Latest / AI DTC WTF / How Tori Rowe Built an AI Operating System for His Entire Agency

Transcript

How Tori Rowe Built an AI Operating System for His Entire Agency · 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
0:18
make your brand or agency
0:19
more money.
0:21
Welcome to another episode of AI D2C WTF.
0:24
Let me tell you, I have an incredible
0:26
episode here for you
0:27
today with Tori Rowe, someone
0:29
that...
0:30
I really respect, Foxwell Founders
0:31
member, who has
0:32
essentially with AI built an entire
0:35
operating system for his agency.
0:37
I mean, he's got an audit maker.
0:39
He's got like so much stuff in here.
0:42
I don't even remember all
0:43
of the pieces he showed us.
0:45
So I'm excited.
0:46
And this episode is definitely going to
0:48
be worth your time to listen to.
0:49
Will, what were your
0:50
takeaways in chatting with him?
0:51
Yeah, I mean, he's a pretty modest guy,
0:54
but this, I will say
0:56
Tori is well ahead of the
0:58
bell curve in more ways than one.
1:00
He showed us a dashboard that he's
1:03
effectively built out
1:04
automated media buying, ingesting
1:06
client calls, having a repository for...
1:10
So much stuff.
1:12
It was so cool.
1:14
It's a little bit of...
1:15
You know, I had a little info overload at
1:17
one point, and then we
1:18
sort of dialed it back
1:19
and he was able to sort of explain
1:21
everything really succinctly.
1:23
And I was just sort of flabbergasted by
1:25
how fast he's moving.
1:27
Yeah, I completely agree.
1:29
So check it out.
1:30
Let us know what you think of the episode
1:31
and hit me up, andrew at
1:32
foxwelldigital.com if
1:34
you have any questions or feedback on it.
1:36
But, you know, we always are trying to
1:38
give tactical information
1:39
to you on these episodes
1:40
and that you're able to just take and
1:42
implement and think about
1:43
and give you some brain food
1:44
on AI.
1:45
So that's what we're here for.
1:46
So let's take it away to the episode.
1:48
So let's take it away.
1:53
Tori Rowe, Floridian legend of AI and
1:56
digital marketing, who I love having here
1:59
and is an incredible founder member.
2:00
Welcome to both of you.
2:02
Appreciate it.
2:02
We are here to talk about deeply AI tools
2:07
that we're using, things
2:08
that we're getting into,
2:09
making sure that you can walk away with
2:11
this show with tactical
2:12
stuff that you can use.
2:14
You know, Tori, you're somebody who's an
2:15
incredible digital marketer that I really
2:17
look up to and respect
2:19
in your thoughtfulness and the approach
2:21
and the completeness of the way that you
2:23
go through thinking through the funnel.
2:25
Right now, what are you absolutely like
2:29
creating on AI that you're loving that's
2:31
really changing results for you?
2:33
Yeah, I mean, there's so much.
2:34
First off, appreciate the kind words.
2:36
There is so much.
2:38
I think the biggest thing that I'm like
2:40
nerding out about is
2:41
probably the reporting side,
2:43
which is drastically different than where
2:45
everybody else is going.
2:46
I feel like everybody's so zoned in or
2:48
honed in on like the creative front.
2:50
There's so many tools.
2:51
There's so much cool
2:51
stuff going on there.
2:53
But like connecting an MCP to like a data
2:55
warehouse is something incredible that
2:57
I've never made an MCU for.
2:59
So like one of the big things we're doing
3:00
right now is mapping all the orders from
3:02
as far as we can reach back,
3:03
first click all the way to the back,
3:05
basically the very last order of someone
3:08
and running an MCP through it.
3:10
Ours is Claude.
3:10
We run it through the Snowflake database
3:12
and basically pull out as
3:13
many insights as we can.
3:14
Right now we have like
3:15
9 billion rows of data.
3:17
There's no way a human can
3:19
actually sort through this.
3:20
So us going through and trying to find
3:23
basically anything we can out of this and
3:26
say like piece this together,
3:27
show me some insights that I wouldn't be
3:29
able to pull on my own, stuff like that
3:30
is super incredible.
3:33
Yeah, there's a lot on the reporting side
3:35
that I'm just like going really deep on.
3:37
Still the hypnotic creative side a lot,
3:38
but the reporting is where I'm spending a
3:41
lot of time the last few weeks.
3:42
So let me just jump in for a sec.
3:43
So you have all of your clients' data
3:47
separately, obviously,
3:48
like in separate warehouses.
3:51
And you're going through
3:52
and you're building report.
3:55
Like what was your reporting before
3:56
versus what is it now?
3:58
And how has it helped
3:58
you make better decisions?
3:59
Yeah, as far as like reporting before,
4:01
standardized spreadsheet stuff, super
4:03
metrics, you know, BigQuery kind of
4:05
stuff, pulling it out,
4:06
thrown into a spreadsheet.
4:08
As good as AI is, you still need like the
4:10
granularity in like a spreadsheet.
4:11
Sometimes I feel like you
4:12
still can't beat it, right?
4:13
So the client facing reporting is still
4:16
like spreadsheet focused.
4:17
But for example, if we standardize our
4:20
naming convention across everything and
4:22
have the exact same UT and parameters
4:23
across every single ad that goes across
4:26
every single ad account, we could say,
4:28
hey, tell me how UGC performs to females
4:31
across every single
4:32
client in the month of March.
4:34
And it would tell us
4:35
exactly what happened, right?
4:37
Tell me how it performs to men.
4:38
Tell me the age of the model, right?
4:41
So if we name the ads, you know, angle,
4:44
model, age range, hair
4:46
color, hook, length, right?
