Latest / 1 to 100 / E33: How to get AI to recommend your brand with FERMAT

Transcript

E33: How to get AI to recommend your brand with FERMAT · 1 to 100

0:00
The product released yesterday was like super fun, honestly.
0:08
So you you guys probably know we don't we we up until now had no
0:12
free like components of our product, right.
0:16
So the lowest price thing that we had was like our funnel
0:21
builder product for 500 a month. We had that for a little while,
0:27
but like the more I thought about what is the end state
0:32
outcome of what we're trying to accomplish.
0:35
And then I saw this AI search thing and how candidly how easy
0:40
it is to build. And like what we are good at,
0:44
which is post the understanding of how you're showing up in AI,
0:50
what do you actually do about it?
0:52
And then the generation of those stores is actually the hard
0:54
part, unless you've already done all the work to build in all of
0:57
those integrations and have the workflows that automate the
1:01
building of all of those stores. And so that's what led to the
1:06
idea of just saying, you know what, why don't we just give
1:08
away the monitoring and a tier of how much content you can
1:14
generate, after which point you'll start to see impact.
1:17
Because actually our alpha customers saw not only
1:21
impressions on that from people being redirected from ChatGPT
1:25
and overviews and things like, so like these things are getting
1:27
sighted, but they actually saw conversions too.
1:32
And once you start to see that, you can sort of see the future
1:36
and then you just like want to do more.
1:39
And so I was like, OK, the job here is to just get people
1:44
started. And then as they want to do more
1:47
than of course, there's like a fair value exchange at that
1:50
point. But the yeah, that was the idea
1:54
behind doing the the free launch was, was just to like put a
2:00
price, put put the price that we thought was right for the actual
2:04
monitoring side, which we think is 0 and then value the thing
2:09
that we actually think is valuable, which is the content
2:11
workflows, so. What what was your timeline from
2:15
the moment of identifying opportunity into like launching
2:18
a product? Three months, so from like from
2:25
conception to like we have a couple dozen alpha partners who
2:33
already saw outcome. So from conception to like first
2:36
customers on was like a few weeks, but then we wanted to see
2:40
how it works and iron out the kings and all that sort of
2:43
stuff. And so then the actual launch
2:45
was sort of three months after conception.
2:48
Yeah, you. Know what's so funny?
2:50
The number one product that we've all talked about the most
2:54
in this podcast that none of us are doing is AI search
2:57
optimization. It's come up so many times every
2:59
time you must have heard him. And like, I know I'm building
3:02
it. Well, I, I just think that a lot
3:08
of people are building it like I, I just think that the
3:12
monitoring part is like, like, I, I mean it sincerely.
3:17
I think it's, I think it's a commodity and I think that the
3:20
workflow afterward, like what do you do once you know?
3:25
I think that a lot of people will have different answers to
3:28
and we have our version of an answer to that question, which
3:31
is like shoppable content for the LM to interact with.
3:34
So that way you can direct consumers there.
3:38
And we have evidence that that works.
3:40
And so we think our version of the answer is a good answer.
3:43
I'm sure there are other good answers.
3:46
And I think just, I think the thing that I didn't fully
3:49
appreciate and someone said this on a different podcast is there
3:52
are spaces that are going to become so big that actually you
3:58
just want to participate in that space.
4:00
And if you can become one of the bigger players in that space and
4:03
that's great. And for AI search, there's just
4:06
like no great vertically curated option for brands.
4:13
And so I, I think that we are going to be the great option for
4:16
brands. And then I think there's going
4:18
to be a great option for SAS and I think there's going to be a
4:20
great option for other, you know what I mean, for services.
4:24
And then I just think there's going to be a great option for
4:27
every vertical eventually. Can you talk about the, the
4:32
product a little bit in terms of like what it, what you're
4:34
actually doing from a monitoring side and, and how that's what,
4:38
what you're tracking and how that plays out in terms of act
4:41
on top of that? Because I think like a little
4:44
bit more detail for anybody who's listening would be really
4:45
interesting. Yeah, yeah.
