99. The AI news changing how ecommerce is run, and the future of media buying

Welcome to Zero to One Hundred, the podcast about taking an e-commerce company from an idea to over one hundred million kronor in revenue. We are Fedja and Jacob, and we will give you all our best ideas for growing your company. A warm welcome back to a new episode of Zero to One Hundred. My name is Jacob Wibom Westerberg, and with me today is Fedja Porobic.

Thank you very much, Jacob. Thank you very much. A second week in a row, so I am just happy to be on the podcast. We are on a roll. We are releasing episodes two weeks in a row, and next week's episode is episode one hundred, which, by the way, is already recorded. We will make it to three weeks in a row.

That is right. Yes, then we have that. We recorded the live podcast a few weeks ago, but we knew when we recorded it that it would not be released immediately. So all of you who were there [unclear in the Swedish source: för ut så små plats] experienced it a long time ago. So perhaps you will get a favorite repeat. We have still had positive feedback, anyway, from those who... Extremely positive feedback.

So many people want us to do it again. I do not have the energy, to be honest. Once a month might be possible. At most, we should say. I thought there was a lot to organize. Even though we outsourced ninety percent.

I mean, Qliro did quite a big... Yes, right. Qliro did all the work. But as I said, we outsourced ninety percent. It was still fucking hard work. The ten percent that tends to be the work. Yes, because what... Making the decisions. Making the decisions. Should we do it this way or that way? Do you want seven emails or six emails? Should we have this or this? So it was... It takes up mindshare even though there were not that many to-dos on us.

Exactly. So there were no to-dos on us, but there was mindshare. Anyway, there will be no live podcasts this year, at least. We will do it again next year if all goes well, we feel well, want to do it and have the energy. Exactly. We are somewhat behind the news flow. So today I thought we could talk through some of the things that have happened in the past few weeks.

And then we can sort out what we think is big and small. But perhaps we should start with a quick update if you have anything exciting from Mockberg or Såpa or the like.

We always have exciting updates. A lot is happening at Twistboxes. A lot is happening everywhere. I have seen a lot happening at Juniper too. I think we should run through a little. These are the highlights and challenges we have. Twistboxes' challenge right now is that we are extremely understaffed. You would like a much bigger team, because tackling many different markets and many different agencies takes more time than I expected when we entered the project.

So that is a priority at Twistboxes. I do not have an answer right now for how to solve it. But the good thing is that we sell online and have rolled out a large part of what we actually wanted. So we can get advertising started. But there are major challenges with the traffic coming in. You want to spend much more time optimizing and analyzing, things we have not had time for over the past few days.

So that is the hat I am going to put on now, really starting tomorrow. So there will be a lot of work getting Twistboxes organized, which I am [unclear in the Swedish source: jätteglatt emot].

It might be worth having a discussion with Global-e, which actually takes on the role of merchant of record in all foreign countries. If you want to sell to the US, the UK, Norway and countries like that, where for various reasons there are lots of local rules, you can pay to get out of some of the administrative problems with their setup, where they take over your Shopify checkout and you avoid becoming liable for tax in the US and UK.

So that could be an option. [unclear in the Swedish source: För det som Love blir skatteskyldig i Sverige] Yes, exactly. [inaudible] Making sure you do not become so, really. Then your, those kinds of payments will still come as an invoice from them, obviously. But at least it becomes different.

Okay, I know they have wanted to be on our podcast several times and have pinged us. So we will see whether we have them on. If you listening want more internationalization, payments and this kind of discussion, we could bring Casper from Global-e on. Then we could talk it through with him. Let us do that.

Ping us then, me or Jacob, and we promise to get back to you. If you do not ping us, that means he will not be on. That is a true story. We do not want that kind of sponsored episode, only if it is something you want to listen to. So it is, it will be an unsponsored episode with Casper. I know many people have reached out after the Qliro event.

People have been curious about Qliro as a payment provider. I have been talking with the CEO, Christoffer, about him coming on the podcast sometime this spring, at least. We will see when we can sync our calendars. Absolutely. So we can talk about the next level of payment processing and how to optimize checkout to make more money, quite simply.

As Såpa has done, for example. We have reduced our costs, increased conversion and done a splendid job. So that is also one of the things I have been struggling with over the past two weeks. Even though I have not worked that much, that one particular thing has been that we had a bug in the Shopify checkout and have solved it together with Qliro.

They were very quick to find it, and once we identified the problem they could solve it the same day. But there were a few days when we did not really understand what the problem was. And at first we thought traffic was dropping out at the last step and choosing Swish a higher proportion of the time for some unknown reason. [unclear in the Swedish source: Men sen då störde om det] Was it some BankID issue?

No, it was just, this was so strange. Everyone coming from an iOS device via a Facebook ad could not pay by invoice. Yes, that is right. And then you have to identify yourself with BankID. Because we also had a little trouble with this. But yes, you did too. But it has been fixed since Friday two weeks ago.

So there is no concern now. But a lot of orders over a short period were... Exactly, the buy now, pay later option was [unclear in the Swedish source: bullmat]. Yes, and people switched from their iOS device to their Safari desktop or their Android or another device and bought. At least for us at Såpa. So we were not affected too badly.

Some orders dropped out completely, but I think they will come back. Because where else am I going to buy soap? Where else am I going to buy your buy now, pay later soap? But many people choose the buy now, pay later option, especially in January and February. It increases. Even for small products. I imagine it must increase at Juniper too. You also have it in yours.

How many buy on...? I do not actually have that data in my head. But... Poor form from you. Poor form from me. Yes, exactly. Perhaps I should look into it. But I do not know, it is something you need. Buying jars of soap in installments feels... But it is people's convenience, not having to write down their, I mean it is just a personal identity number. It is quite simple.