4:50
And just keep going.
4:51
Now, if that's standardized, we could
4:52
say, hey, pull format six.
4:54
Tell me, you know, the like comparisons
4:56
across all of our clients throughout
4:57
March and what has
4:59
the best row as, right?
5:00
Or what has the lowest CPA?
5:01
And then what we're taking on top of that
5:03
is we're tying this to customer lifetime
5:06
value with identity mapping.
5:08
So we'll be able to tell you what the
5:10
customer lifetime value is instantly.
5:12
So as soon as someone finishes an order,
5:14
we push that customer lifetime value back
5:16
and we end up tagging
5:17
that in the warehouse.
5:18
So we're able to say, hey, this is
5:20
generating the highest
5:21
customer lifetime value customers.
5:24
UGC, blonde hair model,
5:25
25 to 34 with this hook.
5:28
Nice.
5:28
I have a follow up there.
5:30
So naming conventions, a lot of folks use
5:36
to ensure that they stay organized.
5:38
A lot of these larger platforms, Motion,
5:41
you know, they rely partly on naming
5:42
conventions to keep organized.
5:44
Say I'm a creative
5:45
agency, you know, for example.
5:48
And there are a lot of different media
5:51
buyers out there who have
5:52
their own naming conventions.
5:54
Have you explored something like using,
5:56
you know, I don't know, let's say Claude
5:59
Sonnet to review the assets rather than
6:03
using a naming convention?
6:04
And if so, like how does that compare to
6:06
like sort of the old fashioned way?
6:08
Like you're outlining
6:09
UTM parameters plus naming.
6:11
Yeah.
6:11
So we're using Gemini
6:13
because Gemini can watch videos.
6:15
I'm not sure about Claude Sonnet, right?
6:16
And so Gemini, we're trying to figure out
6:18
if Gemini can tag these videos, very
6:20
similar to how Motion tags the videos.
6:22
The issue is with it, the granularity
6:24
that we're trying to pull as we get
6:26
deeper and deeper is where
6:27
things get difficult, right?
6:28
And so like there's also not like a QA
6:30
point of this of like somebody has to go
6:33
through and QA all this.
6:34
And when you have 10,000 ads going live,
6:36
so it's easier for us to just dump it
6:38
onto the creative team who already has
6:40
the naming convention.
6:40
And so they're going through and actually
6:42
naming it while they know
6:44
what the file actually is.
6:45
So instead of our media buyers watching
6:46
it refile, instead of us QAing over some
6:49
AI source, I think it'll
6:50
get there, to be honest.
6:52
I just don't see it right now.
6:55
It's something we are working on, but
6:56
it's a tough one to
6:57
unlock, I feel like still.
6:59
Motion surprises me how
7:00
far they've gone with it.
7:02
I mean, they're incredible.
7:03
Totally.
7:03
And I think that like honestly speaks to
7:05
sometimes like the old
7:06
fashioned way is still the best.
7:08
Not everything needs to be totally
7:10
optimized in that sense.
7:12
Again, I agree.
7:13
Like if you do UTM parameters
7:15
and naming conventions, right?
7:16
Like you're going to have a far more
7:17
systematic and like thoughtful database
7:19
than if like you're just, you know,
7:21
either having Sonnet or Gemini sort of
7:24
guess at what you're trying to build.
7:27
Yeah, the way I've been like explaining
7:28
it to our team is basically imagine if
7:30
Triple Whale, obviously they have like
7:32
4,000 people on it, right?
7:33
We're nowhere near this size.
7:34
But imagine if Triple Whale could
7:36
standardize naming convention across
7:38
every single client they have on the
7:39
platform, what type of
7:40
data they would have.
7:41
That's what we're doing internally.
7:43
They should do that.
7:44
It would be insane.
7:46
It would be insane what they
7:48
could pull out of that, right?
7:49
Yeah.
7:50
I mean, I don't know if you guys have
7:53
access to the Motion MCP, but there's
7:54
some of that that's new.
7:57
I mean, that's coming that if you don't
7:59
have access to it, maybe just email them
8:01
and ask if you can get
8:02
access to it or whatever.
8:03
There's a ton of crazy stuff
8:05
that you can pull out of there.
8:06
I think that this analysis, one is, you
8:10
know, yes, this is what it looks like.
8:12
It's specifically this type of ad, this
8:15
type of thing we're putting out there
8:17
that really is going to bring in the
8:19
highest profitable customer.
8:20
I think the future is totally that
8:22
connection between the financial metrics
8:24
of the business and understanding, you
8:27
know, how you, what your inputs are on
8:29
the marketing side and how you bring in
8:31
higher, more high value cohorts.
8:33
So you're building this out.
8:35
This is your, the data warehouse.
8:37
You're looking at this.
8:38
Like, what are some other tools that
8:40
you've used in building this that you
8:43
talked about a couple of options, but
8:45
what are some other connectors or
8:46
anything else to mention there?
8:48
Yeah, it's pretty simple right now.
8:49
It's literally all
8:51
the connections, right?
8:52
All the top funnel paid media kind of
8:53
platforms across the board.
8:55
And also Klaviyo Connects and Attentive,
8:56
all that kind of stuff.
8:57
Post-purchase surveys.
8:58
We're building APIs over from
8:59
post-purchase surveys
9:00
to feed that data in.
9:01
Everything's pulled in through Airbyte.
9:03
Airbyte goes over to Snowflake, stores
9:05
into the Snowflake database.
9:07
And then Snowflake database can kind of
9:09
push anywhere, but
9:10
MCPs over to Snowflake.