4:48
So basically the, the, the like advantage Fermat has is we have
4:57
a ton of data about how people shop because for our customers,
5:02
we are connected to their, all of their ad and marketing
5:06
platforms. So like Meta, Clavio, whatever,
5:08
whatever and to the back end. So we see all the order data and
5:13
when they are interacting on the shop, our shop or the brand
5:17
shop, we have both structured and unstructured data about how
5:20
they're behaving on the store. So we understand like what
5:24
patterns cause shopping. So we take all of that and we
5:29
take third party data sources, so Reddit, social, whatever,
5:32
whatever. And when you plug in as a new
5:35
brand on AI search, we use that information to understand what
5:40
queries to monitor. So that's like the first most
5:44
important difference and why I think verticalizing will end up
5:47
mattering. So we use that data to identify
5:51
like, hey, what are the like hundreds of queries you should
5:54
be thinking about given your contacts, given who else shows
5:57
up, your competitors, all that sort of stuff, how people buy
6:00
from you. And then because we're connected
6:04
to all of that content and we have as inputs like your brand
6:07
voice and all that other stuff from your website, we can just
6:10
automatic like literally against the queries.
6:13
We just automatically generate all of the shoppable content and
6:16
what types of products should be curated into that content.
6:20
And so you can like hit one approve button and just get all
6:23
of this published and it'll get start getting picked up by the
6:26
LMS when people ask the LMS questions of that variety.
6:31
And so that's like what the product does.
6:33
And I think that every vertical will have that so.
6:37
All of that's free, or just free up to a certain number of
6:40
queries. Up to 20 pieces of produced
6:44
content, all of the monitoring, up to like 100 queries and up to
6:48
20 pieces of produced content. So that when you start to like
6:51
actually see the change and then if you want to keep doing more
6:55
of that, great, then you pay us for to do more.
6:59
Yeah. Yeah, that's awesome.
7:00
Do you do you think it's going to be like a arms race, for lack
7:05
of a better term, just like lots of content being generated
7:07
automatically that's like kind of useless outside the context
7:11
of LLMS using it for surfacing information.
7:15
Like is that, is that what's going to be happening in the
7:18
future? I mean, SEO kind of did that.
7:20
Like, yeah, I mean, with with SEO, that's what I mean.
7:24
People have been producing lots of kind of like effectively
7:26
useless content just to rank and keywords like they're not
7:30
actually adding value. And so it sounds like what
7:33
you're describing is that the same thing is going to happen
7:36
with LLMS, but at a much bigger scale.
7:39
OK. I think that there's lots of
7:41
ways to interact with the like content for content sake
7:46
generation point of view. I think that the reality is that
7:52
LLMS are able to suck in way more information than any other
7:57
system previously created. And so when that's true, what
8:03
you want to do is not produce content that covers the gamut,
8:07
including things that are false, because the LLM will also
8:10
identify what's bullshit, which is the whole thing about them
8:14
being able to access both third party and 1st party information,
8:17
right? So what I think is the most
8:19
likely is it will look like this, meaning there's a certain
8:24
amount of content that you should be producing, which is
8:27
accurately and usefully describing what your product
8:30
does. And this is like sort of the
8:33
engineer's dream or the product person's dream, which is like,
8:37
oh, but but our product also does acts and also, and the
8:39
marketers like, dude, shut the fuck up.
8:42
Like nobody's going to sit there and listen to all of the shit
8:44
that the product can do. Just keep it simple.
8:46
So that way someone can buy the thing that they're trying to do
8:48
right now. But the, and then the engineer's
8:50
like, but also this, but also like the, but also this is what
8:54
you now produce for LLMS. And so that's what the whole idea is
8:58
behind the content generation is you want to give unique, useful
9:04
content. But if you start bullshitting,
9:07
the album is going to know that because it has access to other
9:11
information sources, and it'll say like, that's weird.
9:14
Only you're saying that it does that and you're the maker of it.
9:18
Yeah, but I I feel like it's history is like funny.
