that's
16:33
internal. I'm just curious.
16:34
So what I'm trying to build is, uh, I
16:38
would say for most processes,
16:39
it's still human in the loop,
16:40
which means we are still screening the
16:43
ideas and we still say, okay, that's a
16:45
good idea. That's a bad
16:46
idea. Even with designs or with images
16:48
that are created based out of those
16:50
workflows, we always
16:52
need to make sure that there's a good,
16:53
like a mood board, um, at
16:56
the beginning, which is like
16:58
style trading for the AI, I would say
17:00
done by human. And we always
17:02
make sure that the output is
17:04
basically, um, uh, controlled by human,
17:07
um, to really make sure
17:09
that this is on brand and this
17:10
is not sloppy in a way. Be really clear.
17:14
Sorry. One more question.
17:14
So you're, you're taking the,
17:18
let's say that you did that with video.
17:19
So you take some of this
17:21
stuff that the agent has learned,
17:22
the mood board, you then take it to this
17:25
software that you mentioned, or the one,
17:26
I forgot what the name of
17:27
that one is. That is kreia.ai.
17:29
KREA. And then you say to, to kreia, Hey,
17:35
this is all of the stuff that we know.
17:38
Here's a, here's a preexisting video
17:40
asset that we want to be added
17:42
or want to be like included in
17:44
this. Create the, a 30 second version of
17:47
this for me. Or like, what are you
17:48
telling kreia to do?
17:50
You, you're speaking about videos, right?
17:52
Yeah. Yeah. Yeah. Okay.
17:54
It's like kreia is basically,
17:57
it's no difference in using Google V or
18:00
free or for whatever, or
18:02
kreia. It's, it's basically the
18:04
same model. Um, what we do is we, we, we
18:07
create for videos, we create like a
18:09
sketch board. So it's like
18:11
having someone who's thinking of the
18:12
video, we, which scenes do we need in a
18:14
video? And we create
18:16
images for each scene. This is based on
18:19
the idea. And then
18:20
based on this, we say, okay,
18:22
we think C-Dance 2 is the best model to
18:24
animate this scene. And then
18:26
we have 10 different scenes.
18:28
And those 10 different scenes is
18:29
additional B-roll footage for our video
18:31
team. That's how we,
18:32
how we think about this. And the, the,
18:34
the 10 different scenes are being
18:36
sketched out by what
18:37
you're learning from the agent. In some
18:39
ways, yes. In some way it's
18:40
still made by, by us, by our
18:41
creative strategies. It's by you're just
18:43
saying, okay, actually this is what we
18:44
know. This is what the
18:45
client says. And then like, here's the
18:47
scenes and there, and
18:49
these little scenes.
18:50
I do have one good example, which is
18:52
quite hard to, how to shoot as a video
18:55
actually. So usually we,
18:57
we, we still do lots of videos with
18:58
content creators out
18:59
there. We have roughly 300 that
19:01
are working with us. Um, and we have a
19:04
client that's selling socks, sports
19:06
socks. The issue with
19:07
sports socks is it's really hard to, as a
19:10
content creator, it's
19:11
really hard to shoot a video,
19:12
wearing your socks, running and shooting
19:15
a video from, from the back, uh,
19:18
basically recording yourself
19:19
running. That's very, very hard. So you
19:22
definitely need a video team
19:24
or someone else, uh, recording
19:26
that person, which is expensive, which,
19:28
um, makes costs going up. And what you
19:31
can do is you can create
19:32
this scenes or multiple scenes where a
19:34
person is wearing socks and
19:37
it's running basically. And the,
19:38
video is, uh, recording this person from
19:42
the back, from the side, whatever. This
19:45
can be perfectly done
19:46
by AI. And then you can basically mix and
19:49
match the scene of whatever the content
19:51
creator who's talking
19:52
about the product with this B-roll
19:54
footage. So that's quite a,
19:56
quite a good example, I think.
19:57
But, but, but, but then again, which I
19:59
think is really interesting is that
20:00
you're using that so your
20:02
experienced team can, can make the
20:05
decision if it's good enough or good
20:08
enough and give feedback.
