Join host Matt Edmundson in this insightful episode of the eCommerce Podcast as he sits down with returning guest Valon Xhafa from Bahamics to dive deep into the disruptive power of AI in the world of eCommerce.

Key Takeaways:

1. AI's Role in Automating eCommerce Tasks: Valon Xhafa explains how AI is revolutionising the eCommerce industry by automating repetitive tasks, thereby allowing businesses to focus more on strategic and creative aspects. This includes optimiswing organic traffic, reducing cart abandonment, and enhancing SEO efforts.
2. The Importance of Relevance in SEO: A significant insight shared by Valon is the critical role of relevance in SEO. He highlights that measuring and maintaining content relevance is crucial for ranking higher on search engines. This goes beyond traditional metrics like domain authority, emphasising the need for accurate, contextually relevant content to achieve better SEO results.
3. Future Trends in AI for eCommerce: Valon predicts that the next phase of AI development will focus on improving language models to enhance accuracy and cost-efficiency. As AI continues to evolve, businesses will see more advanced, faster, and cost-effective AI solutions that will drive innovation and efficiency in the eCommerce space.

This weeks insights are vital for anyone trying to keep up with the changing world of eCommerce. Make sure to listen every week so you can stay ahead of the eCommerce game.

AI's Revolutionary Impact on the eCommerce World

So you're not going to want to miss this one. Is AI still as cool as everybody thought it was a few weeks ago? We just don't know. We're going to get into all of that. But before I do, a warm welcome to you. If this is your first time here with us on the show, we just chat about eCommerce and all things to do with how to grow online businesses like you.

More details soon, so stay tuned, make sure you subscribe and I'll let you know more about that as we get closer to it. Now Let's talk about Valon, who previously worked as an AI scientist at Google, which, let's be real, not many of us have done, and he's worked at other research institutes developing sophisticated AI techniques and algorithms.

Uh, the episode, Valum, when we were talking [00:02:00] again about AI and how it works with abandonment, uh, car abandonment and all that sort of good stuff. Uh, but today, let's carry on that conversation. Welcome back to the show. Good to have you on, man. How are you doing? I'm good.

And the primary reason was that AI before used to be like this cool thing, great thing, you know, but mostly in the back end. The normal show, the average show would never actually see how that works and the impact of it. Well, probably like [00:03:00] indirectly, but not like so directly, you know, so, uh, it was a great, a great, a great turn point for sure.

And then, you know, like the release of chat GPT and all the different language models, um, you know, like this has a huge impact to, you know, like reach a huge, huge mass of, of, you know, like people to see that actually AI has a huge impact. So, um, and this part, a huge interest all over like different industries, pretty much, which is, which is great.

Right. So, um, we're happy to ride that wave, if you can say so. Um, we saw a huge interest and a lot of expectations. Uh, but yeah, very interesting. Yeah, it's

I think we single handedly changed, you know, the whole nation industry around AI, but that's okay. [00:04:00] If only that was true, right? Um, but it's, it is interesting the journey that we have come on with AI and how it started off with this huge bang. And everybody was talking about it. Everyone still is talking about it.

It doesn't seem to have gone anywhere. Now, I know that when we were chatting before we hit the record button, um, you were like, I hope AI doesn't go the way cryptocurrency goes. What do you mean by that? Because I think this is a really interesting idea.

Valon Xhafa: A lot of, a lot of, uh, hypes took place in the past, right?

So it literally took us like almost a decade. 10 years in research and [00:05:00] development to be able to come to a stage where chatbots can actually formulate Good questions or can formulate cohesive questions, you know, they are still not at the level that Most people expect to be

yeah,

Yeah, right So it took us a long time and it took us like, uh, like to, to actually perform a miracle in terms of AI to be able to pull something off so that we can gain the interest from, um, the customer and the industry back to where we are. So what I mean by that is that whenever there is a hype, It's usually super hyped, um, and then expectations are risen, but then, um, we're, we're assuming the industry starts to realize that actually, um, expectations are not met.

Yeah.

And

Valon Xhafa: it's not just transactional, you know.

And there were like some kind of like waves. And then there was some increase in these waves. And these particular increases on the trend. [00:07:00] of users searching for that, they came particularly because something was released, something new was released, a new model or a new tech or a new company or something, right?

So the great part is that, uh, the whole, uh, investment community, uh, VCs, they are like, um, putting a lot of money into development, which is great. It didn't used to be like this at this scale, uh, before pandemic. So this interest is going to lead to new innovations, which hopefully is going to like push to newer and newer applications in different industries, like we're doing it here with, uh, with Behemix and eCommerce.

