Latest / eCommerce Podcast / How to Actually Use AI in Your eCommerce Business

Transcript

How to Actually Use AI in Your eCommerce Business · eCommerce Podcast

0:04
Well, hello and welcome to the eCommerce Podcast. My name is Matt Edmundson.
0:08
It's great to be with you from our brand new studio. We've only recorded a few
0:13
podcasts here before, so I'm still trying to get used to all the tech and how it
0:17
all works. Very, very different to the old one, let me tell you. Very, very
0:21
different. A lot, definitely a lot more going on. But welcome, welcome to the
0:26
show. We talk about all things ecommerce. Uh, as the sort of show describes in
0:30
the title. Um, if you're new here, a very warm welcome to you. If you don't know
0:35
me, uh, I've been doing ecommerce since 2002. Uh, and so like I love to say on
0:39
the show, if you're a regular, you'll get bored of this now, but I'm a bit of an
0:44
ecommerce dinosaur. Um, I've been around since the early days, uh, and I just
0:48
love it. I've got my own ecommerce businesses just like you. And so I love to
0:53
get great guests on the show, whether that's an expert like today, and I get to
0:57
grill them on how I can make my own ecommerce business better. Or whether we get
1:02
someone like yourself, maybe a founder on the show who just shares their story,
1:06
and we just love to hear people's stories. So wherever you are, wherever you're
1:10
watching it from, you're very welcome here. Thank you so much for joining. Of
1:15
course, if you are a regular, welcome back. I do say this regularly, actually.
1:19
If it wasn't for the regulars, I wouldn't be doing this show anymore. And it's
1:24
because you do tune in and you keep, you keep subscribing. I just love that.
1:28
Keeps us doing the show still. We've been doing it since 2019. Years. Man,
1:33
that's a long time. And just loved every single episode we've done. We've— I
1:37
genuinely go away with notes from every episode on how we can do this better.
1:42
And of course, now we've got AI, so you can take the transcript from the
1:46
podcast. If you're like me anyway, you take the transcript from the podcast and
1:50
go, how does this apply to my business, Claude? And then you have a really
1:55
fascinating conversation. Uh, and it, it's, it can be eye-opening. Um, but
1:59
however you do it, however you consume the content, welcome, welcome back. Uh,
2:04
it's great to have you here. So without further ado, let's talk to— oh, by the
2:08
way, what I've not done is I've not said go to the website. The call to action,
2:13
Matthew, you know, the lifeblood of every ecommerce website. Uh, our call to
2:17
action is just go to the website, ecommercepodcast.net. You can find our entire
2:22
archive of episodes there. You can find out more information about Cohort, which
2:26
are our monthly sort of meetups. Um, they're all done online, by the way, and
2:30
they're all free, but you get to meet e-commerces from around the world. Um,
2:35
we've got on there stuff about the AI systems. We've got stuff on there about
2:39
workshops, I think. Um, anyway, everything's on the website,
2:44
ecommercepodcast.net. So go check that out there. Cyril, welcome to the show,
2:48
man. Great to have you. How you doing?
2:53
Hi, Matt. Thank you so much for inviting me. And, uh, as you said, I can repeat
2:59
that, uh, I have been building ecommerce software since 2003.
3:05
Ah.
3:06
One year ahead of you.
3:08
Yeah, yeah, yeah. So similar ecommerce dinosaur age, Cyril. That's what's going
3:11
on.
3:13
That's it. That's it. And moreover, when you ask me what has been around and
3:21
what changed since 2003, I can answer not too much.
3:29
Yeah, it's all fairly similar. We've still got a keyboard, a mouse, and a
3:34
screen. Nothing much has changed.
3:39
Yeah, sure. The only, the only thing I can recap right now, just what is the
3:46
answer for the question, what should I do if I want to start selling online?
3:53
Yeah.
3:54
And you see that, uh, for 20 years ago, let's say 20, 25 years ago, it was okay,
3:59
I should rent a dedicated server and install a CMS platform and then tune it up
4:04
and start selling. 10 years ago, it was kind of okay, I should register at cloud
4:09
solution and start selling. But what's now? What's today? What's your guess
4:14
here?
4:20
Same thing, just go to Shopify, get a CMS set up, just call it Shopify and start
4:24
selling eBay. I mean, it's a little bit more complicated than that, obviously.
4:27
How would you answer it?
4:31
I'm quite sure that this answer is not true anymore. In the very short time,
4:36
probably weeks, probably months, but not more than 1 year, you will start with
4:41
that, or Claude Code, or Claude Code.
4:45
Yeah.
4:46
Let me start selling online and that changed everything.
4:50
Yeah, it does. It does. And it's fair to say, isn't it? That's your area of
4:55
expertise here. So for those that don't know you, for those that might not have
4:59
heard of you, just give us a quick elevator pitch into all things your world.