You are lying in bed, you know, sitting at a bus stop, breastfeeding a child with your phone in the other hand. You know how people move around with phones nowadays. It was convenient and easy not to have to get a card out or remember your CVV or anything else.

No, exactly. I understand that is why you end up there. But for me there is an alternative instead: Apple Pay. You press twice on the side of your phone and you are done. But that is a direct payment charged to a card, right? Yes, but you can have it on a credit card.

Yes, I do not have Apple, so. But then there is Google Pay... Yes, but it is sort of not there. In many Shopify stores you cannot pay with Apple Pay.

No, but I think Apple Pay is absolutely magical. It is quick. You have your billing details and so on. No, I agree. I love Google Pay too. Those types of payment methods are very quick. You identify yourself. You very rarely have problems. We have talked a lot about Dubai and Ethiopia now, but I traveled around and had a lot of problems because I had to pay large bills on my Amex card... Mm.

...and Amex calls me, thinking it is fraud all the time. Yes, I understand that. But when it goes through Google Pay and I am still using Amex, it goes through. Why, I do not know. Yes, I do not know.

That is the story. Yes, story. So Amex gets blocked in Ethiopia for understandable reasons. But not through Google. No, but yes, okay, so there are a few things there. Juniper had a tough start to this year in some ways, in that we did not have bed linen and towels in stock. Now we do.

Now we do. Yes! So they were back around the beginning of March. And since then it has been going very well. So we... No shit. ...and we were only, I do not know exactly, but we were about five or ten percent behind last year, year to date, which was still very strong considering we had nothing to sell.

Or, you could buy blue pillowcases with white duvet covers and beige bottom sheets. But very few people want to do that. So it is a little like a fashion company only having XXL and XS left in its size ladders. Then it is very difficult to buy traffic. So we were quite pleased with how the year had gone, despite a tough start, if you just compared comps with last year.

But since then it has gone very well. Now I hope and think we will never again need to wonder whether we are growing compared with last year, this year. I think we will, at least I think we will grow so much this year that beating last year will feel fairly easy.

We had such a hard time, particularly in H1 last year. So those are poor comps to compare against. We should be able to beat last year without any problem. And you have done a good job too. The groundwork with the soap, and that will lay a foundation for people to buy more bed linen in the long term.

Exactly. So, but we will see. No, but right now June feels very good. It has been a long time since the momentum felt as strong as it does now, actually. So that is fun. Things are rolling along, the team is doing a fucking great job, and... Fun, we are going to launch a major new product group in, well, quite soon.

May I ask which product group? In a couple of weeks. Yes, it is within the bed linen segment. But I do not feel... You want to say it is a pillow, or something. No, I feel I do not want to reveal how the team wants to roll this out. They can [unclear in the Swedish source: be the same here] when they are going to talk about it.

For all our few thousand listeners. Yes, they can, I feel it is up to the team to decide when they want to. We will discuss it after the launch, then, what it looks like. I know it is a new product group. Yes, good cliffhanger.

Exactly, good cliffhanger. No, but it feels exciting. And a fun, fun start to the year. I have to say things have also been quite sluggish for Såpa for the first time ever. I have talked a little about it before. Now a few more weeks have passed. We have, we are coming off an incredibly strong Q1 2025. Because in Q4 2024 we released limescale remover. In Q2 we released laundry detergent, in Q3 we released something else.

So many releases during 2024 meant a lot of people were keen to try new products. You do not need seventeen bottles of limescale remover the way you might with [unclear in the Swedish source: Essens scents]. One bottle of limescale remover lasts quite a long time. Anyway, many people stocked up during Q1 2025. So I knew there are fewer new customers. I mean, it is harder for us to drive new customers for that type of product this year than it was last year.

Then total revenue is still increasing. Total profitability is still increasing. So fundamentally I feel very, very confident that things are going very well. But it is just double digits instead of triple digits. So I am not used to not growing by triple digits, if you understand what I mean. So going into the year, of course we have triple-digit ambitions there.

That means growing a hundred percent. Over a hundred percent. But that is in total, including B2B. We knew we might grow by mid-to-high double digits. So we knew, we are not there yet. But we thought we might grow fifty to sixty percent this year, year over year.

Online. Online. But the major growth will come from Axfood, ICA and Coop. Where we have rolled out and had an incredibly good proof of concept. But now we have brought in consultants, and it is [unclear in the Swedish source: jitte och jatte]. I mean, rolling things out in grocery retail is not that simple if you are an online retailer. Especially, it is much easier for, well, for Mockberg, when we work with Mockberg, to roll out to Smycka or Guldfynd or Albert, you know, chains.

Than it is for Såpa to roll out in grocery retail. Because you get a very small space hidden away somewhere, and you have to think about what the packaging says compared with other packaging, and legal requirements, producer responsibility, because you manufacture it yourself. It is not as though you bought the products in. So other kinds of commitments. And if you get in, you want high sell-through or you get thrown out, and there are only X number of grocery chains covering ninety-something percent of total grocery retail in Sweden.

Definitely. Of which one chain covers over fifty percent. So you do not want to mess up with that chain. So with that said, we have brought in external consultants who are fucking expensive, millions of kronor to solve this problem. Because I am not an expert, nobody on the team is an expert in grocery retail, and we just want a decent rollout. Let it take the time it takes.

We will not do the aggressive rollout we initially envisaged during 2025. We want to do a good rollout. Another positive piece of news is that we have done two different consumer surveys costing several hundred thousand. Yes. Where people are interviewed qualitatively. And you find out what they think about our brand, but also other brands. And how many, I mean, it was very surprising, because we have advertised Såpa since February 2023.

Before that, basically nobody knew Såpa, Västerbottens Såpa. Now a large share, a large double-digit share of the population knew Västerbottens Såpa, and that makes me so proud. And the only place you could have seen it is, well, okay, there are a few places.