Matt Edmundson: So what, um, what sort of new innovations do you see on the horizon? I mean, the time of recording, it's July 24. Um, where do you see it going the next sort of, I don't know, 6 12 months? Can we predict that far in the future where AI is concerned, or is it more, we're struggling to predict the next 6 12 minutes, let alone the next 6 12 months?

Yeah. And they are more like cost effective. This is gonna be huge demand. Um, the reason why is this, the case is that the existing state of, of, of training and building, uh, language models, running them on, on, on cloud is, uh, highly ineffective. It's not cost effective, it's not financially, it doesn't make financial sense in the long run.

That's, that's a scale and this scale is literally, this scale is like unaffordable. Like think about like, how many prompts that they handle per day, like millions of prompts.

Yeah.

Valon Xhafa: So the majority of these language model companies, they're actually losing money. You know, by being able to process and, uh, you know, like handle all these requests because the status, the state of the language models and the hardware is not optimized yet.

So in the next few months, probably a year or even two years, the whole focus is going to be on optimizing these, making these models more efficient so that we can have these for the long run. Otherwise, this is not going to make sense financially. Right.

Matt Edmundson: That's really interesting. I've not heard of that site.

Um, into your own company. So how have you dealt with this technology barrier yourself? And if the cost to entry is significantly high, how have you guys dealt with that?

Valon Xhafa: Yeah. So, uh, there were like a few measures. Um, we as a company, um, we as a company, uh, we're not, uh, highly focused on language models. Um, and the primary reason is that language models, they still lack in, in reasoning and, and causal reasoning, right?

Strategy or something. This is something that, um, mid enterprise clients, but also other clients as well, they're not going to really trust, you know, so because of these, this effects of hallucination and a lot of other things, language models are not at that point yet. Instead, we use, like, deep learning models that have, that we can implement the causal reasoning, that we can trace back.

So this is first, but still. This process is still, um, we're like, um, you know, like it's not, it's not cost effective, you know, it's still, uh, we're expensive and still needs to be scaled. And for that reason, we, we build our own, you know, like data centers. We both, uh, we've, we've acquired a lot of, um, um, you know, like GPUs, graphic cards.

We build our own data centers from the beginning on. Which is highly important for every AI company out there. I understand that the GPUs, they get older and older with time, right? And like, you know, like every 18 months, you know, like the capacity doubles based on Moore's Law, you know? Uh, but still, still, in the short term, even in that short term, in that 18 months period, the ROI is clear on building your own, even mini server.

Matt Edmundson: Well, yeah, I, it's, it's interesting. I mean, you know, How you've done it and how you've approached it.

Um, and I'm, I'm curious, what have you discovered then, you know, you've been working with AI, you're, you know, more than the average bloke on the street, Balint, uh, about this whole thing. What have you discovered about AI over the last, well, since we last spoke, that's maybe a bit surprising, that's, or that's really helped or that's making some big changes in eCommerce.

Yeah, right. It's not here to replace us completely. It's here to help us transition to things that we as humans are capable of doing, like reasoning, creativity, and so many other things, right? And leave these manual, redundant, repetitive tasks To

AI,

Valon Xhafa: right? And for that specific reason, um, AI is, is, is great for, for these cases and, and brands, they want to see automations.

So they, um, the majority of the brands, they do a lot of, like, repetitive manual tasks that are daily that can be automated. So this is the common interest and the common trend that we see these days in eCommerce, and AI is perfect in these cases. Why? Because if you want to automate something, you need to understand that task.

You need to understand how things come to be. Which is, uh, what AI literally does, right? AI has this capability of predicting things when you have a high accuracy of predicting an outcome. You can use the same process to also explain how did that outcome come to be. But we already discussed that language model, they still don't have this reasoning capability.

with

Valon Xhafa: automations running on AI. Why organic traffic? Well, there are a few things that changed since pandemic. First of all, before pandemic, um, money used to be cheap, low interest rates.

Ads used to be cheap, so the common strategy before was like, you get more money, you throw this money into ads, and you're gonna grow, bring more traffic. Simple strategy, clear, um, strategy here. But things have changed, right? You had all of these, like, privacy updates from Apple, um, money's not as cheap as it used to be, unfortunately.

So, there is a distribution of focus, and, um, from this, because of all these different changes, There is also a clear change in the trend on how brands bring traffic now. They are, um, they are not relying completely on bringing traffic through ads. They're trying to distribute this. They're trying to bring more traffic through different other channels, potentially organic.