9:22
It's just like repeating itself like this is exactly like SEO
9:25
and like the game changes over time and it's just like it's at
9:29
at every stage there's like a new thing to game.
9:33
At one point it was just like content and then it was internal
9:36
links and then it was backlinks and then it was demand
9:39
authority. And like The funny thing with
9:43
all of this stuff is that it is like always a moving target and
9:48
there's always more people, I feel like looking for arbitrage
9:51
and people looking to like push arbitrage away.
9:57
And so I think it's going to be the exact same as as SEO.
10:01
Like I were to say, if you were to like interview any top SEO
10:05
authority in you're to say, OK, like our objective is outcomes.
10:09
Like the last thing that they would say is like we need to
10:13
provide value. They'd be like, well, these are
10:15
the things that we need to do because this is how the game has
10:18
been played and the level of value that we create is a side
10:24
effect of the way the game is positioned at any given time.
10:29
And so it's like it's an interesting product.
10:30
I mean, I, I, I'm like super bullish on it.
10:31
Like I spent like a lot of money this week specifically just
10:34
through LMS 'cause I'm building like a home gym.
10:37
And so like I was like finding all these like random things,
10:39
comparing them and almost every single purchase that I made like
10:43
came from like a link from within ChatGPT.
10:45
So I think, I think it's like a sick product direction to to go
10:49
into, but it's like, it is one of those things where it's like,
10:53
yeah, you got to iterate it like super, super fast as like the
10:56
the models are updating. And it's, it's almost like
10:59
there's even more complexity to it, right?
11:00
Because there's like so many other sort of like models that
11:03
someone could be using or LLMS or chat interfaces that someone
11:06
could be using. Whereas before it was like, OK,
11:08
if you just did Google your, that is your 8020.
11:11
And only if you got super big do you care about like Bing and and
11:14
stuff like that. But.
11:16
That's so. I mean, I think it's just really
11:19
how much do you think of the content is going to be
11:24
decentralized as opposed to coming from the brand
11:28
themselves? Or if you think about SEO, for
11:30
example, I'm sure you guys have all heard about that ginger guy
11:33
from Ireland. He was like an SEO specialist or
11:34
something and he managed to get his face to come up as the
11:37
world's most handsome man, right?
11:40
So if you search world's most handsome man, back in the day,
11:42
it was like Brad Pitt, George Clooney ran an Irish dude.
11:46
You can't do that now, right? That hunting ARB has been shut
11:50
down, not by the intention to. I mean, I guess to some extent,
11:54
Google and ChatGPT and all of these platforms are valued on
11:58
their ability to show people what it is that they're looking
12:00
for, right? So if the product's doing its
12:02
job, it doesn't get gained by a non-technical marketer somewhere
12:05
in the world who's just hunting ARB, right?
12:09
And you would assume that they have, you know, the tools to
12:13
shut that down really quickly. So, yeah, I just wonder, you
12:17
know, to what extent this becomes about decentralized
12:20
content, you know, reviews for a restaurant from verified
12:24
reviewers, people were actually visited.
12:26
That location on Google Maps matters so much more than what
12:29
some random person writing for the New York Times says, right?
12:33
Because if your restaurant has, I don't know, 5005 star reviews,
12:36
right, that signal to Google that this is the right
12:38
restaurant. Whereas like what some random
12:41
dude thinks is is not so much. Yeah, totally.
12:43
And, and I think, I think we're saying the same thing, which is
12:47
like the LLM builders like ChatGPT, perplexity and Tropic,
12:55
they all will, they'll directly tell you like, hey, if you guys
13:00
are going to try to game it like we're going to like our
13:03
objective is to get our LLM to be the most useful and the most
13:08
accurate that it can be for the person interacting with it.
13:11
So then the gaming of that is like, OK, just produce the most,
13:17
just produce every version of the content that this thing can
13:19
synthesize. That is like accurately
13:21
describing what the product is because it's going to identify
13:26
anything that doesn't look that is not that.