5:03
A quick elevator pitch, you mean?
5:07
Yeah. Yeah.
5:07
Okay, good. Okay, good. We are going to make the answer to this question, what
5:13
should I need to start selling online? As I said, with just one phrase that
5:18
you're typing or saying to your lovely agent, okay agent, I want to start
5:24
selling online, and that's all. Probably it must be much, much easier than
5:29
register to any kind of cloud solution like Shopify or something else. Human
5:35
beings must focus on what they are doing the best, understanding what their
5:40
customer needs, understanding how to fulfil your obligations and your promises
5:45
to your audience, and do some brilliant packaging. That's all. All the rest will
5:51
be done by AI agents, no doubt.
5:56
Yeah, yeah, it's gonna be— I'm really curious, you know, to see where it, where
6:01
it all ends up, because it does feel like a bit of a step change. The UK
6:05
government today announced that they're going to perform, um, and I don't know
6:09
whether it's going to be an early, uh, emergency parliament session, but off the
6:14
wake of the head of Anthropic saying that AI development needs to slow down and
6:18
Elon Musk actually agreeing, which shocked the whole world. So everyone's,
6:22
everyone's sort of taken notice. Hang on a minute. Elon's saying the same thing
6:26
as normal people. Well, hang on, we should probably listen. Uh, and the
6:31
government's sort of going, yeah, we should probably have a debate about, uh, AI
6:35
and what's going on at the moment. It's just a fascinating time, isn't it? A
6:39
fascinating world in which we live. And obviously, but you have seen a lot of
6:44
this. You work with a lot of clients over this space. What's the biggest mistake
6:48
do you think we're all making right now?
6:54
Oh, we are all making right now, you mean? Yeah. First of all, it's quite a
7:00
philosophical question, but I have very practical answer. Frankly speaking, we
7:07
are To currently to rely on the general purpose agents and general purpose LLMs.
7:13
You know that the special word of today, harness. Harness is what you have
7:20
around the pure LLM agent that helps you to do the real job and to achieve the
7:26
real result. Frankly speaking, at least half of the wisdom of AI agents we
7:32
currently see is the proper harness. And, uh, probably also noticed that once
7:39
the new LLM is out of the box and gets introduced to the audience, the previous
7:45
generation of the same developer LLM became quite dumber, for example. Yeah. Let
7:51
me say, I'm a longtime customer of Anthropic's and OpenAI products.
7:58
Yeah.
7:58
And just 2 months ago, I have been using Sonnet Sonnet 4 and then Sonnet 5 for
8:05
quite sophisticated tasks. But yesterday I ran the same task on the same Sonnet
8:11
5 just to crawl the list of attendees for the next conference I'm going to
8:17
attend this week. And you know what? This agent spent much less tokens to
8:24
process my request, but it returned much simpler and not so great answer back to
8:30
me. And I start thinking, what's this? Probably Anthropic is forcing me to use
8:37
their latest frontier model much more.
8:43
Yeah.
8:43
For what reason? To make for this LLM more data to learn from or to force me to
8:49
spend more dollars on each 100,000 tokens? I don't know the right answer.
8:54
Probably you have.
9:02
I don't know if I, I don't know if I've got the right answer, but I, I'm sat
9:07
here and I'm intrigued actually by this concept that the same model of AI, um, 6
9:12
months later is producing worse outcomes. I, I'm intrigued to know if that's
9:17
actually true, Cyril. And I, it kind of reminds me of the, um, Do you remember a
9:21
few years ago we had this thing with, um, Apple on the iPhones? I don't know if
9:26
it was ever proved or disproved where they, in effect, they made phones
9:31
deliberately slower so that you would upgrade. So the older the phone got, the
9:36
software was done in such a way that it would make the phone slower. So then you
9:41
would be, you know, you would be forced to upgrade. And I, I think Apple denied
9:46
it. I don't know. I don't want to get sued. It's just hypothetical. I'm not
9:51
saying they did. I'm just saying it's an example. But at the same time, I'm kind
9:56
of— that's what I'm thinking about as you're talking. It's like, it would make
10:00
total sense, I suppose. It's not something I've really thought about, but for
10:05
the frontier models like Anthropic, like OpenAI, to slow down older versions to
10:10
make you use the newer versions so that there's more token usage, so that
10:15
there's more revenue. It's a simple strategy. Yeah. I suppose in many ways, I, I
10:20
don't know if it's valid, but I could see why they maybe would do it.
10:25
It looks, at least it looks like, and it's quite an interesting hypothesis here
10:31
to think about and not just think about, but to understand how should we use
10:37
these in our products, in our tools we are rolling out to ecommerce, to online
10:44
sellers, to people who are selling through marketplaces. or standalone stores.