9:12
They also have their own MCP on there,
9:13
but we're feeding it back into Claude.
9:16
Claude has so much information already
9:19
about us because we've
9:19
been so deep on it, right?
9:20
Of kind of where the business stands and
9:22
all that kind of stuff.
9:23
So we're feeding it back over there and
9:26
then basically running
9:26
anything we want on top of Claude.
9:28
But you're trying to...
9:31
Our goal as an agency is trying to cut
9:34
down on the actual media
9:35
buying part for our media buyers.
9:38
I don't want people
9:38
uploading creatives 40 hours a week.
9:41
Like, that's not the goal here, right?
9:43
If I can get them to spend 50% of the
9:46
time in data and asking questions, and
9:48
like, this is not a
9:49
knock on media buyers.
9:50
This is a knock on myself too,
9:52
obviously, in the media buyer.
9:53
We are not always the top data scientist,
9:56
most data analytic
9:56
person, right, across the board.
9:58
So if we can just ask simple questions
10:00
and start to understand this data more,
10:03
you're basically just empowering people
10:05
to be able to make better decisions
10:07
without having the guess of what they
10:10
think is working or trying to...
10:12
Hey, this has click-through rate.
10:13
Here's frequency.
10:14
Here's this.
10:15
Here's this, right?
10:15
We're just like, no, this is the ad that
10:17
is driving the most people
10:18
for the first touch point.
10:20
This is the ad that is converting people.
10:21
What are the, just as an aside, but like,
10:24
what are the metrics that you use most
10:27
often that are actually correlated to
10:34
results in growing a company that aren't
10:37
the typical ones that
10:38
you hear people talk about?
10:41
Everybody's on CPMR and stuff like that
10:42
right now, which is just funny.
10:44
It's always switching.
10:45
And the one, we actually ran an analysis
10:47
on everybody last year and the number one
10:48
indicator of a creative that ended up
10:51
scaling was cost per ad
10:52
to cart, which makes sense.
10:54
It's a very simple, you know, creative,
10:56
but it's, you can get ad to carts early
10:59
and not get purchases early, right?
11:02
And so it's just a
11:03
higher top of funnel metric.
11:04
If that is kind of in line with
11:06
everything else, everything
11:07
else seems to fall in line.
11:08
There was not a click-through rate
11:09
analysis that ended up showing that that
11:12
was it or CPMs, nothing like that.
11:15
It was always cost per ad to cart was the
11:16
number one indicator for a good creative.
11:18
Nice.
11:19
I sort of want to, if you don't mind,
11:21
sort of go back to the database AI
11:22
component, because I think
11:23
that's really interesting.
11:25
And I guess my
11:26
question is sort of twofold.
11:27
Like, how did you sort of land on each of
11:30
those different touch points?
11:31
Like, why did you pick
11:31
Snowflake, for example?
11:33
Was it just like a cloud recommendation?
11:34
Or did you see like an advantage there?
11:37
And two, I think what folks maybe get
11:40
confused or lost about it's like, okay, I
11:42
have all this great information.
11:43
Like, I know what's performing now.
11:45
Like, how do they sort of take that to
11:47
the next step to actually generate ads
11:49
that will work, right?
11:50
Like, having the information is great.
11:52
But like, what is that human next step to
11:55
either iterate or build new concepts that
11:58
will actually perform?
11:59
Yeah.
12:00
So great question.
12:01
As far as like, I mean, we
12:03
went through a few steps here.
12:04
First, we tried and not trying to knock
12:06
other companies, but
12:07
we tried Supermetrics.
12:08
It was too slow.
12:09
And it couldn't pull as
12:10
granular data as we wanted.
12:12
So we went from
12:13
Supermetrics, found Airbyte.
12:15
We have a bunch of
12:17
engineers we work with.
12:18
So they were able to basically tell us,
12:19
hey, Airbyte's a good one.
12:20
Then databases, we were
12:22
kind of going through.
12:23
And Snowflake, they have Cortex, which is
12:25
their own AI model,
12:26
right, on top of everything.
12:27
So we were able to use the LLM on that,
12:29
which was, you know, just
12:30
something that we wanted.
12:31
Because I wasn't sure how
12:33
MCPs would end up pulling in.
12:36
So that was kind of like the first step
12:37
of just finding what worked.
12:39
Also, as weird as it sounds, we tried to
12:40
go to the ones that
12:41
like the big guys use.
12:43
Because we knew if, you know, an IPO
12:45
company was pulling in 500 billion rows
12:48
of data, we'd probably be relatively
12:50
cheap to pull in five to nine billion
12:52
rows of data was my
12:53
first thought process.
12:54
I was like, we'd be
12:55
the small guy over here.
12:56
So I kind of wanted that as well, because
12:58
there'd be tools that, you know,
12:59
obviously they're
12:59
using and we're not using.
13:01
As far as like the action points on the
13:05
data, there's a few ways
13:07
to kind of go about this.
13:08
One, you already have this creative
13:09
feedback loop, right?
13:10
One thing we've always looked at is like,
13:12
does the model work?
13:13
Yes or no?
13:14
Because we'll take the same exact script
13:16
and give that to 15 models.
13:18
And it'll always come back to like, this
13:20
model works or this model works, right?
13:22
So that's one way of just
13:24
like starting to bucket the data.
13:26
Into basically digestible places for you
13:29
to say, hey, here's the top model.
13:31
Here's the top hook.
13:33
Here's the top.
13:33
For example, we run a jewelry company.
13:35
This is the top background.
13:36
They switched from all white backgrounds
13:37
to all black backgrounds in February
13:40
because we found out that worked better.