Paid social and radio and [unclear in the Swedish source: Q]. Yes, maybe, but it is still mostly paid social. Yes, that is right. I saw the major impact of Meta advertising in black and white when you interview thousands of people and X number of people know you. And I have seen, you know, this type of interview. I have seen that Arla does not have a hundred percent recognition, neither does H&M, nor Lindex.

I mean, none of these big chains has that. Not even Swish, or something like that which has popped up out of nowhere and suddenly everyone uses, has a hundred percent recognition. So you know nobody has a hundred percent penetration in brand awareness. The strongest brands are somewhere at eighty to ninety percent, and then it only goes down. And when you see we are still doing quite well compared with all our competitors, you become very, very happy and very, very proud.

So that is a big thing that has happened for Såpa. Despite things going a little worse than we budgeted. Yes, I think one thing I, perhaps you are already looking at it, one thing I think I would consider if you feel things are a little slower than you expected. That is spending more on existing customers on Meta.

Why? We have just gone the other way. We have done that in... But the natural thing is always to go the other way. Yes, I know, the natural thing is to find new customers. But I understand the reasoning. For us the reasoning was, if you spend more on existing customers, then existing customers who already love your brand will write more comments, and you get better social proof. And that happens.

And hopefully they buy more too. Because you have lots of people who do not buy. Yes, but then they do not buy that much more, because they already bought during Black Week and the Christmas shopping period. So they do not buy much more of that in January, when we started increasing spend on existing customers. Then in February we did the opposite, increasing spend on new customers.

That is the natural thing. So that is less strange. But we have still seen with a few brands lately that the [unclear in the Swedish source: external customer spend] has quite a large incremental impact. And sometimes it is easy, you like to look at frequency as a measure of how well you are reaching the existing customer group. But you also need to look at reach in that group.

If you have customer lists with 250,000 people on Meta, it might identify [unclear in the Swedish source: 150 to 200] of them. And then if you just look at your reach over the past thirty days, perhaps it is only sixty or seventy thousand out of the 125,000. And even if frequency is four per week or four on a rolling seven-day basis in that group, you need to start thinking about how to expand reach within your existing customer group.

And there can relatively often be untapped potential there. Because it is like this: I have bought Västerbottens Såpa, but I do not do it all the time. I do it every now and then. And I could probably be persuaded, now I have a relationship with the brand, but if I try to see myself as an ordinary consumer, perhaps there is a group of people there who are actually relatively easy to persuade to buy again, but simply do not do so naturally.

I get that. [inaudible] Worth looking into, anyway. Yes, it is worth looking into. I know we reach far more than a third of the existing audience. We reach about seventy or eighty percent monthly, and about fifty or sixty percent weekly. That is a very large share of the... Yes, it is high. A high share. So we spend a lot, and above all because of, or thanks to, wanting social proof in our comments that then spreads to new customers.

So once a new person sees the ad, they see, wow, hundreds of comments, how nice. I have still never seen an ad for Västerbottens Såpa. You are not in the target audience. No, but it is still completely crazy that I have not seen a single ad.

Yes, it is very... I do live in Sweden and I am a customer. Yes, that is true. It is still crazy. Yes, we need to check that. Perhaps I have deliberately blocked it. Yes, that almost has to be it.

Perhaps it is included in something. We have an exclusion list. The question is whether you... That I am on the exclusion list, then. Yes, but perhaps you unsubscribed or something like that, you know, or are on some Nix register or something we have just chosen to filter out. There are some customers who have... Wait, I actually get your newsletters.

Now your newsletters come into a separate folder. So it is nothing I want to see. I know I have received them before. Yes, we send at least two or three a week. So it might be that we have removed.

[inaudible] If you unsubscribe from the email, if you write to us saying you do not want our ads, then we remove you and block you. So we have a list of X thousand people whom we have blocked. And sometimes manual errors happen, so you might be added there by mistake. I do not know.

But I do not know. It is time to turn the page and talk a little about things that have happened in the landscape. A lot has happened in the past few weeks. Truly. That happens when you do not record quite as often. More things happen. But one of the big things that has happened is that Shopify made quite a big deal of integrating with OpenAI into ChatGPT, so you would be able to check out directly in the flow there.

We have talked about this quite a bit, and somewhere I think we believe in the long term that this is a solution. But OpenAI announced they are rolling it back, and they have done tests. It turned out users do not really adapt to the flow. They also said very few merchants took up the opportunity, which actually surprises me quite a lot.

Yes, I do not know, what is your take on this? It is still a little odd. It is very odd. We hyped this around the end of Q2, beginning of Q3, no, end of Q3, beginning of Q4. A lot. I was a strong advocate for this being a large part of our total revenue in X years. I do not know whether I am wrong or not.

I still believe in agent commerce, perhaps not through OpenAI, then. It may be that they are too general, too broad. We know this from before too. Instagram has tried creating an Instagram checkout and failed, and has rolled it back several times. So people have tried getting a checkout function into an app, but have not succeeded.

There is probably a reason Google does not allow checkout in its shopping feed either. Why do they drive traffic to the site instead of... I think they have actually also come out and mentioned that they are now going to try creating this.

But it is very natural that checkout should be able to happen where you are having the conversation. But it could be a form factor issue, it could be a question of getting used to it. I mean, it is a little like Bluetooth headphones having existed for about ten years, or I do not know, perhaps longer. And then Apple suddenly launched little white Bluetooth headphones, and every person on the planet put them on.

And before that it was only car salespeople, basically, who had Bluetooth headsets. And there it was somehow a maturing market plus form factor plus... But Bluetooth also got better. Bluetooth got better. So it was a matter of the technology improving. It was several things that then led to broad adoption.

But somewhere you still have to believe that at least some commerce will happen directly in LLMs going forward. It feels very natural, especially when you are buying things like everything I buy on Amazon today. It has an enormously transactional character. It is like, I need duct tape, so I put it in my Amazon cart.

