Elias Malm: Self-Learning AI Is Already Marketing Brands Fully Autonomously // Självlärande AI marknadsför redan varumärken helt autonomt | En AI till Kaffet #102

Welcome to Malmö. It's been great. We'll do this in English because you have fans and customers all over the world. Fans, I wouldn't say, but yeah, I like it. Listen, tell me about the story of EpiMinds.

I know it went from like a weekend hack to the front cover of Forbes in 12 weeks. Have you had time to process? No, and I think it's been quite a journey. I mean, I think we marked our actually, we have a bit of a funny story.

The product shut down three days ago because our certificate with the domain has expired because it was marked for one year. We didn't renew it. That was the first time Mo got really affected.

Yeah, it's been 12 months since we got the ID. So have you renewed the domain now? Now we have renewed it, so you can't steal it. It's all good. No, but I think it's been 12 months now and it's been quite a journey and like a lot of ups and downs, but it's been pretty fun and like the beginning was absolutely crazy.

I mean, I think I quit Google on the day we got the idea after Mo has hacked these agents over the weekend. I know, you know this, the story. And then 10 weeks later, we had raised two rounds, 6.

6 million. And then we were just going for it. Cool. So basically you were working at Google in Ireland. Yeah. So I was working at, I know Mo for like the last, I think, three years before then.

Yeah. And we've always wanted to work together, but I had Google, he was building his own thing and so on. But I mean, he taught me how to code. We talked every day, like almost every week and we had these type of things.

And then over a weekend, he had, he come to me and said, Hey, I, I built 18 million agents. And I was like, No, you didn't. But, but then he basically, what he's trying to do is he's trying to simulate the Swedish population and Switzerland for some reason.

I still don't know why it was Switzerland. Claude probably got it confused. Exactly. Like everyone thinks it's the same, right? But he's taken the demographic data of these countries.

And I think it was like 300 attributes and then he layered it. So he tried to build like replicate agents with those attributes and then he put them in the simulation. And then he wanted to ask them questions about like politics or natural disasters or whatever to see how they wanted to react.

And then he came to me and said, Oh, by the way, I have these agents, like, what can we do with them? And I was like, this is pretty cool. And I've been like in marketing for the last five years at a time.

I had my agency. I worked three years at Google. And I was like, Do you have any marketers? And he was like, Yeah, 26,000. And I was like, Okay, let's, let's go.

Okay. So we built a population. Um, and then, yeah, 26,000 of them were marketers within that population. And I had this like huge, huge frustration about like how marketing is done.

And I thought that it had to change with AI and that if we can take these agentic systems and enable all companies, you know on the globe to have like world-class marketing, that would be amazing. And yeah, so then we, I think we closed our first like 12 clients in two weeks with the Figma file and then we, we went from there. Amazing.

And, and hit the cover of Forbes. I'll just say that again. Yeah. I think, I think that's like when, when you, before I thought that was a, a big deal and I wouldn't wanna say humble now, but it's, it's not that cool, I would say.

I think the coolest thing is like if you actually build something people like, because like all of this like your publicity and raising capital is maybe like, you know, it, it can be fun to attract talent, but like, it's pretty far from what we wanna do. I would say. Yeah.

I guess it's part of the game though, right? Yeah, absolutely necessary. So most of your customers right now using EpiMinds product Lucy, this kind of multi-agent, like marketing system, or how would you describe Lucy?

Yeah, I think our entire thesis is that we wanna bring in the best AI talent and then we wanna see how far we can push this for marketing. And then based on that, we want to build products. The first one was Lucy and how we identify this, it's a multi-agentic system for marketing.

So basically a team of 25 specialized agents that can run marketing from end to end. So they can analyze data. They can make senior decisions. They can execute on those decisions.

And then something we move into now is self-learning. Like how can we actually surpass the senior level marketeer based on self-learning in a way maybe humans can't do. Is that like recursive self-improvement?

Yeah. So like, let's say like you can make it super, super simple as well. Like all of the agents learn from like how you wanna work and what you communicate and what happens in the industry.

But what we do now is that when we run fully autonomous marketing, any type of change, the agents has to force to do a hypothesis with a confidence score and a follow update. As that means that, oh, I want to raise budgets from this channel to this, or I'm gonna move budget from this channel to this because I think it will influence this bottom metric with X amount. And I wanna follow up on that in seven days, 14 days and 12 days or whatever.