13:42
All of their top creators are black
13:43
backgrounds right now.
13:44
So like simple things like this, if you
13:45
like start to bucket things, I think you
13:47
can have like process by analysis and go
13:49
too deep here for sure.
13:51
But the end all be all here is probably
13:54
feed this analysis over to creative
13:57
briefing, have a creative brief, do the
13:59
analysis on or take that analysis and
14:01
then write the briefs for you.
14:03
Your creative strategist on top of that,
14:05
have them kind of check it, but just
14:06
basically speed up the
14:07
workflow across the board.
14:09
Nothing's 100 percent.
14:10
Things probably 70 to 80 percent with
14:12
that last 20 to 30 percent
14:14
being the human touch point.
14:15
But just trying to get it through that
14:16
pipeline a little quicker.
14:17
I think it's interesting that the idea of
14:20
of taking what I've heard a lot is people
14:22
talking about taking a lot of data and
14:24
putting it in by client and making agents
14:26
right to be able to query, which is
14:28
essentially kind of
14:28
what you're talking about.
14:29
But for you, it's even more.
14:31
It's bigger.
14:31
It has more implications
14:33
for the entire business.
14:34
I could definitely even see a situation
14:36
where it's like you have access to not
14:40
just their Shopify data, but
14:41
like their financial metrics.
14:43
You know what I'm saying?
14:44
Like as a marketer, like because then
14:46
you're able to look at obviously highest
14:48
margin categories and then you're able to
14:50
say, OK, when we do these things in
14:52
succession or, you
14:54
know, let's combine them.
14:55
You know, the black
14:56
background thing is a simple example.
14:57
But the more that we're able to start to
15:00
connect these lines, the better off we're
15:01
going to be, because I think thus far
15:03
what's been happening to a lot of people
15:05
is like that's not good enough.
15:07
If you look at like Zach Stuck, for
15:08
example, right, who's one of my close
15:09
friends, like what he does is and with
15:11
his team is basically he'll go and say,
15:13
that's not good enough.
15:14
Let's do a better offer.
15:15
That's not good enough.
15:16
We're going to do a better offer.
15:17
And then they're going to
15:17
they keep designing the flow.
15:19
So the ad creative, they're testing and
15:21
designing and then making
15:22
sure that the lander matches it.
15:24
Right.
15:24
And that's how they're continuing to
15:25
expand on persona development.
15:27
So like with this, the more that you give
15:29
it, obviously, the more powerful it is in
15:31
moving, moving forward.
15:32
And it's powerful for your team because
15:34
then you're higher
15:35
leveraging them, as you said.
15:36
Yeah.
15:37
So like to the persona thing, it was like
15:38
we can sit there and tell you, hey, 25 to
15:40
34 year old women coming from Instagram
15:43
reels convert at X percentage and have a
15:45
higher customer lifetime
15:46
value when they go to these pages.
15:48
You know, pretty quickly.
15:49
Right.
15:50
I can tell you that in probably 10
15:51
seconds just by a simple
15:53
question over to clock.
15:55
Right.
15:55
So stuff like that of like
15:56
we can dive deeper into this.
15:58
Like I said, I think the biggest thing is
16:00
trying to stay focused and
16:01
not go too deep at the moment.
16:03
Right.
16:03
Of like we're not trying to.
16:06
Yeah.
16:07
We're not trying to
16:08
go reinvent the wheel.
16:11
We're trying to do make
16:12
the wheel a little faster.
16:13
Right.
16:14
Now, will we go down that path?
16:16
For sure.
16:16
At some point.
16:17
Like let's get the basics down and do
16:19
incorrectly and the rest of
16:20
this will fall into place.
16:22
Nice.
16:22
Yeah.
16:22
That makes sense.
16:23
It is so easy to probably.
16:25
And I think folks listening probably can
16:26
relate to this just
16:27
like analysis paralysis.
16:28
Right.
16:29
Like we have all of this
16:30
incredible information.
16:31
Like what the fuck do I do next?
16:33
Right.
16:33
So you were saying the database helps you
16:37
effectively come up with raw scripts.
16:38
And at first when you're talking about
16:39
models, I was like, I thought you're
16:40
talking about different AI models.
16:42
But then I was like, oh, it models,
16:43
creators in the wrong headspace.
16:47
But so what you're saying is you have
16:50
effectively you have the model say this
16:52
is what is working well and maybe
16:53
generate like an initial brief for it.
16:55
And then that brief goes to a human being
16:57
that, you know, cleans the brief up.
17:00
Then that ultimately goes to the creator.
17:02
Is that sort of the flow?
17:03
Correct.
17:03
Yeah.
17:04
Like it's basically, I mean, if you want
17:06
to go deep on it, like we have from the
17:08
very top of like what we build with AI.
17:10
We now have a creator platform that we
17:12
cold email creators on Instagram.
17:14
They can go to this platform.
17:15
They sign up.
17:16
It stores into a backend database for us.
17:18
So our team can filter
17:19
through the creators.
17:20
Right.
17:21
So they filter through the creators.
17:22
We actually push out
17:23
product to those creators.
17:25
Claude writes the
17:25
brief to those creators.
17:27
Now that's like the top of funnel.
17:29
This filters back in after those ads run
17:31
back into the data warehouse.
17:32
And that cycle continues.
17:34
Right.
17:34
And so we say, hey, it's
17:36
a, you know, brunette model.
17:37
It's a guy who's 45 years old.
17:41
Sweet.
17:41
Go back to the creator database.
17:42
And eventually this
17:43
will all be connected.
17:44
You just tell Claude,
17:45
hey, go find these creators.
17:47
Now it goes by.