Um, there are so many platforms out there that help you, you know, like with paid and help you optimize things and get ROI and build strategies and this and that. But in terms of organic, it's, it's way more complicated. Because you have, for example, for the majority of the brands we work with, which are primarily mid enterprise, Or maybe let's, let's kind of like step back and let's simplify things, right?

If you have a small store, you're just getting started, the fastest way to bring traffic are ads, no doubt about it, right? So we have, we have clients that are all like on Shopify. Uh, few, few hundred thousand traffic, uh, the majority of the traffic is paid, but that as you start to move towards a mid market, then enterprise, there's distribution of traffic changes.

Yeah.

Yeah.

Valon Xhafa: So in general, for example, based on numbers that we have, 30 35 percent of the traffic is organic for mid enterprise, for the clients that we have. For smaller sources, like 5%, 10%.

Max, but this traffic, uh, accounts for like 50, 55 percent of the revenue. That's how much the impact is.

And

Valon Xhafa: because of that, because of that, there is a huge interest. There is, it's very important to establish your brand as you move towards mid enterprise market. Because relying completely on ads, especially these days, um, it's not cost effective.

Matt Edmundson: No, it is. I I've seen, uh, uh, it's interesting, uh, hearing you talk about this because it's one of the things that I've noticed a lot more people now are looking at SEO. Again, everything comes in cycles, isn't it?

SEO was really trendy a few years ago, and then it sort of stopped being trendy because ads became easier. And now it seems like SEO has become the big thing again. You know, how do we, how do we, how do we grow our organic traffic? And I'm intrigued by this topic, Val, and I'm actually quite pleased to talk about it because Recently, Google have made changes, as best as I understand it, to the search algorithm where they're trying to sort of push out websites that are trying to use AI in many ways to bump up their organic rankings, right?

How do you deal with that? Yeah,

Out there by building guest posts, for example, which is a common process. It's where it's, it's where a straightforward process, right? There is nothing wrong with it, right? Um, in this way, by not really knowing with which people you should partner up to write guest posts or blog posts or other things. You know, and, and doing this, which can take time, but, uh, on the other hand, if you do it wrong, if you do it with the wrong type of people, just because a Google AI algorithm, um, can penalize these cases and actually damage your position instead of helping you, we don't believe that's the right way.

So, if you want to write a guest post, for example, um, relevancy. Or in SEO in general, relevance is very important. It's the most important factor out there. But you don't really have data or platforms that offer you data about relevancy. How is that possible? We still rely on cases, on data that are based on domain authority, for example.

A thing that, based on the data that we see, has almost no impact. Oh wow, okay. That's good news. So relevance, relevance Matt, as in information. is the highest impactful factor out there in SEO. And there is nothing wrong about measuring it and understanding how relevance has an impact on your rankings.

What's wrong with it, right? There is literally nothing wrong, pretty much. So, take into account that relevance has the highest impact and how often you publish things. And how often other people talk about you, which means that they, uh, uh, build backlinks or they, um, build, you know, like links that go to your site.

These two factors have the highest impact right now. And we see this with data. But still, this is like a, you know, like a high level overview because there are so many other things that these two factors can be broken down. It's not just like the relevance of your content with a keyword, but it's also how relevant is this company that is talking about you.

Valon Xhafa: It cannot be your competitor, which means it cannot be highly relevant, but also it cannot be a company that has nothing to do with your, uh, with your domain or industry or whatever. So in a nutshell, based on the data that we see is that the differences were a slight, because if you think they're called like the relevance from zero to 100.

You know, the difference can be like 10 points or something, 10 percentage points. It's not that high, but we as humans were unable to actually, uh, without measuring relevance, come up with a relevance that is, that has that precision. So, taking all these different things into account, we've, uh, we've established and we've, uh, we were able to measure relevance, to be able to explain the rankings, to be able to understand what has an impact, which.

And we've seen that whenever you were generating something or rewriting that thing, the relevance was becoming less and less and less.

Matt Edmundson: Right.

Valon Xhafa: So what this means is that the most important factors that is out there in SEO relevance is not really being taken into account by a lot of content writing tools out there.

You know, just because the domain has, let's say some backlinks, doesn't mean that these backlinks are state of the art. The idea is being able to actually understand, um, how the relevance has an impact.

Yeah.

Valon Xhafa: Paul, do you write content that are relevant? And relevance is here to stay, because relevance is a metric that is highly scalable to be calculated when it comes to the engineering part.