And based on that, it can do deep analysis about why certain things happened and why certain things didn't happen. And then put some things in the short term memory and some things in the long term memory. So that means that it not just can learn from what we see, you know, when you talk to it and what type of humans want to, like a lot of people do that.

They try to pour the human marketeer knowledge into the agents and we do that. But what we're trying to do is that, can't we look at the performance? So can we actually surpass what a human can see?

And so that's how we do it. Wow. So what's your vision then? Yeah, I think like for us, if we wanna be the default in-house team on all companies on earth and work with these amazing agencies in-house teams, and if we can enable all companies to have like, you know, exceptional marketers and also enable exceptional marketeers to be able to do more, we would be really happy, I would say.

How do you like envision a future marketing team? Like what would it look like? I think there is two phases to this. One that I'm not as sure of, but one I'm really sure of.

Okay. So we're talking about a lot of marketing engineering. And what we talk about that is that like, I think the hottest role in marketing and what would be the most fun role to work with is a marketing engineer.

And that's moving from like, Hey, I optimize keywords. I do all of these things to, I use my taste as a human. I use my knowledge about the business to make really good decisions and training large scale of, of, of agent teams.

So that's how we define a marketing engineer. So instead of you managing it kind of, you instead manage a 25 person team that can do all channels and do a lot more in that sense. So that's like short term, we think that people be marketing engineers and we think that will, you know, enable a lot for, for companies.

The long-term thing is that there is this theory. I don't know if you agree, that we'd love to hear your thoughts is that a lot of people, and especially if you work at Google, you get this question like, you won't be replaced by an AI. You won't be replaced by AI.

You will be replaced by human using AI. And I don't know if I believe that. I think it's a bit naive and I think it's a lot of bullshit. So I think like what can happen is that you realize that full AI, so fully AI, AI, like a team with just AI in marketing will surpass someone that is using a human using AI just because of the feedback loops.

So I think that might be a thing in the future, but then I still think that there is parts for humans, like with taste, creativity, and so on that maybe AI can't cover. Yeah. So those are the two phases that we're looking at.

Okay. So obviously you have tons of, you know, you fundraised, have tons of money, but you decided to, and you did your kind of fundraising rounds in the States, no? Yeah.

I mean, we were in the States. Like we were in the 7 to 8 weeks, I think we were in, but we raised from like, we raised from i8space, so it's a global, of course, but we raised for the European office. Yeah.

And we met both the US VCs and European VCs. And so you could have been, you could have kind of based yourself, you know, wherever in the world, but you decided to be based in Stockholm. Yeah.

What was the driver behind that? I think it was like, at first it was either SF, New York or Stockholm, I would say. And I think all of these things with like location, I get the stats.

I really get the stats that like, Hey, if you build in SF, you will have a higher chance of like a likelihood of success. And I don't think we can escape from that. I

Making changes with a lot, right? So I think that was like the first layer, like, how do we save time and how do we enable marketers to make really good decisions? What we're moving into now is like, more like, you know, autonomous.

And what basically that means is that how can we make sure that eight out of 10 conversations you have with agents starts from the agent side? So that means that our agents check your accounts or your marketing, you know, a few times per day and then proactively reaches out to you and say, hey, I think we should do this. This is something we've flagged and so on.

And then the next step where we have a few of our clients, but also our marketing, is fully autonomous. So that means that, hey, you have this goal that we figure out together. You do whatever you want within certain guardrails to be able to get it.

Then you report back to me if you need to break those guardrails, like, hey, I need more budget or I need these type of things. So that's like, you know, three different levels where also we see our clients start. Like they start here and then they go here and then they go fully autonomous.

Very cool. So basically, Epimind is already running fully autonomous. Yeah. For a few also companies in Stockholm, we have done it for a while now, which is pretty cool.

Can you mention any of the names? Yeah, there is one investment, like a fully AI native investment bank, that's called Leon, which is amazing. And we've been running there since November.

And I think we transitioned to fully autonomous like two months ago. Still like small budgets and so on, but it's been really, really good, especially when you give it a lot of feedback in the beginning, then it can run. It's like, I don't check their accounts now and I haven't checked it for two months.