10:50
For example, if you see the map or the chart of AI-native startups that are
10:56
building something for ecommerce, you will be surprised. Most of them, I'm not
11:02
sure about exact numbers, but probably not less than 75%, they are buying, they
11:09
are building the buy-side agents. So something that are more B2C-focused that
11:15
should help buyers to streamline their buying operations, but for what? To buy
11:21
cheaper or to check the best bargain or I don't know what. And surprisingly, on
11:27
the second hand, the field is not overcrowded with the companies. They are
11:33
building the seller side or merchant side actions. Why is that? For me, it's
11:40
kind of surprise because I have been building B2B solutions for decades and
11:46
never built any B2C solutions. And all of us know that it's much easier to chase
11:52
large B2B customers rather than thousands of small B2B users.
11:58
Yeah, that's an interesting one. I don't know if I've come across— and now we
12:03
talk about it, you watch, my Instagram feed will be full of said companies. Who
12:08
will be offering me their services to use AI to buy product? I think it's an
12:13
interesting idea, isn't it? I guess there'll be a race because I think people
12:18
would look at that and go, well, if I want to be like the AI Amazon or the,
12:23
maybe the AI eBay might be another good analogy where I'm not the guy holding
12:28
the stock. I'm just processing the payments. There's money to be made in in that
12:32
somehow, somewhere, if I've got, you know, if I, if I'm the first to capture the
12:37
market in terms of buyers, I think it's an intriguing proposition. I don't know
12:42
anybody that's doing it well, maybe Amazon. Uh, but other than that, I don't
12:47
know anyone that's doing it well. Um, have you, have you come across anyone
12:52
playing well in that space?
12:57
Oh, uh, not right now, probably. Uh, but I'm thinking about the different angle
13:03
for this question, actually. When you are working on the ecommerce solution that
13:09
must make the life of online sellers easier or much more important to increase
13:16
their margin, to increase their profit, should it be created by a major, for
13:22
example, an Amazon which sits on the terabytes of terabytes of data and hundreds
13:28
of thousands of sellers? Should it be created by midsize player like Helium 10
13:35
or something like that? And is there any room for small third-party players that
13:41
have not enough data and not enough audience to engage with? That's extremely
13:47
interesting for me as a small company founder because I've been on different
13:54
sides. I even spent 7 years in early-stage venture capital and you know that The
14:00
main question that early-stage investors are asking to startup founders, what
14:06
if, then put the title, the name, the trademark of the huge company. What if
14:13
Anthropic, for example, or what if Amazon will develop it on their own, directly
14:19
in-house, and then roll out to all of the hundreds of thousands of their
14:25
customers?
14:32
Yeah. It's an interesting one, isn't it? I, I'm with you. I, thinking that
14:37
through is quite fascinating. I can't see it being Amazon. Um, but I think
14:41
Amazon's a closed platform. I don't think they like— I don't think they play
14:46
well. Maybe they will play well with others later in the day. I don't know. I
14:51
would love the answer to be— I don't think it will be, Cyril. I would love the
14:56
answer to be a guy sat in his garage who becomes the next unicorn. And they're
15:01
talking about this idea, aren't they, of now with AI that actually the ability
15:05
to create a unicorn company where it is just you in your garage is actually
15:10
quite a possible scenario right now. I would hate for it to be Shopify. I'm not
15:15
a fan of Shopify. I'm sorry, Shopify. I'm just not a fan. I would hate for it to
15:20
be Amazon. And again, I think Jeff's had enough money, you know, to find Do you?
15:25
You did well, dude. Just let someone else play. So I would hope it's the little
15:29
guy. I would love that story.
15:34
Same for me. And which is much more important, we can see that it will come in
15:41
place and it's coming every decade, every year and every quarter, for example,
15:47
because in most cases, the old good major cannot create something really
15:53
important because in most cases, it means that they must kill their own business
15:59
in its current stage, for example. And I must say that for companies like, I
16:06
don't want to name exact, but let's do that. You mentioned eBay already in our
16:12
conversation today. And I remember those good old good times 15 years ago when
16:18
eBay decided to build the entire vertical ladder of their solution, not just Not
16:25
just as it was, okay, start selling something online. But what if they want to
16:31
fulfil the ask from medium companies or larger companies? And for corporation,
16:37
the obvious answer is start buying another smaller companies. And it was, it was
16:44
in place. eBay have bought CGI Commerce, or sorry, GSI Commerce. I don't
16:50
remember these exact 3 letters. Founded by cousin of Roy Rubin, the black box
16:56
for huge brands that have been going to start selling online. It was, oh, it was
17:03
15 years ago already, probably 16 even. And on that time eBay have bought the
17:09
PayPal company because if you're selling something, facilitating selling online,
17:15
You must have your own in-house solution. And, you know, from the very strange
17:22
moment, it appears that the entire valuation of the eBay corporation on the
17:28
stock market is getting lower than the sum of the parts that incorporated into
17:34
eBay.