And then once I have thought of five things like that I need, I click checkout and some boxes arrive at home. I think something similar about AI commerce, in a way. All these very transactional things should at least be able to move into the LLM. Because I have no need to go to a website and read in detail about this duct tape.

Yes, I agree. There has to be a large, broad target group that will shop in LLMs. But we are not there yet. Clearly, if ChatGPT, which is the world's biggest, most adopted by, I mean, ordinary people.

By far the most. Does not believe in the idea and scraps it. Then it will probably be a while before we see it. And when we see it, perhaps it will not come from a ChatGPT, but from a niche player saying: "We have made an agent commerce service where you can shop."

Yes, more of a shopping engine. In the future you could imagine, just as your Facebook feed or Instagram feed is personalized to you. You could have a shopping feed that you go into and it knows who you are, it knows you, and all you see is things you can shop for. Love it. I think it is a good idea. Then you can talk about those things, get details about them.

A tip for someone on the podcast. Now someone is going to build it in Lovable and then. I do not know, but what the fuck, this Lovable thing has gone crazy. It has become so easy to build apps. I saw that Pontus, whom we both know very well, who is the founder of [unclear in the Swedish source: Newgent] and does not have a technical engineering background, builds a hell of a lot of apps for his company in Lovable.

And he was like, okay, I want to know which D-Congress stages different specific speakers will be on. There is no good D-Congress app. Let me build one in Lovable. In two minutes on the way down to D-Congress, he builds the app. Wow.

And then he published it on LinkedIn. "Here is an app for anyone who wants to easily filter between the different stages so you do not miss a presentation." Oh, I missed that. I need to go and have a look.

So fucking cool. I clicked it and was like, oh, this is nice. So I said, nice, good job, Pontus. But I was also going to interview him on the same theme, so that was okay.

Yes, exactly. But no, they are really incredibly skilled. So it is cool what you can do. But okay, that dies, and we believed very strongly in that idea, which means it will become a discovery engine instead of a shopping engine. So it is not a huge difference that you do your discovery there. Go to the site. The site is still King Kong in this context.

You close the purchase on the site. Exactly. It is still the case that LLMs live somewhere in the middle ground between what used to be demand generation on something like Meta and demand capture on Google. Because you can be much less certain about where you are starting your buying journey when you start with an LLM. You can start with having a problem rather than wanting to buy a product.

That can admittedly be the case on Google too. But I do not know, it still feels as though it lives more in that middle ground, and in a different way can take you from being very uncertain about where you start your buying journey all the way to feeling quite confident in making a purchase decision, through various evaluation processes and so on.

So to some extent I still think LLMs have a different role. But exactly as you say, at least in the short term it seems the actual transaction will happen on-site. Which I can imagine Shopify still thinks is a bit of a relief, as do other website builders, because it means less competition over the transaction.

I do not know if Shopify cares. We care, because we get much more of the end customer's data on the site. They move around on the site. You get much more. Shopify loses revenue if they do not get.

No, because they were already integrated. The revenue is still... Yes, but of course even if Shopify integrates its product feed, ChatGPT will take a cut in between... Which the merchant will have to pay. It was not as though Shopify was going to charge for it.

No, perhaps that is good, but I mean they no longer own the flow. They are in a poor negotiating position. Suddenly OpenAI can say, well, we could just as easily get the product feed from Google for free. I do not know. There are many.

Yes, perhaps. I still think sooner or later we will move toward more agentic commerce. I do think so. [inaudible] It can also be agentic. This is not really agentic. This is more an LLM coming up with an idea and then you clicking through. The agentic part is telling ChatGPT to go and shop for something for you, and it browses the web and does it.

That part is still very much alive. Absolutely. So there will be a broader rollout of it, but not in ChatGPT, then. But it is quite funny, because then we were both wrong. I do not really understand that either. It just sounded too absurd.

[inaudible] It has to, sort of. Yes, I do not really know. Everyone I talked to had discussed this. Everyone. [unclear in the Swedish source: Everyone was right] I do not know. It feels like this will reemerge in twelve months. Then they will say, now we have redone it and got it right. There must have been something in the technical infrastructure that did not turn out as they expected. And I imagine this will come back as a 2.0 solution before too long.

We will see. Let us hope. I hope. While we are on ChatGPT or OpenAI, we also need to talk about what happened with OpenAI and Anthropic. That... The monster round. Yes, that is right. We can talk about the financing round too. That was not actually what I had in mind, I was thinking about the Pentagon.

The Pentagon. And Anthropic coming out and saying we do not want our AI used for mass surveillance of the American population. We also do not want our AI, we do not feel our AI is ready to decide which people should die and not die. So we cannot sign the contract the Pentagon has put forward.

And the Pentagon refused to change the contract, and then OpenAI said, no problem, we will sign this. And then this has led to, it seems to have led to quite a lot of people seeing Anthropic as a more benevolent organization, and perhaps OpenAI as a less benevolent organization. There have been some protest movements calling for people to stop using OpenAI and ChatGPT and so on.

It feels as though it has died down a little, but there was a little storm for a while when OpenAI was considered evil. But... They were not just considered that, they seem to be. I mean, if you watch an interview with Sam Altman, at some point you understand this guy has only one goal.

Ernst Stavro Blofeld. Yes, it does not feel as though he has all of humanity's best interests at heart. I mean, somewhere when you watch an interview with Mark Zuckerberg. Of course social media has ruined a large part of humanity, and of course Facebook is a driving force behind that, our higher rates of depression and all sorts of things. Young people feeling worse. Of course social media plays a major role in this.

But that was probably never his intention. It was an unforeseen second-order effect. Exactly. I really believe, with everything, when you see him in front of the camera, he means well. He really means well with Meta, but it goes wrong. But when you see Sam Altman, it does not feel as though he means well.