17:47
And so it's just hopefully cyclical at
17:49
one point where all of
17:50
this is just connected.
17:51
Anything we build with AI, we try to take
17:54
the next step from what we last left.
17:56
And I think I found this out pretty early
17:58
for myself of like, I would go over here
18:00
and build something.
18:01
I'd go over here and build something.
18:02
Right.
18:03
Instead of like piecing things together
18:05
and eventually just kind
18:06
of connecting them all.
18:07
I would go do two things on the opposite
18:09
side of the room, which
18:10
don't benefit you as often.
18:12
Yeah.
18:13
That definitely makes sense.
18:14
Yeah.
18:17
And I think the impetus behind, or for
18:20
many of us rather, is just like, why
18:22
would I have Claude just do this one use
18:24
case when I can have it do 10?
18:26
And like, you know, why would I just make
18:27
this internal when I can turn this into a
18:29
tool for everyone to use?
18:30
It's just like, there's such a rabbit
18:31
hole to sort of go down.
18:33
So how do you sort of prevent yourself
18:34
from going down that rabbit hole?
18:36
And then even more so, like, what, at
18:38
what point do you realize, like, I've
18:40
built something that is
18:41
sufficient for my use case?
18:42
Like, do you, are you sort of coming up
18:45
with a concept beforehand?
18:48
Like, here is what my end goal is.
18:50
And once I hit that, I'm
18:51
not going to move any further.
18:52
Or is it more so you just like
18:54
intrinsically realize
18:55
when you've hit that?
18:56
It's a really good question.
18:58
I will say, like, I'm in the rabbit hole.
19:00
I'm climbing out of it, probably.
19:02
So, like, I've been down it quite a bit.
19:04
I think the easiest way for me is, like,
19:06
I'm trying to solve a
19:07
problem when I go to this, right?
19:08
And so I go to my entire team.
19:10
And, like, we sit down.
19:11
We sat down for, like, two hours one day.
19:13
And it was early.
19:14
And I'm sorry that my
19:15
team had to go through it.
19:16
And I was like, tell
19:17
me all your problems.
19:18
And I just sat down,
19:19
recorded the whole meeting.
19:21
And then basically was like, okay, what
19:22
problems do we need to solve here?
19:24
And what problems are
19:25
adjacent problems that I can go fix?
19:27
And so that was, like, the first thing
19:28
of, like, the internal side of the
19:30
business was, like,
19:31
fixing all of this, right?
19:33
Now, on the external side that you hit
19:36
on, as far as, like, SaaS platforms and
19:37
people kind of pushing things out,
19:39
one of the things we wanted was, like,
19:41
bulk ads launcher, an email writer, like,
19:43
all this stuff of, like, a platform that
19:46
we could use to media
19:47
buy emails, read images,
19:49
do all of our reporting all in one
19:51
singular place instead of clicking.
19:53
I mean, we have, you know, 30-plus
19:55
clients that were, like, clicking through
19:57
ad account to ad account, right, having
19:59
to check everything.
19:59
So I wanted it for us.
20:02
We built that, got it
20:03
probably 50% of the way there.
20:05
And then we hired three engineers, two AI
20:08
engineers and a product engineer.
20:10
And they came in and
20:11
they're overhauling it.
20:12
And so we wanted to make sure it was done
20:13
100% accurate, as
20:15
clear as we possibly could.
20:16
And that is a SaaS
20:17
platform that we're launching.
20:19
And that comes out in four weeks because
20:20
we went to the rabbit hole.
20:22
And I was like, hey, if we want this,
20:24
everybody else will want it.
20:25
Right now we have, like,
20:25
250 signups waiting for it.
20:27
What point, that's really interesting.
20:30
At what point did you say, like, we need
20:32
to sort of bring in,
20:33
like, the big dogs here?
20:34
Like, was it just, like, you found
20:36
yourself spending way too much time in
20:37
cloud code and you were just, like, you
20:39
know, bug, bug, bug?
20:41
Or is it more so just, like, you know,
20:43
I've reached an impasse?
20:44
Yeah, it's, it's,
20:46
everything was functional.
20:47
Even something as simple as, like, the
20:49
bulk ads launcher, right?
20:50
Like, it takes so much
20:51
time to launch creative.
20:52
Of, like, I could get it working.
20:54
It's going through and
20:55
everything like that.
20:56
But, like, think about how good an actual
20:58
engineer or a developer
20:59
with cloud code can be, right?
21:02
And so they're going to
21:03
take it to the next level.
21:04
And so that was kind of the goal of,
21:05
like, if I can get it here,
21:06
they can get it there, right?
21:08
And I want to make sure that it's as
21:10
efficient as possible
21:11
for everyone in the room.
21:12
So that was kind of the next thing is I
21:14
knew that I'm not an expert and I just
21:15
wanted someone to make sure, hey, you
21:17
know, dot the I's and cross the T's and
21:19
say, hey, this is good
21:20
enough for everybody.
21:21
Yeah, it's that kind of line of taking a
21:25
step back and understanding what's going
21:26
to be useful to you
21:27
versus the team is huge.
21:28
And, yeah, it's interesting.
21:29
And good job, by the
21:30
way, on launching a SaaS.
21:31
I feel like if you don't have if you're
21:32
messing around with, you know, AI and you
21:35
don't have a SaaS that you're launching,
21:36
it means that you're you're not trying
21:38
hard enough right now.
21:39
I feel like I was like every day I'm
21:40
like, I could launch that.
21:42
And then I'm just and then
21:43
I'm like, no, this is garbage.
21:44
Nobody's going to pay for this shit.