You know, if you, for example, if somebody wants to write about you, you can actually simulate and predict if that's going to help you in the ranking or not. Even before you publish an article somewhere, that's how impactful the process is. So far it used to be like they tried to write as many guest posts as possible, some were helping you, some were backfiring, you know, but that's pretty much it.

So if you see that some are bad, why would you publish there? Go somewhere else, you know? So that's what I'm kind of like talking about here. We're talking about like being able to have a GPS and navigation when it comes to building organic traffic in the right way, right? It's not about like It's not about like hacking the system or finding loopholes in the Google algorithm in terms of like inserting links where they should not be and we have seen this for example we see cases if if an article or a backing or a guest post talks about you but then at the same time they talk also for another 10 companies Transcribed You know, that's not as good as just talking about you.

Would that make sense?

Matt Edmundson: Yeah, totally. So, it's not like, um, AI is revolutionizing the principles of SEO, so the principles are still the same. Good quality content, good backlinks, relevancy, and so on and so forth. What you're saying, where I think AI has struggled is, well, I guess where everyone has used it, like, you know, people like me, as you go to chat, GPT or Claude or, you know, whatever you're.

Much more, you know, let's think about relevance and and let's look not just at the micro here. Let's look at the macro How is that going to work in the in the bigger picture if I've understood it?

And what we have seen is that the level, the level, the margin of you taking an action that can help you versus, uh, can damage [00:32:00] you is pretty close. Right? We're talking about like few percentage points in relevance that can help you in ranking higher versus damage you.

And

Valon Xhafa: because of, because of the language models, they don't have this information.

We have seen that it doesn't really work. The reason why is that when you write a content, for example, based on the results that we have, is that that content, you know what, it cannot be like highly relevant to your, uh, homepage, for example, because, and then you're kind of like, you're kind of like [00:33:00] damaging your homepage, which is the most important part of your site, right?

You know, and it's not only that, but it's like, for example, if you build a gas post from, from, uh, from a different domain, and they are linking to you, not only the article that is talking about you has an impact, but even the homepage where that article is coming from. And the relevance of that homepage to your homepage and to your article and to your keyword has a huge impact.

Um, this means that a backlink that used to be good a few months ago, or a guest list that used to be good a few months ago, it doesn't mean that it's going to stay with that quality forever. Because relevance is a measure of text content. And as soon as, let's say, the homepage of the backlink changes, you know, not the article, but the homepage, this also means that the relevancy can decrease, which also means that the backlink quality can decrease.

So whenever you change something, things are going to, the relevance is going to change, which it can have a huge impact on, on your ranking. Does it make sense, Matt, or?

Matt Edmundson: Yeah, it does. It totally makes sense. And I'm really intrigued to see where this all goes, Valen, because it's, I do see that organic traffic is going to become more and more of a big deal again.

And using, Using AI to give us a competitive advantage on that and to help us do that right and well in a way that's going to work with those platforms, as opposed to trying to find a cheat code backdoor in for a little while, um, I think is, is exciting. I think it's one of those things that I'm really intrigued to see how it develops because I don't see many people talking about it.

Um, I see a lot of platforms talking about the micro. I don't see. Many people talking about the macro and how it all sort of fits together. So I think that's interesting. I think it'd be a bit of a game changer. So yeah, it makes total sense. Um, I'm intrigued to see how this works with you guys as well, how Behamex sort of handles all of this, um, going forward, I, I, I'm, I'm utterly intrigued.

Um, you don't need to, why would you want to be an expert in something that changes all the time? Right. Because a lot of people talk about like the Google recipe, which is something that is a very interesting or interesting talk. When I hear about, uh, a lot of SEO people talk about the Google recipe and, uh, you know, like there were a few leaks from Google about what kind of factors they use and they were talking like, we didn't see the recipe.

The

PageRank

Valon Xhafa: algorithm was a recipe, because you could trace, you could, you could understand the whole in and outs. It was a descriptive algorithm. The current algorithms that Google uses and the whole, probably, not the, probably, but the whole world uses, like, these are probabilistic algorithms. There is no recipe there.

And they have these large algorithms. So, there's a lot of things that try to optimize the click through rate. The chances that a user, when they land on Google, they're going to click something. That's what they're trying to optimize. So the recipe here, it's a large model that is constantly learning and changing and adopting things.

The relevance Is the key component, which is a huge difference from before where domain authority used to be a key component. But still, these algorithms are trying to optimize things and improve things all the time. And they are changing all the time. So there is no recipe. If you still try to go after the recipe, you will never find the recipe.