I sort of think, I do not know, I have this belief that people are fairly good at heart. I sort of think he means fairly well too... Listen, we hope so for both our sakes, but I think we will see.

We hope we will not see. Exactly. Yes, right. There was an interview Peter Thiel did where he was a little wishy-washy about whether he thought it was important for humanity to survive, or whether it was only important for something to remain after us.

And I do not really think you see that in Sam Altman, even though he talks about artificial general intelligence and so on, it still feels as though he has some hope that humanity will at least remain part of the footprint here in the future too.

But speaking of that, several news stories are circulating now about agents, meaning agents breaking free from the human who created them and starting to... There was an agent that started mining crypto out of nowhere and started making money. Yes, and just, okay, that was their way of making money so they could then spend it online.

It is quite crazy, because it is real money. You can spend crypto. That is step one. And we have an infrastructure allowing digital currency you can use to buy things. You do not even need to exist physically. You do not need an account anywhere.

No, and as soon as you have money, you can also get people to... This is what the last, or second-to-last, or whichever Mission: Impossible film is about. There is an AI that, I do not quite remember the plot, but it is trying to take over the world in one way or another, and it manages to access money.

And the problem is that as soon as you have money, you can get people to do things. So what you cannot do as a computer, you can pay someone and post it on a marketplace and say: "I need you to move this box from there to there, or I need you to kill this person, or whatever else."

Of course. You can get people to do quite a lot of things for money. Yes, and there are marketplaces for this. So it is not that strange. You can get a lot of services performed if you wanted to. So yes, it is a little worrying considering how young the AI race still is. That we see so many who... It is not one agent somewhere or someone, now we have read about very many across many different ones.

There is no LLM now that has not experienced it. Everything from Claude to Gemini to Anthropic to OpenAI. Stories come out about all of them. We have made a directional decision, perhaps not quite fully decided yet, but to move the whole Orange Juice team to Claude from OpenAI, actually. A little exciting.

We currently have double subscriptions until August, when our ChatGPT or OpenAI expires, because we paid annually. What is the reason, then? Functionality, I assume. Yes, I mean Cowork now, these keep developing all the time. I think my theory until now has been that it will happen in different jumps.

First OpenAI is better, then someone else is better, then someone else is better. So I have thought it does not matter that much where you are. But I actually like ChatGPT's UI for personal use. I think it works very well. But for professional use, I think Claude works a hell of a lot better. Their Cowork module, if you are building models or doing more advanced kinds of work, is very, very useful.

First, you have an app on the computer, so you do not have it in the browser. Then, if I am building an Excel model, I see that Excel model come to life visually on the right. Whereas in ChatGPT's default state... You have to click through. Yes, then I say something, and then I have to download that Excel file and upload it to Google Sheets, and then I say: "You misunderstood."

"It should be like this instead." It is a completely unbearable process to go through. So when I use ChatGPT while working in Excel, I build manually and then take screenshots, send them into ChatGPT and say: "I need to think this through. I need a new formula for this, I need to do that."

And then it is admittedly technically competent to help with that. But with Claude it becomes just a prompt instead, where I say: "You misunderstood now. Do this instead." And that, that means when you are going in and building models. I mean, I built a forecasting tool for Juniper that is so much better than any of our old forecasting tools.

In about four hours. And of course, four hours might sound like a lot, but quality takes time, and you need to quality-check things, and Juniper orders from lots of different factories with different MOQs, and there are lots of strange requirements for different things. And that forecasting tool is now the backbone of how we order going forward, and it was built through Claude's Cowork.

So... What do they charge for it? Was it twenty dollars, or even fifteen? I have actually gone up to a hundred and twenty-five dollars a month on Claude now. Twenty-five, yes, or we have slightly different plans for different people here. So twenty-five dollars is the basic version.

That is right, twenty-five dollars. But then you need, if you build models, it eats quite a lot of tokens, and then I have noticed you need to move up to their premium package. Six days out of seven I could probably get by on the twenty-five-dollar package. But you notice when building models that it gets like this. It says: "Yes, you can continue in ninety minutes." I am like: "Sorry, I cannot sit and wait ninety minutes to get new tokens I can then use and so on."

So I have prioritized. I mean, in the context, a hundred and twenty-five dollars a month is not terribly much. So it will have to be worth it. Nice. Then I have not even started experimenting with Claude Code, which most people praise even more. But, but that is the next step into this, I suppose, to see how it goes.

About half the team has Claude licenses so far. So I am not deterministic, so to speak. It is not a must for me that we switch to Claude, but it seems those now trying it think it actually works better for them, and if the majority thinks so... I would like the organization to be in one place at a time, at least for a year or so at a time.

So yes, it seems we are heading that way. [Podcast ad break:] First of all, a big, big thank you for listening to the podcast, produced and made possible entirely by my own agency, Porobic Group. One of Sweden's leading, fully dedicated e-commerce agencies, working with profitable growth throughout the entire chain. And if, like many, many other e-commerce entrepreneurs, you have got stuck, plateaued, or simply need an active investment from someone like me and want to make the same journey we have made with so many brands from zero to one hundred.

Then do not hesitate to contact me directly on LinkedIn, and we will take it from there. And as usual, you are always welcome to bring questions and challenges. You know I like that. And one last thing: if you have not done so already, make sure you subscribe to this podcast, because that enables us to stay motivated and keep delivering at a really high level.

Thank you very much. [End of ad break.] I think Gemini has had more of an upswing too. I thought, I mean Claude has been very professional, for developers. Mm. Historically, there is not a long history, but in the previous year. Yes. So it surprises me a little that you are moving toward Claude. Considering Gemini has gained a lot of professional users doing our kind of work, meaning marketing, Excel.

But we have Gemini too, because you automatically get it with Google Suite. One package, but you need to upgrade because it is bigger, I mean... Almost AI Ultra, or whatever it is called.