17:41
Yeah.
17:41
Yeah, because you cannot focus on just one part of your business and say, okay,
17:47
and all the rest will come I don't know from what way. So it must be the clear
17:53
focus. And second part for our audience, it must be the clear shift in the
17:59
entire environment around you. Today, it's quite obvious and quite clear. Today
18:04
is that the general purpose agents are the point of contact between the
18:10
customer, the business, and their data. That's, uh, quite obvious. So the next
18:16
question is, if you are using cloud solution or if you are selling on Amazon,
18:21
uh, who owns your own data, your information about your buyers? You or probably
18:27
the platform itself?
18:33
This is, uh, this is the age-old problem, isn't it? In, in many ways. Um, I
18:38
mean, just going back to eBay, eBay I think are stuck in an old business model,
18:43
aren't they? And they can't really seem to get themselves out of it. They're
18:49
sort of like the first major player, it feels like, to be like Blockbuster. Do
18:54
you remember Blockbuster and its sort of— eBay feels like the sort of digital
18:59
equivalent. But yeah, going back to the data question, who owns what? And it's
19:04
such a murky field now. Yeah. Um, like when we sell products on Amazon, I don't
19:10
own any of that data. Uh, yeah, I'm— at least I'm not aware of any data I could
19:15
potentially pull out of it. Um, there may be a few reports. I don't— I certainly
19:20
don't own the customer data, but obviously I, I do on my own site. Um, and so
19:25
you've got the— on one hand, the proliferation of these marketplaces, but on the
19:30
other hand First-party data is becoming more and more important, isn't it? And
19:36
so how do we balance that? And how do we balance that in the age of AI where it
19:41
seems to all be changing yet again?
19:46
It seems so. And I'm quite sure that there are a lot of newcomers on this field,
19:52
including myself, including the company we're currently building. Let's see this
19:58
from the following angle. For example, for decades it was quite obvious that if
20:04
you're selling something, you must have a storefront. That's obvious. And, uh,
20:11
you see, if you start thinking about that, it came from the real offline world.
20:17
When you're coming to a brick-and-mortar store, you see the goods, the products
20:23
on the shelves. Actually, today it completely reflected in an idea of the
20:29
storefront. Yeah. But is it necessary actually? Because your traffic, your
20:35
visitors, and your prospective buyers, probably they might don't want to see the
20:41
storefront itself. You probably know these 2 different concepts of female buying
20:47
and male buying. In most cases, when the gentleman is coming to the online
20:53
store, it's for clear purpose to buy something very specific. And the main task
20:59
and the main challenge for online seller is to let this customer to find this in
21:05
your inventory quickly and to make less friction in 2 or 3 clicks, went him to
21:11
checkout. Yeah. But when a kind of female buying visitor is getting there,
21:18
probably without even a clean intention, probably she wants to see what's on the
21:24
shelves? Should I try this? Should I consider buying or not? Or just have a
21:30
look? And you feel that these 2 types of customers, not necessarily men and
21:36
women, but they must be treated differently. Hmm. If you have clear intention,
21:42
please, I want to give you exactly what you need. If you want to understand the
21:48
entire picture, the entire inventory, okay, I'm ready to show up. But today,
21:54
online sellers in most cases have no understanding what it's all about. Probably
22:00
just the channel where this buyer came from or the particular ads that led to
22:06
this visit, but probably not to the final purchase. And that's the problem.
22:12
That's what I'm talking about when I say you're not owning your data. And when
22:19
we are talking about Amazon sales, it's much worse because your buyer is just
22:25
one click away from buying from your competitor. That's a huge problem.
22:31
Yeah, it is. Yeah, it is one of the many problems, isn't it, of Amazon? So how
22:36
do we deal with it?
22:42
Good question. First of all, my own intention and what we are building now is
22:48
the idea, okay, try to get from your platform or from the marketplace you are
22:55
selling on, try to get as many data, as rich as possible access to API. So you
23:02
can connect, as I said, your agent, doesn't matter from what provider, and start
23:09
asking clear question, asking not only for advice, but for actionable insight,
23:15
let's say. For example, What, what I can say here, if you see the new tool on
23:22
the market, in most cases it's not a help for you. It's kind of a new burden.
23:29
You have to learn how it works. You have to learn how to work with these
23:36
outcomes, how to make exactly your profit higher and increase your revenue. So
23:42
the next point Working with AI agent and your data must be much easier to you
23:49
and not creating new questions, but only providing answers. Next, what if we can
23:56
provide the goal? Okay, I want to sell more, or I want to clear shelves, or I
24:02
want to increase just my solid profit, but I'm not going to increase my
24:09
operations out of that. Just put this goal to your agent.