13:30
And so that is actually the equivalent of like making sure
13:33
all the reviews are visible and making sure those reviews are
13:35
accessible and making sure those reviews are from as like
13:39
identifiably real people as possible, syndicating it to 3rd
13:43
party marketplaces or having it be in third party places like
13:47
Google Map. Like, I guess what I'm saying is
13:50
like in the SEO world, there was like white hat, Gray hat and
13:54
Black Hat, right? Like nominally that's the
13:56
spectrum. And I think in the LM world, I
14:02
think Gray and black will get quashed much faster.
14:05
And so the techniques you want to deploy look more like the
14:07
white hat techniques from the SEO world.
14:10
That's, that's basically all I'm trying to suggest.
14:11
I'm not trying to suggest that there's not like gaming, I just
14:14
think the game is more white hat than Gray and black.
14:19
I actually think too when it comes to specific products, I
14:24
think like Open AI for example, has incentive to pull in as much
14:27
data as they can right now. Because like when I actually
14:30
think about the product recommendations I get through
14:32
ChatGPT, I don't think they're very good.
14:34
I think category recommendations are quite good.
14:37
So in terms of like, like I just had this the other day.
14:40
My mother-in-law staying with us.
14:41
She's in the basement, the basement's cold.
14:43
I was looking at good heater options and ultimately like I
14:46
got the type of heater that I wanted out of ChatGPT, but the
14:51
right, the specific products they recommended weren't very
14:53
good. I found better options going and
14:55
searching for that on Amazon do. You want to know why this is
14:57
fucking crazy but I can't believe this is true.
15:00
You can ask ChatGPT how are you recommending me products?
15:03
It'll tell you. Oh, interesting, which I thought
15:05
before. It turns out, what do you think
15:08
are the top 3 criteria for ChatGPT to recommend a product
15:11
to you? Think one of them, and I, I
15:14
think this relates back to what you were saying about the amount
15:17
of leverage and the amount of data that you have on the
15:19
customer. I think one of them is whether
15:21
it's just the product is defined by that one use case only.
15:26
So if it's spread across different use cases, it dilutes
15:30
its ability to be fetched by ChatGPT, but I'm not sure.
15:33
That's I bet. I bet price is really high.
15:37
That's the issue, the price. Yeah, this is exactly the
15:40
problem. Prices like #1 by a long shot.
15:43
And and because price is number, price is number 1 and then
15:48
convenience and then variety. OK.
15:51
So like it it, it roughly tells you exactly what it's going to
15:55
do. And the and it overweights price
15:56
by massive amount. And so you have to specifically
16:00
prompt it with I am not optimizing for price.
16:05
Yeah, in order to get the best option.
16:07
And here's what I am optimizing for.
16:09
And then it will actually give you good recommendation, which
16:13
is crazy. My girlfriend's putting together
16:16
her registry right now. She says she's using Gemini
16:19
because every time she asks Cha Chi PT for the Google reviews,
16:23
it says I can't get them. So she's using Gemini because
16:26
Gemini can actually get Google reviews and give her a better
16:29
recommendation. So I'm.
16:31
Yeah, I think that's interesting.
16:32
How? How does ChatGPT not have access
16:34
to Google reviews? Yeah, I, I like, I really don't
16:40
like ChatGPT anymore. I think I mentioned this in the
16:43
last episode. It's seemingly shit at
16:47
everything. Like, yeah, I don't know what
16:50
the fuck's going on, but like, there's been snow many times
16:54
when I'm like, you've got to be fucking kidding me, I swear.
16:56
You could have done this three months ago or like, yeah, it's
17:00
like the opposite feeling of using ChatGPT before.
17:05
And so I've almost been like exclusively Claude and Grok,
17:11
like almost 100% Grok is incredible.
17:16
And like, I, I personally believe from first principles
17:21
over like a long time horizon, it wins because leadership is
17:27
like so focused on truth, which like, if you were to like watch
17:32
a bunch of interviews with Sam versus a bunch of interviews
17:35
with Elon, OK, the number one thing someone needs wants from
17:39
an LLM is truth. And like Elon will mention truth
17:43
about 100 times like ad nauseam as like the only thing that
17:47
matters. And Sam Altman seemingly never
17:50
like mentions like the word truth.