21:45
But it's but but it is like it's a sign
21:48
of like, oh, wow, this
21:49
is actually really useful.
21:50
This is super interesting and it helps
21:51
you kind of connect dots.
21:53
So under that, what what other things are
21:56
you have you been doing with AI lately
21:57
that outside of this that have just been
21:59
absolutely mind blowing for
22:01
you that have been awesome?
22:02
Weavey is for sure the top thing that I'm
22:05
like wowed by right now.
22:06
So how do you spell this?
22:08
W-E-A-V-Y.
22:11
Let me just.
22:11
Yeah.
22:12
Pretty.
22:13
Yeah.
22:13
Yeah.
22:13
Weavey.
22:14
Yeah.
22:14
Yeah.
22:15
Weavey.
22:15
It's like Weavey Figma or something like
22:17
that is what it's called.
22:19
Everybody like I mean, we've been working
22:21
on AI creatives for, I
22:23
don't know, a year and a half.
22:24
Right.
22:24
I think everybody's always been since
22:25
ChatGPT came out, everybody's trying to
22:28
been like finagling with it.
22:29
And I'll say like they're pretty good,
22:31
but it's still like a
22:33
60, 70 percent hit rate.
22:35
Right.
22:36
Like they're not all perfect.
22:37
You throw a lot in the
22:38
trash and stuff like that.
22:39
This is 100 percent every single time
22:42
exactly what you're looking for.
22:43
It has been absolutely incredible.
22:45
Basically, what it is, is you're building
22:47
a pipeline with prompts, images and all
22:50
of this kind of like basically framework
22:52
to get to your end result.
22:54
So you could the guy who came
22:56
over to us, we just hired him.
22:58
He came from BarkBox and something for
23:00
like BarkBox is like they really struggle
23:02
to go do a photo shoot with a bunch of
23:04
animals and get all
23:05
their toys inside with this.
23:07
Right.
23:07
It's not easy for them.
23:08
So what they did is they ended up
23:09
building this thing
23:10
through Weavey, this pipeline.
23:12
It basically says like,
23:13
hey, pick the dog that we want.
23:15
You know, like Golden Retriever.
23:16
And then they're like,
23:17
OK, pick the plush toy.
23:18
They picked a plush toy and it basically
23:20
has like a 3D rendering of the plush toy.
23:22
And then it's basically
23:23
feeding this into a prompt.
23:24
They're describing the background.
23:26
Then it feeds to another prompt.
23:27
They're describing where
23:28
they want the dog to lay.
23:29
Then they're feeding it
23:29
where they want the toy to lay.
23:31
And then it feeds into an image and that
23:33
image comes out after you run this.
23:35
And you can basically just
23:37
do this on a pipeline nonstop.
23:40
So basically getting Claude Code to
23:43
basically build something to feed the
23:45
briefs and the prompts in here and just
23:46
kind of put it on a flywheel.
23:48
Cool.
23:48
It looks like it
23:49
accesses pretty much any model.
23:52
Right.
23:54
Yeah.
23:54
It's insane.
23:55
Will, you're going to love it.
23:56
Yeah, I'm stoked about this.
23:58
You're using it.
23:59
So like you're doing
24:00
this with clients now.
24:02
And it's who I mean, like what are the
24:04
verticals of the
24:04
clients you're using it with?
24:05
We use it all over.
24:06
I mean, it's honestly like we haven't
24:08
found a vertical we
24:08
can't use it in, to be frank.
24:10
Like I think the biggest thing is
24:11
founders who are open to it.
24:13
That has been the biggest thing that
24:15
we've kind of ran into is like some
24:16
people are like, hey, it's not authentic.
24:18
Right.
24:18
And I get that.
24:20
I get both sides of the table here.
24:21
So it's like hard for me to be like, hey,
24:24
you should use AI or you shouldn't.
24:25
So I mean, supplements is a very simple
24:29
one, kind of easy across the board.
24:30
But also like we've
24:31
used it in cowboy boots.
24:32
We've used it in CPG quite a bit.
24:35
What else we got?
24:36
Apparel is super simple.
24:37
We'll also use it just to do like
24:39
backgrounds for emails.
24:40
Right.
24:40
Like we like really cohesive backgrounds
24:43
for emails of like generating a
24:44
background that is still branded and then
24:46
kind of overlaying the
24:47
template on top of it.
24:49
Like just simple stuff that would
24:51
normally take a designer.
24:52
I don't know, an hour, hour and a half to
24:54
kind of create something like this.
24:55
This is done in five minutes.
24:57
Very cool.
24:58
Sorry.
24:58
I'm just like lost in there.
24:59
I'm lost in their flow.
25:00
So would you find this is sort of give
25:05
sort of like, you know, what is like the
25:07
one use case that folks
25:09
can sort of use it for?
25:10
Like, like it's because it seems like
25:11
very manual in the sense like you need to
25:13
sort of pick your model.
25:15
You need to pick the prompt.
25:16
Like, what would you say is like the one
25:18
sort of like start starting point for
25:20
maybe someone to jump off on for?
25:22
Yeah, I think you.
25:23
Yeah.
25:23
Just start your flow.
25:25
Right.
25:25
If like there's some like very simple.
25:27
Take your hero
25:28
product and start the flow.
25:30
Right.
25:30
Start to build that
25:31
prompt and stuff like that.
25:32
But you can start to basically put these
25:34
in a drop down of like you have 50
25:37
prompts that are already there
25:38
and you're just selecting it.
25:39
Right.
25:40
And so you can start to kind of mix and
25:41
match and do what you want to do.
25:43
So like, for example, back to like
25:44
Artbox, they're just selecting the dog
25:47
they want and they have 50 images of that
25:49
dog that they've
25:50
already like uploaded into it.