24:16
Yeah.
24:16
And see what could be done with the data, not just with the data, state of the
24:23
platform. But if your agent could put the actionable commands back to this
24:30
platform and see this executed and see enough tracking of the results, because
24:36
just for example, you know that Amazon is telling outside that their AI
24:43
assistant, about 20% of its suggestions are confirmed and accepted by sellers.
24:49
Okay, that's great. But in this case, how they are going to stand out of the
24:56
crowd, stand out of the competition and gain more sales if they are accepting
25:03
exactly the same suggestions that 90, 89% of your competitors are doing?
25:09
Yeah, exactly.
25:10
So why don't you try the independent third-party solution to see if you can make
25:15
this competition better for you than for your rivals?
25:20
Yeah, it's a really interesting point. And I, I'm kind of— there's so much
25:26
there, Cyril. I'm trying to figure out where to start. I guess just rewind,
25:32
listening to you talk about AI and another tool coming out and not making life
25:37
easier for you necessarily. And, you know, we're all entrepreneurs. We've all
25:43
got shiny object syndrome, haven't we? I guess, is there a Is there a benefit to
25:48
creating your own agents versus SaaS, for example, versus hiring in every month
25:54
somebody's version of their agent, if you like? Like, where do you see it for
25:59
founders and operators? Are they kind of, should we all be getting pretty good
26:05
at creating these basic systems ourselves using whatever AI model you want to
26:11
use, whether it's a Frontier one, whether it's sort of self-hosted or whatever
26:16
it is, should we be creating things that connecting with various APIs? For
26:22
example, the APIs which are obvious ones to me at the moment, like a Google, um,
26:27
API which monitors, you know, your Google Search Console. For example, you can
26:33
you can obviously connect into that. Um, there's a million tools out there, or I
26:38
could write my own, but of course there's an opportunity cost to writing my own,
26:44
but at least I own that information. What's your, what's your viewpoint on this?
26:50
Agents or SaaS? Do I do it myself or do I SaaS it?
26:55
Yeah. Uh, my personal feeling here and, uh, my personal vision, uh, what we are
27:02
building, uh, what we are building on right now, First of all, any improvement
27:09
in the work of the agent, it should be in your favor. But anyway, in some cases,
27:15
it might be like if you are hiring a PhD to just calculate simple numbers in
27:22
your Microsoft Excel sheet, for example. You must choose the exact type of model
27:29
that is quite enough, but not more for your particular goal right now. For
27:35
example, we are constantly doing experiments and trying to understand what's the
27:42
actual amount of data inside the midsize operating online store, for example, to
27:49
understand the current context and how to map it to the amount of tokens you
27:55
have to spend If you're connecting, for example, the frontier model like Fable
28:02
made by Anthropic. And what's interesting, if you want, for example, to connect
28:09
Claude to your Amazon store through SP-API, it will cost you about 4 million
28:15
tokens, mostly input tokens, less output, when you're processing, let's say, an
28:22
inventory of of 2,000 SKUs. But that's too simple and too dumb operation to do
28:29
it with a Frontier model, or at least even Opus or even Sonnet. So it must be
28:35
the layer. I don't want to name it SaaS because it's just, it's not a solution.
28:42
It's just an intermediate layer between the wise, sophisticated LLM and the poor
28:49
data. So your data must be, first of all, well prepared. Do it quickly. In most
28:56
cases, it must be completely deterministic. So, you know, there is no way for AI
29:02
agents, for LLMs, when the task is completely deterministic. It's just the
29:09
question of software building, just an algorithm. But it must be done under the
29:16
supervision of a simple model to decide what part of data is really important
29:22
for your current goal. For example, let me provide a clear example. If you are
29:29
changing any part of your product listing, you are doing it for purpose. You
29:36
want in the same time to be more affordable or adorable even for your buyers,
29:42
for your visitors, but it must still keep very relevant for search. I'm not just
29:49
saying search engine optimisation, it's also relevant for Amazon Ads, for Amazon
29:56
Search. So 2 players at once, your visitors and the algorithms. But the 3rd
30:02
player is also in place, the agents. They are crawling your product listing to
30:09
understand how to map what you put on listing and what Their customer, their
30:16
human being asked to buy or to consider buying from you. So is it possible to
30:22
fulfil all 3 in one time? I'm not sure about that. So you should decide what is
30:29
more important for you right now in this current situation.
30:36
That's a great question. And I love this idea of thinking about it in layers.
30:41
Because at the moment, um, it feels a, you know, very black and white. It's
30:45
like, um, I can get AI to do this or I could get SaaS to do this. But I like how
30:50
you've broke it down into layers. And actually there are things, there are
30:55
layers which are so simple you don't need AI to do it, but it's going to burn
31:00
through the tokens if you do. And so this is something I'm seeing more and more
31:04
now as people are becoming more and more aware of the costs of AI. Um, this
31:09
breaking down into layers. This is a complex task. Let me give this to Fable.