So originally when I met you, it was kind of in this kind of CMO gathering setting and you demoed your, you know, Lucy to like this huge group of CMOs. And it was like incredibly eye-opening. I could kind of see all these marketing leaders from like big Swedish companies being like, shit, this changes the game for real.

This is something truly different. And I'll have to rethink what my marketing teams look like. Like what work do you think there is left to do for a marketer? And how should, you know, what should a CMO today think about?

I think I was speaking yesterday and I think one thing we talked about that was like one of the big points was that you should invest in things that exponentially become better when the underlying models become better. And what I usually said, and I think the CMO dinner as well, I like the CMO session as well, is that like, hey, what I've showed you today, plan that this will be two times better within the next six months. And I think, to be honest, it's been almost two times better now.

It's been like four months. So it was quick. So what I would do with CMO is that like, you just look at everything that you have, like look at what we do and all of these things and say, okay, if this becomes two times better, how should I build my team and what should I invest in?

And that means like skills, tools, people. So what I would say is that I think exceptional marketers will have the best career in their life just because they can do so much more. They can be these marketing engineers and have an exponential impact.

Like, do I believe that you can run, you know, manage marketing for a few hundred millions by one people, like one person in the NES, maybe not attribution and so on, but I think one person can do a lot now. So I would invest in those like exceptional talented people and I would pay an insane premium for them because it's got to be worth it. And how would you describe those insane talents?

Like what are the attributes that these individuals need to hold today? I would say like, I think like, you know, talk about the T-shape engineer and all of these things. I think that should be for marketing as well.

I would say that something that is a big gap that is definitely needed when it comes to marketing and to get these systems to work is of course attribution tracking and deep understanding of that. Like, I think that's like key, right? Because that's still a question that maybe we haven't seen AI be able to solve, like actual marketing attribution.

That's going to be a big thing. Then also like creativity. Like we bench our models on something we call the marketing seniority score. So we do like a study and we base us against all of the underlying models as well, but also actual marketers.

And we, let's see if you take 10 categories, we beat them in eight out of 10, right? Yeah. But the one that we don't do is like creative intuition and taste. The agent's pretty bad, to be honest.

I think they score like 0.81 compared to like a one, which is an ordinary marketer. So these type of things I would also say. Those are the two things I would say like we can't cover today.

And then someone is like really hungry, really intuitive with this system. So really want to drive it forward, I would say. Yeah. Great. So there's a role for human creativity moving forward.

Yeah. And I do think so. And let's say like in a year from now, like, let's say 80% of all performance marketing will be run by agents. Then it's also like then everyone has that tool.

So then it's more about the same thing as that. That's the same thing as everyone has smart bidding in Google ads, then you still have to, you know, take a layer on top of that and do something different and all of these things. I'm also really bullish on like brand, right?

Like, you know, if everyone has access to Hollywood style production, everyone access to really world-class marketing, like who do you buy from? And I think you buy from firstly best products, but also like best brand story. Celebrities.

I'm pretty bullish on like all of that, I would say. Okay. Yeah. Interesting as well to think about, like, who will the future celebrities be? Like, will they be actual humans or will they actually be AI generated ones?

Maybe there needs to be a bit of originality in our celebrities as well. I mean, maybe I could be wrong, but I hope so. Elias, I know a while back, you were invited to the Google Nordics office to meet their marketing team and kind of do a panel discussion.

And would you share kind of the predictions that you had for marketing? And I think like one of the themes coming out of it was like marketing, like how effed are we? Yeah.

I think it was also just general, not just maybe just marketing, but just like, I mean, these people are working in marketing and they're working in sales. They're working across the Google org. And I think like, I mean, we talk a lot about like what will happen and so on.

And I think what we owe people is also to be really, really concrete about what we see happening in the industry. And I feel like I get to talk with these amazing CEOs of agencies in those teams for the biggest brands in the world. And I think what I just think that we owe everyone is just to be super transparent about what we see and not sugarcoat anything.

Like, hey, if this is an area where I think AI will take over a hundred percent, I will say that right. And I had this point in where if I do this prediction, it's like prediction six to 12 months and how fucked are we? And I think that's like, I saw the person who invited me to that to speak at Google started to sweat a bit because I didn't show him the slides before.