17:52
And so like I think from first principles when like the only
17:55
thing that matters the truth in LMS like I, I feel like my bet
18:00
right now is like on Grok or something with like similar
18:03
leadership and direction. I.
18:04
Was going to say. I was literally going to say
18:06
that the single reason why grok doesn't succeed is because
18:10
they've made it so hard to use, which is insane.
18:16
I can't believe that's the reason why it won't win.
18:20
You know what I mean anyway, but that's that's.
18:24
Something they can change eventually, right?
18:26
Like let's say 2 years from now. They're, it's very clear that
18:29
they, they have the thing that everybody wants to use more, but
18:32
it's hard to use. I think you can.
18:35
You can change that behavior. And in the meantime, they're
18:37
still making it better and better.
18:40
Yeah, I, I, I suspect it's because their goal is not to
18:43
have it actually be the most used for what it's worth.
18:47
And I actually think that that's where the magic is, is that Elon
18:52
is good at telling people what the only thing that matters is.
18:56
And other leaders, I mean, like, we're all guilty of this shit.
19:00
Like it's so hard to describe to people.
19:03
Like here is the main thing and keep the main dislike.
19:07
All the Peter Thiel's sort of like friends and family, as it
19:10
were. They all are very good at this,
19:13
which is just like only focus on one thing and like literally do
19:17
not pay attention to anything else.
19:20
And so he's like focus on truth, focus on truth, focus.
19:22
If he said focus on truth and distribution, guess what
19:25
happens? You don't focus on truth.
19:29
So my my hunch is that it's hard to use because they don't give a
19:32
shit about distribution right now.
19:34
That's my hunch. It also is like, I do think that
19:38
our interaction with LLMS are probably not going to be in, in
19:42
volume is probably not going to be as much on the screen as they
19:44
are like right now. Like even like I interact most
19:48
of trachy, but when I still use trachy BT, I use it for voice on
19:51
my phone, like when I'm getting ready in the morning and things.
19:54
And I think if you think about, OK, what are the opportunities
19:58
that Elon has with Crock? It's like 1.
20:00
You can put it in the robots that are going to be in our
20:03
homes. And then the other one is like,
20:05
it's in Tesla and so it's in your car.
20:08
And so if like their distribution actually eventually
20:11
could just be through like hardware and so.
20:17
I think the only like big sort of thing that they haven't
20:21
figured out is like, yeah, like the phone sort of like
20:23
distribution, like piece. Like the idea that Siri is still
20:26
the Siri is so fucking insane. I wonder where we get to AI that
20:30
isn't majority chat based, right?
20:34
I'm not sure if any of you guys have used Cursor, but one of the
20:37
nice things about Cursor and how it's evolved over time is you
20:39
can just hook it up to everything and then it can kind
20:41
of just keep going and doing the job, which is really cool.
20:45
Like I really like it because it was, you know, it's, it's broken
20:49
down a lot of the refraction, a lot of the requirement for me in
20:54
the process. And yeah, I want to see more
20:56
products like that. I want to see more hands the AI.
21:00
Majority of AI products will not be textual interfaces that
21:04
they're just not going to be. They're either going to be voice
21:06
or they're just going to be actually like even our AI search
21:10
product is literally 100% agentic.
21:14
Like it is going and retrieving information and it's creating
21:17
the content. It's, it's like it's doing the
21:20
whole thing and it's like, hey, boss, do you approve And you hit
21:24
an approve button and then and then it's done.
21:27
The whole thing is UI oriented or like visually oriented.
21:34
The outputs are visually oriented.
21:37
That's what things are going to look like.
21:39
It is going to look like you assign a task to someone and
21:41
they just like go and do it. And then they come back to you
21:44
with a finished product and say, what do you think?
21:47
Does this look good? Like literally our UI is a hot
21:50
or not UI or like a Tinder UI. It's like literally swipe left,
21:54
swipe right. I'm not joking.