31:14
This is a medium-level task. Let me give this to Sonnet. This is, um, this is
31:19
deterministic. Let me give that to Excel, right? Or, or, or whatever. It's, it's
31:24
kind of, um, it, it seems to be becoming more and more common now, thinking, uh,
31:28
in these, in these types of layers. Do you think, do you think that's a smart
31:33
move or do you think it's all gonna sort of even out anyway and we're just
31:38
trying to make life more complicated than it needs to be right now?
31:43
I'm quite sure that it is a very smart move and that's moreover, it's the only
31:49
way that the special purpose tools, AI tools, I mean, should be created. Once
31:56
again, just imagine, You are hiring very sophisticated, probably wise PhD, but
32:02
you must map this guy's abilities and knowledge and skills to the exact task you
32:09
want to dedicate it to this person. And this matches with the modern AI agents
32:16
completely. For example, you are going to get some data prepared. Okay, but
32:22
don't forget that your context, context window is limited for understanding by
32:29
AI agents. And which is much more important, it is moving. It is moving by the
32:35
time axis, because what was completely true a week ago might become completely
32:42
wrong on the next week. For example, if your agent has enough knowledge about
32:49
the seasons, about the holidays and the special events that is extremely
32:55
important in selling online. Everyone knows about Black Friday, Cyber Monday,
33:02
everyone knows about Thanksgiving and Christmas season. Okay, but what about the
33:08
local specialities? What about the local days, special days? Or what about your
33:15
competitors? For example, if you are running the competitors' monitoring tools,
33:22
most of them now are based on the algorithms. It's okay. But what about if we
33:28
put some AI capabilities into this workflow? Just imagine in your store,
33:35
apparently you've got 2 times less conversion for running sneakers. For example,
33:41
what's going on? What's the reason? Nothing has changed in your store and in
33:48
your traffic, for example, but 2 times less sales on this product category. As a
33:55
human being or using the old tools, you have no idea about that. But what if you
34:01
have AI tool for monitoring your competitors at scale? Not 2, not 5 direct
34:08
competitors, but 100 of them. And what if your tool have insiders customer
34:14
accounts at each and every of your competitors? The robots can locate and
34:21
understand the internal loyalty campaign very quickly. They can map it to your
34:28
own inventory because probably if you are competitors, your audience somehow
34:34
intersects with those guys. And your AI agents can start almost the same loyalty
34:41
campaign at the same time, even probably before the first email reached to the
34:47
inbox of main part of the audience. It's a real game changer because you are not
34:54
reacting. You are acting at the same time on the same issue, same field as your
35:01
competitors are doing right now.
35:07
That's super powerful. I love that. I did, going back to something that you said
35:13
there, so do you, Robots, whatever language you want to use, you know, in terms
35:19
of AI monitoring my competitors, you know, having AI actively sign up to emails
35:24
and to social media accounts and to monitor them to understand what my
35:30
competitors are doing. Do you see that happening more and more with AI, or do
35:36
you think it's kind of Do you think it's kind of one of those things where we,
35:42
we, I, I, what's the phrase I'm looking for? We might, I, I think when it comes
35:47
to competitor research, a lot of us are very well-meaning, um, but very poor on
35:53
the follow-through. Um, but now we've got AI, it, it strikes me that it be— it's
35:59
never been easier to monitor your competitors. And therefore, if that's the
36:04
case, should I be doing it? And consequently, how do I deal with the fact that
36:10
my competitors are probably monitoring me an awful lot more now than they were?
36:16
Hmm. I do not have the direct answer, but I have the general one. If you see the
36:23
new technology on the market, every time it is almost the same. First of all, a
36:30
lot of people start using this technology in a very very straight way, probably
36:37
even dumb way. For example, my LinkedIn, my own LinkedIn profile is quite
36:44
popular. I've got almost 10,000 followers on LinkedIn and a lot of people are
36:50
writing in direct messages to me. Today, I'm— my inbox on LinkedIn is overloaded
36:57
with the same, very similar generic messages.
37:04
Yeah.