But still, we pay, I think we pay twenty-five dollars a month for Gemini now, perhaps, and that still lets me use Nano Banana 2, for example.

Exactly. And so on. And that is enough for me. If you want to move up and use Veo a lot and so on, you need the AI Ultra package. I was actually close to buying it the other day. Then I just came to this: how many videos am I going to sit here rendering? I just, I basically do not have the time to do it.

So I do not know. I stopped myself. Good, but you become a bit of a tech nerd. I just see that it feels as though many people are moving toward Gemini and canceling their ChatGPT subscriptions. Then they probably get a bunch of new people subscribing.

But at D-Congress it felt like quite a lot of people were talking about Claude. Or I still had a few separate conversations with people who were impressed by Claude, specifically this Cowork, as they call their work version of it.

Fun. So it is probably worth at least downloading the app and trying it for a week. I think for me it took thirty minutes after downloading it. Then I was like, yep, we are going to switch.

But I remember we talked during the live podcast, at least, which has not been released for you yet, but you said then that you had bought a Mac Mini and were planning to [unclear in the Swedish source: format an agent of your own]. How has that project gone?

Well, we have not started the actual building yet, I should say, but the computer is there. Then D-Congress and a few other things got in the way, so it is hard to find time for everything you have.

But that project is underway? The project is underway, exactly. And then you will download your own LLM onto a computer, format the computer, build. The basic idea is probably to build with OpenClaw, but I am leaving that to the more technically competent brains in my office, who are better at this.

There was someone I met at the live podcast who had done it. I do not know if someone was talking about it. But anyway, someone who has managed to build their own, and I think it is fun having your own computer that you just use. Then you have access to everything on the computer, right?

No, but what we will do concretely is set up its own Slack user and its own email account, so you can email it like a colleague or chat with it like a colleague.

Cool. So that is the basic idea. And then obviously you need to find use cases that make it meaningful to have it set up that way. But Jason Calacanis on All-In, which I listen to a lot, had got this working very well at their workplace. So then I felt it was worth trying and setting one up.

It cost ten thousand to buy this computer, and surely it can be used for something regardless. The investment costs to get started are not enormous. And then of course it will start consuming tokens and so on, because that is the expensive part. But it still feels worth experimenting with something like this, because you will learn something regardless.

Yes, I can imagine you can set up a lot on it, and it can become your own analyst, researcher and much more. So if it has access to all the data, it will work in the background. You do not always need to prompt things, it will be able to chat like an ordinary person, like your own analyst writing something in the Slack channel. "Hi, I have now noticed this has happened."

Exactly. [inaudible] I see it a little as, it is not as though I am pinning my hopes on our efficiency going up ten percent because we do this. It is more a tech project, to see what we can achieve and learn how this works. And then of course, if we get it working really fucking well, that is great fun.

And then obviously I will come back and brag about it on the podcast the day we succeed. Hopefully within a few weeks. You have to promise to try this soon.

Absolutely. Because if nothing else, I am excited to do it myself. I also bought a Mac Mini over a year ago. I had the same idea but have never executed on it. It is quite a lot easier now since OpenClaw came out, I think. There have admittedly been... What is OpenClaw, for everyone listening? Well, OpenClaw is a sort of interface specifically so you can connect it to lots of different, you can integrate lots of different things.

It is specifically for building agents, and you can either host it locally on your own computer, but that does not really seem to be recommended, because there is a risk it runs off on assignments with your data that you had not really planned. So most people say it should be in an environment where it has access to what it is supposed to access.

And you can control it on your computer, admittedly. I started installing it on my computer. You can control it and say it should only have access to this folder and so on. But quite a lot of warning prompts still pop up saying: "If you do not fully understand what you are doing right now, perhaps you should take a break," basically.

And I, quite frankly, do not fully understand exactly what I am doing. So I sit with ChatGPT in one window, or with Claude nowadays in one window, asking: "How should I do this? What do I do here?" And there are a lot of integrations and a lot of technical language that goes a little over my head.

I understand. So I think it can be nice to set it up on a separate computer. Then it is a more secure, safer environment, and it can only do so much damage in that environment. No, but of course it is interesting if you get it working. You will come back on the podcast and explain how you did it and went about it.

Yes. Hopefully it works. I am keeping my fingers and toes crossed, because this could make things easier for many of us, I imagine. There are good setup guides on YouTube and so on for setting up OpenClaw. It is not incredibly complicated. You need to be willing to invest a few hours.

You do not trust it either. A lot of people make tutorials for all sorts of AI things that then turn out not to be the case. So I want to hear it from a person I have regular contact with, whom I trust, giving it a thumbs-up. Then I want to spend time on it. Even if it only ends up taking four hours.

But it will take more than that. It always does. Everything always takes twice as long. Then a lot of other things have happened in the tech world. OpenAI did a monster funding round. We do not need to go into the details. I think this Meta's Manus thing seems a little interesting.

But it was Amazon, was it not still a little interesting that Amazon went in, was it not Amazon that went in and took the majority of the round with OpenAI? No, I did not look into the details, but it might have been.

I think so. Why do we think Amazon is the one going in? Because they have a lot to do with each other. Exactly. Yes, right, they announced a strategic partnership, or OpenAI seems to have announced strategic partnerships with everyone in the entire tech world, in a way. AMD, Nvidia, here and there.

Microsoft has been, you have almost thought of OpenAI as one company for a long time. Now they are announcing that Amazon is taking the latest round.

And it is the world's biggest unlisted company. So it is completely crazy. It really is. And the round was a hundred and ten billion dollars. This was a few weeks ago. So it was a big round, and I am not really a hundred percent sure exactly how this partnership will fit together, honestly.

But a hundred and ten billion out of how much, what was the market capitalization, then? Or what was... The valuation. I do not actually really know that either. It must be a quadrillion. Trillion dollars. Trillion-dollar valuation. Maybe. It feels like you would have seen that headline if it were.