20:09
So it's, it's, it's just like you're
20:11
working with a AI photographer, uh, and
20:15
the rest of the workflow is
20:16
pretty much the same that it's always
20:19
been. Is that kind of correct or?
20:22
Kind of. I do think that human taste is
20:24
still super important. Um, but everything
20:27
else should be enabled and
20:29
enhanced by, by AI. And, uh, that that's
20:32
how we think about this.
20:33
Yeah. I also built another quite
20:36
fun workflow, something else I did. I did
20:38
build quite a fun
20:38
workflow, which is pulling the best
20:40
performing image ad out of the ad account
20:43
and creating iterations
20:45
based of this best performing
20:47
image ad, which is quite easy for AI. So
20:50
it's basically downloading the best
20:52
performing image ad,
20:53
uploading the best performing image ad
20:54
into Google drive done in cloud. Uh, then
20:57
it's sending the image
20:58
into cloud cloud is a reading basically
21:01
what's on the image. It's
21:03
creating new hooks basically. So
21:05
text overlays for the image. This is then
21:07
again, double checked by
21:09
another, uh, um, prompt basically,
21:11
because sometimes the grammar was wrong
21:13
or the spelling was wrong. And then
21:15
basically it's creating
21:16
three different versions of the same
21:17
image with different hooks
21:19
and overlays. This is end to end
21:21
automation of, of creatives. But from all
21:25
what we guys probably know is that
21:27
iterations are not like
21:29
the thing where you should spend too much
21:31
time on because of meta
21:32
andromeda and that are there
21:33
creative diversification. So I would not
21:35
spend too much time on this.
21:36
That's why I thought, okay,
21:38
it might be a good idea to automate this
21:40
end to end. That's, that's what you do
21:42
with, with image ads
21:43
and iterations. Yeah. It's pretty cool.
21:45
And, and, um, are, are the, are these
21:48
versions, um, performing
21:50
well when you're testing them? Like the,
21:52
the flow is working, but it is giving
21:54
good quality as well.
21:56
It's, it's working. We still double check
21:58
every output because I
22:00
mean, I would not upload random
22:02
stuff into ads manager, at least not when
22:04
I'm the one who's responsible, uh, for
22:07
our clients. So I always
22:08
want to have someone double checking
22:09
this. Um, and yeah, their
22:12
output is good, but, um, from
22:13
everything we see before this cloud
22:16
transition and AI shift,
22:18
iterations are not the thing that
22:20
are driving performance in ads manager
22:22
that much, at least sometimes yes,
22:24
sometimes no. That's why we,
22:26
before we had this workflow, we decided,
22:27
okay, we don't put, put much
22:29
effort into iterations just
22:31
because of that shift on meta side. And
22:33
we spent more time finding
22:34
new concepts. And that's why I
22:36
thought it might be a good idea to have
22:38
to still like autopilot in
22:39
the background who's creating
22:40
animations, but we don't spend human time
22:42
on this. Yeah. Yeah. And how are the
22:45
clients reacting to,
22:47
to this kind of shift? Do they, do they
22:50
know, do they care about it
22:52
being more AI in the picture?
22:56
How, how are, how are they feeling? One
22:57
year ago, I would have been afraid,
23:00
honestly. Um, and I thought,
23:02
um, that, um, that maybe clients don't,
23:05
um, at least for, I'm
23:06
speaking now for the German speaking
23:08
market, which might be different to the
23:09
international market, just to be, just to
23:12
have this disclaimer
23:13
here. And looking at the German market, I
23:16
thought that, uh, most people
23:18
think that we basically spend
23:22
less time managing their things, which is
23:24
not true at the end. And, uh,
23:27
then our, the perceived value
23:28
of what we do as an agency is going down.