37:05
Every time I have a special template, I'm answering Okay, man, what kind of AI
37:11
suggester, AI copywriter you are using? Because it looks like completely
37:17
relevant to my profile, but it's completely irrelevant in terms of your ask. Why
37:23
don't you, why don't you directly say what you want from me and what you're
37:29
offering, what you are asking for? That doesn't matter. So First of all, new
37:36
technology is getting its usage at scale, but not in a good way. Then the
37:42
markets start differentiating, differentiating, and solutions are dividing into
37:48
2 types: the cheap solutions that are affordable to any business, and more
37:54
expensive but much more wise and targeted solutions. And online selling, it
38:00
should be very similar. I'm not quite sure that we still should keep the term
38:06
ecommerce. It should be transited to AI commerce because both sides, I mean
38:13
merchant side and buyer side, will have their own agent almost any time,
38:19
probably from the beginning of 2027. And in this case, we must understand how
38:25
they interact with each other. As a human being, as a buyer, how can I be quite
38:31
sure that both agents on the merchant side and my side are chasing the best
38:37
win-win deal? I have no idea how to understand that. Should I buy the cheapest
38:43
option or should I buy the most expensive? But my buying agent said, okay,
38:49
Cyril, that's exactly for you and for your budget. Okay, come on, guy. You know
38:56
everything about my budget because you have access to my Gmail and access to my
39:02
payment solution for bank account, at least read-only account. But okay, you
39:08
don't know anything about my intention. Probably I'm buying this good for my
39:14
lovely wife and I want to buy it slightly over my available budget to make her
39:20
happy in this case. Why not?
39:26
It's an interesting one, isn't it? I am curious with AI commerce. Should we stop
39:32
using ecommerce and go AI commerce? I think it's an emerging technology, isn't
39:37
it? And I, I guess I see them intertwined a little bit more. Um, I remember when
39:42
ecommerce became a thing, the whole world was going, you know, the high street's
39:47
gonna die, right? And shopping malls are gonna be a thing of the past. And for a
39:53
time they were. Um, and the high street, you know, many places is still dead and
39:58
some of the shopping malls are still dead. Because you just couldn't compete as
40:03
an in-person store, brick-and-mortar store with the costs of the online store.
40:08
But I still want to go to the shop, right? I still want to touch, smell, feel.
40:14
And I think with AI, I think there's going to be the whole convenience thing.
40:19
And I think Amazon in some respects have to be the ones that do this because
40:24
they've They're the convenience people. Like, why do I go to Amazon? Because
40:29
it's convenient. I don't want to browse. I just want to go there, bang, buy
40:35
that, dush. I want it next day and I want it, I want it, you know, easy as.
40:40
Whereas I don't buy clothes on Amazon for a reason, right? And I, and there's
40:45
got to be some part of that journey in there. And I think figuring that out is
40:50
going to be The trick, it's going to be, are websites going to die? No. Are they
40:56
going to become a bit more like the high street? Yes. Do we need to rethink it?
41:01
Yes. How do these things all work together in a seamless sort of play? And the
41:06
shops I think that do well at the moment are the shops that mix online and
41:11
in-store really well, don't they? They, you know, you've got click and collect,
41:17
you've got all kinds of things, go into the store, we'll deliver it to you and
41:22
and so on and so forth. Um, I think that's a really good sort of— it took us a
41:27
while to do the, the hybrid, um, thing well, I think. And I— it's part of me
41:32
thinks, I wonder how that translates in the world of AI, right? Where— how do we
41:38
do the hybrid thing well?
41:43
Oh, no one knows, at least now. Uh, this is once again, this is why I'm not, uh,
41:49
dropping the idea of continuing my early-stage investing as angel investor
41:54
because the main part of this game when you're investing into very early stage
42:00
innovative solutions from younger entrepreneurs, younger generations of
42:05
entrepreneurs, they are, they see this crazy world in a different way.
42:11
Yeah.
42:12
For example, you know that for the quite recent time a lot of real stores
42:18
offline was quite kind of a showroom. So a lot of people getting there, uh,
42:24
looking around, touching the fabrics, uh, understanding how it looks like, but
42:29
then later they are ordering online.
42:35
Yeah.
42:36
That's good probably for online stores. That's, uh, really good in terms of the
42:41
branding, branded stores that for a single brand only. But you know what?
42:46
Recently I had a conversation with a young, young woman. She producing the
42:52
lingerie and she had her own brand presented in several of regional stores. I
42:57
mean real stores.
43:02
Yeah.
43:03
And I asked her, okay, you are planning your online presence. Why don't you
43:09
start with selling on Amazon at a European department, European marketplace? And
43:15
you know what was the answer? I'm caring about my brand value and I don't want
43:21
to be easily found on Amazon. For me, it's kind of controversial. I see no
43:27
problem if your brand is presented on any kind of marketplace and especially at
43:33
Amazon because for some kind of customers, let's say reliability in delivery may
43:39
be much more important than recognising the brand or even the price. For me,
43:46
it's really important. I will likely buy, let's say, a bit more expensive thing
43:52
that I'm quite sure it should be, it will be delivered overnight via Amazon
43:58
facilities rather than get some savings out of this order but without estimated
44:04
delivery at all, probably a week later or something like that. And once again,
44:10
That's completely important for sellers to Amazon to understand their market
44:16
position here and their intention of their prospective buyers. What is more
44:23
important for them? Delivery, price, keeping the value of the brand or something
44:29
similar? Let's see.