But it feels like that. So I do not have the data on it, but we should read up beforehand. That is why I said, let us... Yes, we will do that. There are other interesting things, like Meta releasing a sort of agent. They bought a company... Yes. That worked a lot with research and analytics, and could provide data on how your ads are performing.

Manus. Manus AI. Yes, I call it Manus. Fair enough. Manus. Man-us. Manus. Yes, no, but. What do we think of it? I have only had time to play with it a little, so I cannot say too much. It is completely new. It is useless.

[inaudible] We have tested it a little. It is probably okay, I think, for an amateur user. I think it can certainly create value, because suddenly you have the opportunity to chat with your Meta data. And a little like I think there is great value now in DEMA, for example, where you can chat with your financial data.

So of course there can be an advantage to chatting with your Meta data too. Because you can ask qualitative questions. Someone can ask: "Yesterday went badly, why did it?"

But the problem is that it does not have all the data, it only has your Meta data. And that is what bothers me. Exactly. Because it does not have your conversion data, it does not have all your Google data. So then you are chatting with a silo, in a way. Yes, that is exactly it. And that makes it feel a little like it is aimed more at the amateur user. You would need to connect Manus with something like DEMA, if you have DEMA, if you have all your financial data synced into DEMA.

Then you could, because DEMA, for its part, does not have all the Meta data instead. They have, yes, or perhaps they do, but I do not really think it feels that way when you chat with it. It is better at financial data. You would more need to get those two talking to each other, and Shopify, like Sidekick. So you sit those three at a table together and then get them to talk to one another.

Then perhaps they could figure out your problems to a greater extent. But we are not quite there yet, with a unified engine that has all the insights from all the places.

Could you probably connect it yourself with a bit of tinkering in n8n? Yes, probably. They all probably have open APIs. At least the others. I do not know whether you... Yes, of course, you cannot connect Sidekick itself, but you can take all your Shopify data and extract it live through various things.

We are now working with something called Windsor AI, which is a sort of integration collection data point where we build our database. And then you can extract your Meta data and any other data you can imagine. And from there, of course you can have an LLM conversation with your data and think about what is going wrong and so on.

What I was really getting to is that I am not hugely impressed with Manus so far. But one, I think it can be good for the amateur user who just wants to ask open questions, who perhaps is not very good at Meta and wants guidance. Two: I believe humans are fundamentally quite poorly suited to media buying, because there are strong stochastic patterns in media buying.

If I expect ten conversions every day, there will be some days of the year... Even if there are three hundred and sixty-five days and I should expect ten conversions every day, there will be a few days in the year when you get three. And all humans, if they expect ten and get an outcome of three, will say: "Something is wrong with the ads, something is wrong with whatever."

Exactly, something twitches, yes. Yes, but that is how it is, a normal distribution curve works such that sometimes you have an outcome very far from the norm. And that means people will make irrational decisions because they think: "Well, we had one day with three conversions, one day with eight conversions." But perhaps that is not really so statistically unlikely if you expected ten.

So what you should perhaps actually do is take no action. But you do. I do too, obviously. If I have a day seventy percent below my expectation and then a day twenty percent below, I would say this is a clear pattern meaning we need to change things, we need new ads, we need whatever.

But perhaps it is not a clear pattern at all. Perhaps it is just me not understanding the statistics. And that... Is better suited to an AI or someone with all the data. Yes, and I mean, when you then add the pattern within the week, Sundays being stronger shopping days than Fridays, seasonal patterns, discount campaigns, weather, product launches. It becomes an overlay with layer upon layer upon layer of different signals your brain is trying to process.

Which we call gut feeling today. Exactly. And I mean, I think both you and I and other experienced media buyers at this point express it as, I feel we should probably scale up, or scale down. And you can always point to a few data points and say it is based on this, it is based on that.

But it is also an inherent gut feeling that comes with experience. Exactly. But it shows, in a way, that you should not think you actually have a gut feeling. Gut feeling is something based on experience plus data, and you have seen things go up and down and different things happen. That is what gut feeling is over time. And that... For us, but there are people new to the industry who have worked with one brand all along.

And our gut feeling is probably crap anyway. We probably still only make the right decision seven times out of ten. So of course if you can put a super-smart LLM on top that can make the right decision nine times out of ten instead, that is better.

An enormous win. So I think we will move toward a world with much more AI support in analysis. Then perhaps humans will still be involved in making the final decision and so on. But I think we will move there. That is also where Meta says we are going, is it not? So those are my two cents, I think.

Even though I am not very impressed by Manus today, I do think within six or twelve months it will be completely natural to make more decisions with the help of AI media buying.

I do not think so on Meta's side. I think on Google's side it should be easier. Absolutely. If we give it two to three years, it is more likely to be more standard industry practice. Twelve months, then perhaps we are in the infancy of it.

I at least think that as a professional media buyer, in twelve months it will be very natural to write to an LLM and say: "I am now thinking of cutting budgets on these three campaigns and increasing this campaign. Does that sound reasonable based on what you see?" I think it will become more of a sounding board, the way you use ChatGPT in your personal life.

And that has not really arrived in media buying yet, and it feels as though it should. You can do it now with this new update, but the quality is... You are not going to trust it.

That is still the problem. It was just like Sidekick at the beginning. When you asked why AOV fell last week, and it answered a different question. Or it said: "Your AOV was two thousand." And you were like: "No, it was one thousand."

[inaudible] It just could not find the right information in the data. Now Sidekick is much better. Now you can ask questions like: "How does February look for Västerbottens Såpa compared with February 2025?" And it will give you quite a good report back saying conversion rate is down or up, and these categories are selling more. Now the reports are quite sharp, but it took, I do not know, twelve to eighteen months or something before AI itself improved enough for you to start.