23:31
That was my, uh, very, to
23:33
be honest, but turns out it's
23:36
completely different. People really love
23:38
when we do this. People
23:39
are really open. Uh, people,
23:42
all clients always, uh, yeah, they, they,
23:46
they're super happy when we
23:47
create new stuff. For example,
23:48
another thing we do, and all of our
23:51
clients are, are really loving, loving
23:53
this is we, we create CGI
23:56
type of content, which is basically you
23:58
take the product and you
23:59
put it somewhere in front of,
24:02
I don't know, the Empire State Building,
24:03
for example, that's
24:05
trademarked, would not do this,
24:06
but it's possible technically with AI. So
24:08
you take your product and you put it
24:10
somewhere on a monument,
24:12
someplace that most people know. And this
24:15
is the CGI type of stuff,
24:17
which is quite cool. It's still on
24:19
brand and people love this when we do
24:22
this. That's one example
24:23
where we understood that our
24:26
clients really like AI stuff. Before I
24:29
was hesitant and I was skeptical, but
24:33
right now I'm all in. I'm
24:34
going all in into this and I'm pushing
24:35
this really hard, um, to the team and the
24:38
team is really open.
24:39
Um, our clients are super open. We do
24:41
lots of content around this and we are
24:43
very, very open to our
24:44
clients that we use AI in different
24:46
scenarios. They trust us that we are not
24:49
creating AI slop. Uh, and
24:51
I take care that we are not doing this.
24:54
Um, and that's why I
24:55
think it works quite well.
24:56
So if it's done right, AI basically makes
25:01
the work that you're doing
25:02
more effective because you're
25:05
able to give the client more creative
25:08
options and therefore more scale, right?
25:12
Because scaling is so
25:14
much dependent upon persona development
25:16
and the expansion of
25:18
matching personas into that. Um,
25:23
what do you think in terms of places
25:26
you're spending your time
25:28
and what you're seeing and
25:30
hearing from other leaders? I think
25:32
there's, there's a lot of people throwing
25:33
shit against the wall right
25:34
now, but what is your opinion about the
25:38
future of this moving forward and where
25:41
you are spending time and
25:43
where you'd recommend others spending
25:45
time? Because I think a lot of us
25:46
initially with this have gone
25:48
into better agency owners have gone into
25:51
spending time on reporting
25:52
and automating the reporting and
25:54
spending time on creative iterations and
25:56
obviously building up more iterations and
25:58
workflow and processes
25:59
around that. The next phase of this to my
26:03
opinion, I will, I won't
26:04
poison the well. So I have an idea
26:06
of like what I think is going to happen
26:07
with this, but I'm curious of your
26:09
opinion of where it's,
26:10
where it's headed and where people should
26:12
be spending their time. Um,
26:14
you, uh, you mean spending
26:15
time as an agency or in general in, in
26:17
this digital market? As a
26:18
marketer, let's just say,
26:19
as an agency, you could, you could
26:21
certainly answer it any way
26:22
you'd like. Um, I do think,
26:25
I mean, there's a, there's a saying, and
26:27
this is quite common that the person
26:29
using AI is definitely
26:30
outperforming the person not using AI. So
26:32
there's no way not using AI
26:34
definitely. And I would be
26:35
curious. I would be open. Um, um, and I
26:39
would, I mean, at the
26:40
end I, I would spend time
26:43
thinking about what's actually a good
26:45
system and how can I
26:46
connect together multiple,
26:49
multiple things, databases, tools,
26:51
whatever, to find a good and effective
26:54
workflow. So being more like
26:57
an architect being more like, uh, I mean,
27:00
pretty clear, you need to
27:01
move from, uh, the day to day
27:04
work more to a strategic work. I mean,
27:05
that's nothing new that
27:06
happened already with media buying
27:08
five, six, seven years ago with iOS and
27:11
every, every change there.
27:13
Um, this is, this is, um,
27:15
accelerated by AI and you need to, you
27:18
need to be the one who's really
27:20
connecting multiple dots and
27:21
maybe multiple channels insights from
27:23
multiple channels. For us,
27:25
this means we, we are, we evolve
27:28
more into like a consultancy, which is
27:32
also doing agency stuff
27:33
basically for our clients. So we help
27:35
with multi-touch attribution,
27:37
cross-channel, branding, upper funnel, I
27:40
don't know what else,
27:42
which is usually that's not what we do,
27:43
but we know this stuff and we can help
27:45
with this stuff and we
27:50
really, I think with AI as an agency, you
27:52
have the option to really cut
27:55
down costs for your clients.