44:35
Yeah, some really deep questions here, Cyril. Listen, I feel like this is going
44:40
to be a fascinating conversation in so many ways. But I'm also aware of time and
44:45
the quick departure of it. Let me, before we go further, let me get my question
44:49
asked. The question that I like to ask all my guests, what's your question for
44:54
me? This is where I go. I take your question, I'll go away and answer on social
44:59
media. But what's your question for me?
45:04
Oh, my lovely question that I'm constantly hearing from prospective investors to
45:10
Jinnify. Don't you afraid that you will be substituted by AI solution? I mean,
45:17
you personally as a businessman or an entrepreneur. So Matt, what do you think
45:24
about that? Could you be replaced by AI podcast host?
45:30
That's one of my favourite questions. Could AI replace Matt as a podcaster? One
45:34
of the things I would love, dear listener, is actually if you answered that
45:38
question for me. Should AI replace me as the podcast host? Would it make for a
45:42
better listening experience? There's probably a strong argument that it would.
45:46
Anyway, um, before I digress too much, I will of course answer that question on
45:50
social media. So if you want to know my answer, come follow me at Matt
45:54
Edmundson. Now, um, so I had a— sorry, that question just tickled the back of my
45:58
brain. Uh, how do people reach you? How do they connect with you? How do they
46:02
find out more about what you guys are up to? What's What's the best route to
46:06
you?
46:11
I like to keep connected to like-minded people over LinkedIn. So LinkedIn, Cyril
46:16
Golub, it's very easy to find me. And don't hesitate to put some comments under
46:22
the actual posting and reach out to me in direct messages.
46:27
Fantastic. We will, of course, link to Cyril's LinkedIn and company information
46:32
in the show notes, which are free for For you in your favourite podcast app.
46:37
They'll be in the YouTube description. They will be on the website at
46:41
ecommercepodcast.net. They will be in the newsletter that goes out every week.
46:46
And of course, they will be in AI. So, because why would they not? So basically,
46:51
there's no excuse for not finding them. So why not reach out to Cyril, ask him
46:55
your questions and say hello, especially about all things AI. The man knows what
47:00
he's— what he knows. And, um, I feel like we're just sort of— I, I do say this a
47:05
lot, and I appreciate I say this a lot. We're just scratching the surface.
47:09
You're starting to get somewhere. Um, and sometimes an hour is just never
47:14
enough. But Cyril, thank you so much for joining us, man. Genuinely appreciate
47:19
having you on the show. But before we go, I like to do this thing called saving
47:23
the best till last. And so in the, in the sort of the few minutes that we've got
47:28
together, um, Cyril, I'm curious. Someone's sitting here listening to the
47:33
podcast. They are an e-com business, um, turning over maybe a million online, a
47:38
small sort of e-com brand. How— and I think they're looking across the world of
47:42
AI and feeling slightly overwhelmed, if I'm honest with you. They know how to
47:47
get Claude to do a few things, maybe fix a few things on their Shopify site. How
47:52
should they be thinking? thinking about AI right now, do you think?
47:56
Think about AI like your assistant that must be supervised and must be clearly
48:03
set with the goal and the expected outcome. Don't think that it is kind of a
48:09
magician or wizard that will, will do everything without you and without your
48:16
supervision.
48:24
Love that. It's not a wizard, um, as much as we all like to think that it is.
48:28
Uh, but yeah, I'm sure Claude, when Claude listens to this episode, which we
48:33
know Claude does, is going to go, hang on a minute, Matt, I am actually a
48:37
wizard. Uh, I, I, I'm, I'm, I am. We'll see if AI responds. Uh, but Cyril, thank
48:42
you so much for coming on the show, man. Genuinely love the conversation,
48:47
really. Really appreciate you being here and appreciate you, man. Genuinely.
48:51
Thanks, Matt. Thanks for inviting me and high five to the audience.
48:56
Yeah, high five, definitely. As it goes. So there you go. Another episode all
49:00
wrapped up of the eCommerce Podcast. Thank you so much for joining us this week.
49:04
Like I say, make sure you like and subscribe and do all of that good stuff
49:08
because we've got more shows coming up. And of course, I don't want you to miss
49:12
any of them. I do love doing this, podcast. Hopefully you get some value out of
49:16
it. Like I say, like, subscribe, check out ecommercepodcast.net for all the back
49:19
catalog archives and even the notes from today's conversation with Cyril. But
49:23
yeah, that's it from me. That's it from Cyril. Thank you so much for joining us.
49:27
Have a phenomenal week wherever you are in the world. I'll see you next time.
49:31
Bye for now.