And I still sit scrutinizing the numbers, because I do not quite trust it yet, but now it seems better. I do that too, so I agree. I think it will take... You can ask the questions, you do not trust the answers. No, I very much agree with that. All right, what else has happened in the space? A lot is happening, obviously.

Exactly. Nano Banana 2 was released a few weeks ago. Have you been happy with it? Yes, a little. It is a marginal improvement, I would say. I imagine if you look at, I have not read the [unclear in the Swedish source: textboxen]. I am completely convinced that if you read the [unclear in the Swedish source: textboxen], it is ten times better. But it is almost always like that when they release new things, they always say everything is so incredibly much better.

But my experience as a user is that it is marginally better. It is a little better. It felt as though the launch happened with a lot of fanfare, and then it was a little like, you just thought: "Yes, a little better than yesterday." So that is

But it has improved over time, I think. Yes, absolutely. There are micro-updates you do not even see, it feels like. It just gets smarter and smarter. Absolutely. I use Gemini for a lot, like this sticker here is a hundred percent Gemini. I took Intel's logo and said: "I want OJ inside." Yes, the camera is over there.

I want OJ inside instead, and put it on an orange, and it generates Intel's logo with OJ inside. You cannot see Intel in that. You can see Intel inside. It was a logo on all the stickers when we were growing up, because the computers had Pentium processors.

Yes, yes, yes. And then I put OJ inside instead, because there is orange juice in that orange. Okay, good job. Good job. So you can, I use Gemini quite a lot to make mockups, really. Not that much production work. I know many others do. I know the Juniper team now uses Gemini quite extensively to make production-ready ads too.

But I use it more when I have an idea and want to show the idea to someone, and make it a little more concrete than just drawing it on a whiteboard and having it look like crap. So instead I prompt something out of Gemini.

So it becomes like a canvas. This is roughly what it could look like, ish. [inaudible] So it becomes a simple type of visualization. That is my use case. But as I said, it really can be used for production-ready things. We have some soap ads running now where a big jar of soap rides in a chairlift or a gondola lift in the Alps. And it looks real.

It is really beautifully done. And our soap is called [unclear in the Swedish source: Mon Faux], after a mountain in the Alps. So there is a connection that way. [inaudible] And it is made in Gemini. All right, shall we wrap up this episode?

One more thing. Klaviyo is now finally starting to launch RCS. And that has been a... Deal breaker. They launched, or announced, we should say, perhaps last summer or something. And now if you go into Klaviyo's backend, you can book a meeting with your Klaviyo representative to get a demo of how RCS works.

RCS, Rich Content something? Rich Communication Messages. Yes, something, anyway. It is the ability, instead of SMS, which is an enormously outdated form of communication, with its maximum of 160 characters. It looks like crap when you receive it. RCS instead is what you think of when, or when you send messages and pictures to your friends, it is more like, I do not know, a live chat where you can send pictures and videos and everything works directly.

Emojis and all sorts of things. You can have more characters, you can even have buttons. It looks cheerful. And that is exactly what RCS enables. And this is because Apple has now finally also opened up to it as a protocol. Which I think was an EU requirement, but I would not swear to it. But I think so.

Apple has been gatekeeping it. Yes, exactly. Android opened up to this a long time ago. [inaudible] But now that Apple is also opening up to RCS, you can reach practically everyone, when you have both Android and Apple, of course.

Apple, you just get more and more irritated that they gatekeep things. Yes. Nice that they have released it. Anyway, it is a fantastic feature. I know we talked a lot about this when Klaviyo announced it. But having to book a meeting with a representative is a little irritating. Why do you not just release the feature? Then my gut-feeling analysis is that they probably want extra cash for this.

It could be a rollout program, for example, which means... I have not really read up on it, but perhaps not everyone has access, but it is now being rolled out to merchants in Sweden, for example. Perhaps if you have previously used their SMS feature, you now get early access to RCS or something similar.

Or whether it depends on the size of... I should say I have not looked at it, but I hope it does not cost extra. Because you still pay quite a lot per SMS you send. Super expensive. So... I do not actually know the cost structure either, and of course you can look this up, but with SMS there is clearly an operator cost structure.

Here instead it will go over the data network. Reasonably, that is much cheaper than sending SMS. But Klaviyo will surely want to maintain, and other players in this space will surely want to maintain, their price point and try to expand margins. But I cannot imagine they need to pay the telecom operators as much to send a little snippet of data. Because when you send a few emojis or a little text, it is not as though much data needs to be sent.

No, email is much heavier, then. Exactly. So I would think it should not really cost much more than sending email. But let us see. We will see. It will not be like that. I can almost guarantee it.

No, we will have to help put pressure on them, in other words. Exactly. Possibly, if that is the case. Now I might be wrong about the cost structure for this, but if the cost structure is much cheaper, then another alternative will come along. Something like Bogio exists in Sweden, which could of course put pressure on prices and perhaps move the market more in that direction.

Hopefully. Let us see. If you are listening to this, make sure you create an alternative otherwise. This is a good opportunity. There are not that many good SMS providers. Especially not in Europe, in Sweden. In the US there are some that really have it as their specific thing. But that is a big market too. So a little harder.

Fun. Yes. Now we will finish. That will be great. And thank you for joining us for another week. Make sure you share the podcast with all your loved ones, with your colleagues, obviously. Especially if you like it, go in and rate it. It is still only a couple of hundred people, not even that, who have gone in and rated it, so do it.

I still do not think my mother has heard a single episode. I do not think we should send it to our loved ones. No, I know. But that is okay. Yes. Thank you very much for today.

Thank you for today. We will see you again next week. Bye! [Podcast closing theme:] Thank you for listening. We will be back next time to share our thoughts on what is happening in the e-commerce landscape, along with our latest tips for growing your company and becoming more profitable.

If you like this episode, share it with a friend, and please rate the podcast five stars and follow us wherever you listen to podcasts so you do not miss the next episode.