25:52
So you're starting to get these
25:53
variations instead of like going to Higgs
25:55
field or Kai.ai or whatever.
25:57
And you're actually like manually
25:58
uploading these things.
25:59
It's basically storing it.
26:00
So, yeah, it's a pain in
26:01
the ass the first time.
26:02
But once you kind of have that template,
26:04
then you're just copying it, duplicating
26:07
that template, changing out the products,
26:08
changing out your prompts and saying, OK,
26:10
this is hero product number two.
26:12
Right.
26:12
And then going back to like what you and
26:15
I have talked about before, well, the
26:16
iteration process is the easy part.
26:19
The real I think the real winner is a net
26:22
new concept from AI is the extremely
26:25
difficult thing that I haven't been able
26:27
to find that is like
26:28
mind blowing until now.
26:30
Hmm.
26:31
And so you're saying you
26:32
have like a winning ad.
26:33
Like what would be your
26:34
process in terms of iteration there?
26:36
I mean, we have it from the Figma board.
26:40
We built a plug in from Figma to Kai.ai.
26:44
And so we can say go make 10 iterations
26:46
and we can select change product
26:49
placement or change the background.
26:50
And so we can do either of those and we
26:52
just select how many numbers we want.
26:54
Very easy for us is we do nine by 16 with
26:56
a four by five safe zone.
26:58
That's kind of like the
26:58
limits to stay within this.
27:00
We just go to Figma, select eight, nine,
27:02
ten, whatever we want.
27:03
Press run.
27:04
It comes back, gives
27:05
us exactly what we want.
27:06
No prompts.
27:06
No, no.
27:07
And so it'll just
27:08
kind of run through that.
27:09
And then as we spoke about again, well,
27:10
it's like then that kind of duplicates
27:12
and we do the exact same
27:14
thing with the headline iteration.
27:15
So the H1s up top and then
27:17
we just say, hey, 10 new H1s.
27:19
It runs back to Claude, runs through the
27:21
Claude project for that company, that
27:24
brand, runs to the Claude project.
27:26
Claude project is connected to Notion,
27:28
runs to the Notion
27:28
dashboard all the way back, right?
27:30
And then it just
27:31
generates the H1 copy across.
27:32
That's crazy.
27:33
So the H1 copy is unique, written by
27:36
Claude every time is what you're saying
27:38
for every iteration.
27:39
Every single time.
27:40
Yeah.
27:40
I mean, you could do, I think we have it
27:41
up to like 25 right now, but it probably
27:45
takes five seconds for it to do it.
27:47
It's so quick.
27:48
Yeah.
27:48
Okay.
27:49
So I want to sort of drill down on one
27:51
thing you mentioned.
27:51
One of the issues with these models, I
27:54
think we can all sort of agree like Nano
27:56
Banana 2 is probably the best model for
27:58
static generation, is like
27:59
getting your text to look right.
28:00
My process is like building, you know,
28:03
building spec cards and having it
28:04
reference the spec cards, but still it's
28:07
maybe like 80, 85% success rate.
28:09
You mentioned something the other day
28:10
that I thought was fascinating.
28:12
It's like in this Figma thing, you are
28:14
manually creating the
28:16
initial one in Figma, right?
28:18
With the actual copy,
28:20
with the actual font file.
28:21
And then through your iteration process,
28:23
you're changing that through Claude.
28:25
So, you know, maybe like at a fifth grade
28:27
level, can you sort of like explain
28:28
exactly what's going on there?
28:29
Yeah.
28:30
So we'll go generate, let's
28:31
say you have your image, right?
28:32
You're like, hey,
28:33
this is the image I want.
28:34
That lives in Figma.
28:35
The designer comes in is like, hey, this
28:38
is where I want the H1 copy.
28:39
I think not only is like the font hard to
28:42
find, but it's also like the placement
28:43
when you kind of use AI of like getting
28:46
that placement perfect.
28:47
It really looks aesthetically pleasing is
28:49
really difficult, I feel like.
28:51
So basically all we're doing is we built
28:53
a plugin from Claude code
28:55
to run back to Claude, right?
28:57
And so all they're doing is they're
28:58
basically designer comes
28:59
in, makes the first iteration.
29:01
Like that's basically the only thing they
29:02
have to do is come in,
29:03
make that first H1 copy.
29:05
And then we just hit the plugin.
29:07
It runs back to Claude.
29:08
And then it just duplicates the image
29:10
across and just puts the H1 copy in the
29:12
exact same place with the new H1.
29:14
Got it.
29:14
I think, yeah, I think that like what
29:16
you're saying is like so valuable because
29:18
the amount of DMs and questions I get
29:20
about it's like my font's
29:21
not rendering correctly.
29:22
Like this is like the easiest solve that
29:25
people just like don't realize.
29:26
It takes like one extra second of effort
29:27
in Figma to just like actually write out
29:30
the copy and then like
29:31
iterating becomes so much easier.
29:33
Yeah, 100%.
29:34
The thing why we started doing this, one
29:37
is I believe it was, I don't know if it
29:39
was the CMO at Claude or someone, but
29:41
there was someone at Claude who spoke
29:43
about doing this with
29:44
Figma in a different capacity.
29:46
And I was like, I watched that video and
29:48
I was like, we're just going
29:49
to change it and do the H1 copy.
29:51
Because the biggest thing that pissed me
29:52
off is we'd go generate an image and the
29:55
image would look great and the font
29:56
screwed up the whole image
29:57
and start from scratch again.
29:59
Right.