27:57
Um, and that's what we try to do with the
28:00
photo shoots we do. This is
28:01
a photo shoot you do once,
28:03
and then you have perfect imagery for
28:06
endless AI creations of your, of your
28:09
product. Um, and that's
28:11
how we try to help being more agile in
28:13
terms of content creation. Um, which
28:16
basically means you can
28:17
prompt your, your product into every,
28:19
every situation. And before you are, you
28:21
would need to create a
28:22
photo shooting. Um, that, that, that's
28:25
how I think about this in general, I
28:27
would be interested in your
28:28
take, uh, because I mean, the end goal
28:32
of, I would say every ad platform and
28:35
also meta clearly is to automate
28:37
ad creatives and to basically give us
28:39
your money. We do everything for you.
28:41
What do you guys think? Do you
28:43
think that's, that's happening and when
28:48
would it happen? What do you think?
28:49
What's your take on this?
28:50
You started the loop. So you just, just,
28:53
uh, what's, what's the next phase and
28:56
also specifically around
28:57
these platforms and automating
29:00
everything? Yeah. So I think there's a
29:02
couple of different pieces of
29:03
this. One is as it relates to the way
29:07
you'll use it as a marketer right now,
29:09
it's been used in silos.
29:11
So we've been utilizing it in creative
29:12
because that's where so
29:13
much of us spend our time in
29:14
terms of scaling an account. I think that
29:16
it'll continue to expand
29:18
into utilizing it for full
29:20
funnel. So one of the most powerful
29:22
things that we can do is build a better
29:23
offer as build better
29:25
landing pages that match personas. And
29:27
then the ads match those
29:28
personas too. So that it's a full
29:30
thing where if you have a particularly,
29:32
and then you, and then you cross
29:34
reference this with financials.
29:36
So you have a product that if you're
29:37
selling water bottles that
29:38
has a better margin than a
29:40
product that if you're selling
29:41
microphones and looking at things on my
29:42
desk, um, then you're
29:44
going to push a, the water bottle more
29:47
and AI will say, Hey, look,
29:49
your margins are better on this
29:50
product and you have more of this in
29:52
stock and your new customer
29:54
acquisition on this particular
29:55
product or this particular funnel is X
29:57
over the last 50, 60 days. I recommend
30:00
that you start doing
30:01
more creative for this and it will then
30:04
spool up persona development that's
30:07
around the water bottle
30:09
angles that you have in the markets. And
30:10
then it'll have corresponding landing
30:12
pages or corresponding
30:13
advertorials that are built and you'll be
30:15
able to say, yes, try this
30:17
idea. Um, and then it basically
30:20
builds from there looking at, you know,
30:22
competitive analysis and continues to
30:23
like go on the things that
30:25
you've been talking about. So I think
30:26
it'll become a more full funnel play. And
30:29
I think that that also
30:30
extends to agencies. I think that where
30:31
it comes with agencies is
30:33
really going to be, how are you
30:35
making the case for AI with your clients
30:39
in agencies to say, look, this
30:40
isn't just me messing around,
30:42
right? And Florian, you talked about AI
30:43
slop. This isn't just us
30:45
trying things. And I think to some
30:46
degree in agencies, this is actually
30:48
people messing around. But I think right
30:49
now, what you want to try
30:52
to go forth with is look, we're going to,
30:54
this is allowing us to go into deep
30:55
research with more ideas,
30:57
and it's allowing us to build our process
30:59
better so that we get to
31:00
spend more time effectively
31:02
analyzing data for you and doing and
31:05
being a better advertiser. It's not just
31:07
advertising all products
31:08
and saying you have a four row as it's
31:10
like looking at them by category, looking
31:12
at them by contribution
31:13
margin, looking, you know, looking at
31:15
sort of the results in relation to the
31:17
financial metrics and
31:18
calculations. And if you're able to do
31:20
that, utilizing AI as an agency moving
31:23
forward, you become a true
31:24
marketing partner. You're not just one
31:26
person that's building the
31:28
meta ads, right? You're doing
31:29
a lot of other things. This could even go
31:31
into things like building
31:34
incremental channels, right?