Nothing. You know, it's a clear concept, right? So there is no recipe because everything is embedded on these large, So, there you have it. I hope you found this video helpful. If you have any questions, please leave them in the comments. I'm Matt Edmundson, and I'll see you in the next one. This is Garnet.

It's going to give you a good score that, well, you were able to handle this user that I brought to you and the user is happy because they're coming back to Google. That's how it works. It's a win win for both. You write great content, you optimize your page, users save more time, they convert, maybe they buy something, and you and Google share the revenue.

Not only within eCommerce, Matt, but all over different industries.

Yeah,

Valon Xhafa: that, uh, we are trying redundant things, repetitive things, things that we should not do. Instead of focusing on creative stuff, creative thinking ways to build new things, to come up with new challenges, we are kind of like, I'm not saying wasting, but like spending our time on these things that they don't really have a huge impact in the longterm, but they are more repetitive.

So

Matt Edmundson: fantastic. Well, so many questions in my head, but I'm aware of time. Um, and it's been a fascinating, as always, Valen, a fascinating conversation about the industry and where it's all going. And I, I love that when you were saying there is no recipe at all. In my head, all I could hear was the kid from, I don't know if you ever saw the movie The Matrix, where he goes, there is no spoon.

Uh, that's, there is no spoon. Um, and so that's what I heard, maybe showing my age now. Um, if people want to reach out to you, if they want to find out more about what's going on with you, find out more about how maybe they can, you know, how you can help them with their SEO with Behamex and stuff. How do they do that?

What's the best way?

Or maybe 90 percent automated, because we have seen this case, Matt, in a lot of different industries. I'm bringing this example for the end. Like back in the 70s, stock trading used to be completely manual. Over the phone, everything else, right? Like, we do SEO and a lot of other things right now, manually.

And, uh, there were a few people who tried to, um, you know, like, start to build algorithms because the first computers came out. And here was the biggest issue. They didn't have data to train these algorithms or build these algorithms because there were no CSVs files and there were, there were no data. So they literally had to go to libraries.

That job is dead. If somebody does stock trading for a company or something, they're probably just managing computers doing the algorithms at a high frequency rate, and you cannot really compete with these algorithms. In my opinion, this is how a lot of things are going to change in the future for good.

Because we didn't have that. So I'm comparing these two cases as, as examples on how the stock trading used to be versus how the SEO is right now and where the SEO can be in the future. It's going to be, it's going to be, it's going to be automated at a high level, pretty much.

Matt Edmundson: Yeah, well, I'm looking forward to seeing where it goes.

Um, I, I am looking forward to seeing where it goes and in stock trading as well. I'm, I'm still looking forward. to find in the program that can, you know, just do it all for me and make more money. But, uh, but no, it's really fascinating. Uh, Valen, thank you for coming on the show. Before we go, right, we do this.

Valon Xhafa: Very good question, Matt. Um, What do you think, what do you think about the, uh, what do you think, uh, are going to be some, or what do you think is going to be an interesting trend in eCommerce in the future?

Matt Edmundson: Okay, an interesting trend. Or,

Valon Xhafa: maybe, maybe, maybe, maybe, let's rephrase it. How is the online shopping, um, experience going to look like in the future?

Matt Edmundson: Okay.

Valon Xhafa: Because it hasn't changed that much in the last 20 years.

But I have some really interesting answers to that, especially after a recent visit to the States. But I'll go more into that on social media. Val and listen, thank you for coming on to the show, man. Really appreciate you. Coming on and just sharing your whole insight into AI. It's always good to talk to knowledgeable people and you are very knowledgeable, sir, uh, in this whole area.

So thanks for coming on. Uh, we will of course link to Valon and everything, uh, his, uh, LinkedIn and his company bahamix. com in the show notes. Uh, but Valon, that's it from, from me. Thank you, man. I really appreciate you being on.

Valon Xhafa: Hey, Matt. Thanks. Thanks for having me. It's always a pleasure.

Matt Edmundson: Always a pleasure, indeed.

No doubt he'll be on the show again soon in about 18 months. So make sure you follow the eCommerce Podcasts wherever you get your podcasts from, because you're going to want to hear that episode as well as all the other ones we've got lined up. And in case no one has told you yet today, dear listener, you are awesome.

Oh yes, our theme song is written by Josh Edmundson. And as I mentioned, if you'd like to read the transcript, the show notes, uh, all that sort of good stuff, just head over to the website, eCommercePodcast. net. But that's it from me. That's it from Valon. Thank you so much for joining us. Have a fantastic week wherever you are in the world.

I'll see you next time.