29:59
And I was like, I don't
30:00
want to keep doing this.
30:01
Right.
30:01
The image is right or the font's wrong or
30:03
the font's right and the image is wrong.
30:05
I'm going to do them separately.
30:07
And so the first thing I did was go make
30:09
the images and then I'd come back and
30:10
upload that image and
30:11
try to get the font right.
30:13
And then it still wouldn't be right.
30:14
And so I was like, okay.
30:15
And a lot of it was
30:16
placement of where I wanted it.
30:18
Like even just trying to slide it over,
30:20
like a quarter of a
30:21
centimeter on 9x16 matters, right?
30:24
Like it's a huge difference.
30:26
And so just having them place it and then
30:28
you just name it H1 and then Claude just
30:31
knows that this is the H1 text box and it
30:33
just runs it across.
30:34
Great.
30:35
Very cool.
30:35
I mean, we would sort of be remiss not to
30:37
mention that Claude
30:38
design came out yesterday.
30:41
Have you played with it at all?
30:42
I mean, I blocked off some time tomorrow
30:44
to spend on it, but it seems like, you
30:46
know, Figma stock is down.
30:48
People are, you know,
30:49
talking about this a lot.
30:51
I don't know if
30:51
you've gotten in there yet.
30:53
I haven't touched it.
30:54
I know it's out.
30:56
I mean, we had Opus come out, right?
30:57
Everything like this.
30:58
But I have not touched
30:59
either to that capacity.
31:01
I've been in meetings to meetings the
31:02
last two days, but that is this weekend.
31:05
The fiancee will be pissed when I sit on
31:06
the couch all weekend
31:07
playing with Claude design.
31:09
Yeah.
31:10
What used to be a Friday night of just
31:12
like drinking beer and watching sports is
31:14
now just like drinking NA beers and
31:16
playing with Claude code.
31:17
Exactly.
31:19
It's a life.
31:19
It's a life I dream.
31:20
Sounds awesome.
31:21
Yeah, it sounds awesome.
31:22
So, you know, what do you think, Tori,
31:26
just as a way of, you know, getting
31:29
people ahead that the way that you think
31:32
about AI and how you're, you know,
31:34
integrating in what you're
31:35
doing, setting aside time.
31:36
And where do you think you recommend
31:39
media buyers and agency owners to spend
31:41
their time moving forward with AI?
31:45
Like, is it around the
31:46
creative optimization?
31:47
Is it around thinking
31:48
about the whole funnel?
31:49
Is it both of those things?
31:50
Like, how do you sort of allocate your
31:51
time in terms of really
31:53
seeing client results?
31:54
Because I think a lot of us struggle with
31:55
like, all this stuff's cool.
31:57
I don't know what to do.
31:58
And so it's really a matter of where can
32:01
you focus that time?
32:02
Yeah, I mean, I think the biggest thing
32:04
for me, and like,
32:06
obviously, like, I'm a founder.
32:08
And so like, I have a team is I wanted to
32:10
provide more value to two
32:11
people, my team and my clients.
32:13
That was like the only
32:14
thing I really cared about.
32:15
Like my time, listen, I'm going to work
32:17
100 hours a week anyways,
32:18
because I'm a psychopath.
32:19
So I'm going to figure out a
32:20
way to like work 100 hours.
32:21
It doesn't matter.
32:22
I wanted to let them get more done within
32:24
their work week or provide
32:25
more value on the other side.
32:27
The biggest thing I would tell people and
32:28
how we kind of got started and how I tell
32:30
my team to get started
32:30
is like whisper flow.
32:32
If you know what that is, WISPR, you
32:34
know, you just hold a button on your
32:35
keyboard and you talk to it.
32:37
It's a very easy way to talk to an LLM
32:39
and actually like get good results.
32:40
And you can sit there and kind of process
32:42
and think, double tap the key and sit
32:44
there for 10 minutes, literally 10
32:46
minutes, I would say.
32:47
Like, don't go any shorter and just go
32:49
describe everything in your business and
32:51
all the problems you have.
32:53
Like, hey, you know, we're
32:54
not the best at reporting.
32:55
We need to fix this.
32:56
Our spreadsheets are manual.
32:58
This is how I want to do this, etc.
33:00
And let Claude, Chow GPT, Perplexity,
33:03
whatever it is, prompt you back and tell
33:05
you, hey, these are
33:06
the things you can do.
33:07
Everybody's business is so different.
33:09
Some people are really
33:09
strong on the creative side.
33:10
Some people are really
33:11
strong on the analytic side.
33:12
So I wouldn't say like, hey, there's a
33:14
one size fits all year.
33:15
I think just go talk to it.
33:17
Describe where your issues are and let's
33:19
see what it spits out so you can start to
33:20
focus on what's going to move your
33:23
business forward the most, not what I'm
33:25
going to recommend because everybody's a
33:27
little different here.
33:28
I think that's a good call, right?
33:30
In terms of sitting down and just talking
33:31
to it first and starting there.
33:34
That's something that it's easy to forget
33:37
when you're getting a lot of information
33:39
pushed to you about like, these are the
33:40
10 things I solved, etc.
33:42
Well, Tori, appreciate
33:44
you hopping on with us.
33:45
Appreciate you sharing some just
33:46
absolutely golden nuggets.
33:49
If people have questions or want to
33:51
contact Tori, you can find him on X or
33:53
you can email me, andrew at
33:54
foxwelldigital.com and
33:55
I'm happy to connect to you.
33:57
And Tori will share some resources in the
33:59
notes here, but I
34:00
appreciate you hopping on.
34:01
Awesome.
34:02
Thank you, guys.
34:02
Appreciate it.