31:35
Where, look, we have this idea, we're
31:37
going to launch connected
31:38
TV now, and this is how we're
31:40
going to be able to do that. So, you
31:42
know, I think this is, it's
31:44
headed in the direction of
31:46
making it more effective with more people
31:49
and, or with less people
31:50
involved, I think. But with the
31:51
people that you have, it's going to be
31:53
more data that's going to
31:54
really actually be able to really
31:57
help folks. So that's kind of what I
31:59
think. I don't know.
32:00
Thomas, what do you think?
32:01
I agree. And I think that, which I also
32:06
feel like Florian's point
32:08
in this whole conversation is
32:09
around that you're still going to need
32:12
real people with real experience,
32:15
understanding the data,
32:17
understanding the workflows and
32:19
communicating it clearly to the clients.
32:23
So I think that people
32:24
will be more and more valuable, and at
32:27
least people who have experience in this
32:31
industry from before.
32:34
And I also think that it's going to be
32:35
harder and harder for younger
32:37
people to kind of get up to a
32:39
certain level and get that gut feeling
32:41
experience because, you know, younger
32:45
people are just agreeing
32:46
to whatever AI is telling them more than
32:49
us grumpy old farts who are
32:51
just like challenging and no,
32:52
not that way. And you think like this
32:54
and, you know, all that stuff. So I
32:57
definitely think it's going to be
32:58
more of the whole funnel. And I think
33:02
that Meta, Google, everyone
33:04
is going to try to build that
33:06
platform. But I think that the more the
33:11
agencies can take ownership
33:12
to creating their own software
33:14
and their own platform and plug into the
33:16
different APIs, I think
33:18
that's the one who can do the best
33:21
job for their clients going forward. It's
33:24
quite funny because I was
33:25
playing around with Cloud,
33:29
which is basically, it's pretty easy to
33:31
duplicate product detail pages using
33:35
Cloud and then change
33:36
something and messaging on a product
33:38
detail page or any landing page. So it's
33:40
pretty easy to spit that out
33:41
because I had the same idea in like,
33:44
okay, we have the messaging on the
33:45
creative, but it's not
33:46
really, really matching the landing page.
33:49
And the idea was to
33:50
basically create a system that's
33:53
that's using the existing page and
33:55
changing the headline,
33:56
changing the image to,
33:58
to have a better match with, with ad
34:00
creatives, which could be
34:02
done quite easily using Cloud,
34:05
honestly, and Netlify. But
34:08
yeah, that's a different story.
34:10
Yeah, I mean, so much of it is just like
34:11
trying and testing and
34:12
I'm thinking about it,
34:13
but I think you have to make sure that
34:14
you're trying and testing
34:15
things in a strategic framework
34:17
that add into the greater whole. So it's
34:19
not just like, oh, it's like,
34:21
you know, so much of this is
34:22
like, you see this on X, oh shit, I got
34:23
to try that. You see this on X, I got to
34:25
try this, I got to try
34:25
this. And it's like, you can try it if
34:28
it's, but you also have to have a
34:29
destination because otherwise,
34:30
you're just taking all these side
34:32
journeys that aren't necessarily
34:33
improving the overall things.
34:34
It's like, where are the hard costs of
34:36
our agency? Where are
34:37
the hard times that were,
34:38
or where are the hard costs of working
34:40
with a particular client? You know,
34:42
is it client communications? What are the
34:44
things that we can do to
34:44
make that more efficient and
34:45
effective and saving money there? So I
34:48
think it's like, if that
34:48
contributes to that goal,
34:50
then absolutely. Right. But yeah,
34:52
Florian, I appreciate you
34:53
joining us. Florian will
34:56
be able to answer any questions. If you
34:57
have them, you can email me
34:58
andrew at foxwelldigital.com.
35:00
And all of his details and notes, if you
35:03
want to follow Florian,
35:04
are going to be in the show notes. But
35:05
thanks. Thank you for being here.
35:06
Thanks for having me guys.
35:08
Bye.
35:09
Healthy options.
35:10
ããç¨åº¦ æå ä½¿ç¨