21:55
It's not even textual. Yeah.
21:57
I will say that's my like biggest gripe right now with
22:00
using sort of LMS and like day-to-day work like one now
22:03
pretty much only use Bisper flow for like dictating prompts,
22:08
which like now, honestly the only AI product outside of like
22:11
the foundation online. So like I couldn't live without
22:13
now is definitely whisper flow. It's the only one.
22:17
And but like I've noticed that the cadence of community like
22:21
the beautiful thing with whisper flow is that it unlocked it.
22:24
It unlocked the barrier of like my, my brain speed towards my
22:28
typing speed, but the but the return isn't coming at the same
22:35
velocity cuz I'm like handcuffed by like my ability to read,
22:38
which is probably subpar. But then like the dictation is
22:43
almost like 2 verbose. And so like there's a part of me
22:46
that's just like I'm I'm potentially craving like neural
22:50
link. Like like it feels disorienting
22:56
to like have this like difference in cadence of like
22:59
communication. And I almost like feel like the
23:02
moment that it can just be as fast as like thinking and then
23:09
receiving is like when there's going to be like incredible
23:12
productivity gains because like right now it does get like
23:15
tiring. Can you share with me when you
23:18
use whisper flow, are you sitting or are you walking?
23:21
I'm mostly sitting. I I I'm actually like a only
23:25
work at my computer type guy, which I think we've talked
23:27
about. Yeah, but when I'm walking, I'm
23:32
very much just walking like I, I don't like look at anything or.
23:36
Yeah. Yeah, imagine that.
23:37
Hey, imagine if you go directly from thought, not just all the
23:42
way through to execution, right? You're thinking and things get
23:46
executed. So I know you're on a plane and
23:48
you have an idea and it gets run the whole way through before you
23:51
can even talk to anyone else. Well, that's for sure going to
23:54
happen between like Starling and Neural link and.
23:58
I mean, you with knurling the guy.
23:59
I was just on a walk. I was just on a walk and like I
24:03
had a bunch of ideas on that walk and like I, I voice noted a
24:07
couple of them just to myself or like transcribe them to myself,
24:10
but I'm not working. I'm just like, OK, write this
24:13
down. But like, the idea that I don't
24:18
have like a fathom summary of my walk seems insane because it
24:24
would be like super valuable if I could just like be like, oh
24:29
like what was that fucking idea that I just had?
24:33
But The thing is, the only reason you need a fathom summary
24:36
is because you have to use that information again to go execute.
24:40
So imagine no need for a summary, no need to dig up
24:44
information because the information is being used in the
24:47
most productive way as soon as you've thought about it, right?
24:52
I mean this doesn't exist and it might be a while to exists.
24:56
There's a tool that we use that takes in, it plugs into wherever
25:01
you get your customer feedback. OK, I, I don't know what systems
25:04
you guys use for us, that's like Slack, literally.
25:08
OK, so it's in, it sits in Slack, it's looking at feedback
25:11
from customers, it's gathering it.
25:14
It's then creating tickets out of it.
25:18
From those tickets, it then actually writes the code to
25:22
solve that customer problem. And then it pushes the PR and
25:27
then an engineer can review it. But the whole thing so like you
25:33
have not instructed. So it's like very, this feature
25:37
is like basically already here. It's just unevenly distributed
25:40
as it were. But like a customer is filing a
25:43
ticket and then there's a pull request sitting waiting to get
25:46
reviewed and approved is happening without anyone talking
25:51
to anyone or doing anything. You, you guys should try this
25:55
product. If you're interested, I'm happy
25:56
to share it with you guys afterward.
25:58
It's called Edge. They they, they literally they,
26:00
they won't be able to onboard multiple people if too many
26:03
people sign up. Because it's like, yeah, it's a
26:06
small startup. And so the only reason I wasn't
26:10
going to share it is they literally wouldn't be able to
26:11
handle the inbound. Not that I'm saying we would
26:16
generate a ton of inbound, but yeah.