And then we talked about all of these predictions and also like, you know, what you should do if you're like, let's say, 25, 35 early in your career. Like what should you, what should you spend time on? Okay, so let's hear your predictions.

I think, I think first of all, I had one that was that the line between software and service will completely blur. And I think that's kind of goes without saying. And what we mean by that is that like, I think Sequoia famously said that the next trillion dollar company will be a software company masquerading as a service company.

And I can kind of agree with that. And especially, let's say, you know, let's say if you take the traditional agency market, I know that they will come up fully AI native marketing teams that you can tap into that performs the same role as an agency, but even more closely within those teams and do that for a fraction of the cost and outperforms. Like how do you do it then?

And that's of course like a really good positioning for us, but it's also like a big thing that I think will happen. Like you won't be able to tell the difference. And then of course you want to work with humans, but you get what I mean.

So I think the line between software service will completely disappear. I think it's one prediction that you should think about. Prediction number two. Prediction number two was the thing we talked about a bit is that talent will consolidate.

I also would say that tools will consolidate as well from like the MarTech space. We've already seen now that a lot of people are killing their tools too. And we take over what they currently do.

But what I would say with talent consolidates is like, let's say we hire now, we hire for G10, we hire for tech at Epiminds. And if you do the math, I would rather pay maybe five

Low-ticket sales, like will you be able to, will an AI be able to sell you a ticket of $100? Yes, probably in the future. But then we go up and go up and go up. And what I would say is the highest ones is like, all right, if you go into a company, like a Fortune 500 company or a really big company, and you're going to negotiate and win a multi-million dollar deal.

How do you then do like the discovery call, like stakeholder management, like really high, high levels sales, that will be a high defensibility. And then I would invest in those specific skills. So like, that's how I would think about it.

And for marketing, that would probably be like, you know, maybe one channel performance marketing will probably get automated within like two months, right? And then you have these different levels and then like maybe the end is like advanced analytics and attribution for the best companies and, you know, cross-level marketing across like these type of things. So like, that's how I usually think about it.

And that's also how we do as a company. We do these layers and then we try to see like where is the defensibility in that. Yeah. Building the tiers, like where is the, you know, where's the human component?

So it's still going to be a really, really important one and focusing there basically. Yeah. And then let's say if you get to a point, let's say to go back to the sales example.

Let's say if you have a point where like software is going to be really easy to build, you know, there's going to be like really high ticket items and things will consolidate. Like, imagine if you are the salesperson that is the absolute best at winning the largest deals. Like that, of course, is going to be insanely, insanely valuable.

And when everything would like, you know, outreach, the meetings, the proposals, all of that is automated, then of course you will also be able to do so much more, I would say. So just thinking about like the whole software layer and we are seeing kind of more and more functionality in Claude, for example, and kind of doing outreach email is, is even quite easy right now with Claude. So what becomes like the role of, you know, a software layer like Appy Minds within this?

Yeah, I think also if you go back to these layers, right, we sat down and then we said like, okay, we believe that analytics will be able to talk with your data. It's going to be probably be done within like, you know, three months. And we see that's already layered that Claude can do really well.

Second layer is like cross-channel execution. Like, can you actually execute in a really nice way across all channels? And that's like, yes, you can somewhat build that in a good way right now.

But like, I think our solution is like six months ahead on that type of layer. But then where we see the defensibility is and where we want to be is that, you know, how do you make the best marketing decisions? And that's also why we work with like these biggest brands in the world, because if we can make 20, 30% better, you know, handling a marketing budget and so on, that makes a huge deal for when we handle half a billion.

Like we handle half a billion in media spend right now per year. That makes a big, big difference. So where we think that we see it is like in the revenue intelligence, making the best absolute marketing decisions and with self-learning.

So let's say if you worked with our product for a year, that means that your agent team has learned every day, multiple times about every single action that has been done across all channels and your CRM, Shopify, and so on, and builds up this insane marketing brain that can make really, really good decisions that outperforms any type of LLM. That's where we feel like the defensibility is. And that's why we get to work with like these big six agencies and these biggest media spenders in the world, because there it makes a huge, huge difference.

Do I believe that we're going to be insanely relevant for someone with one channel and a few thousand bucks in media spend? No, you shouldn't. But that's like where we, we miss it.

So I think it's the data long-term and then short-term is just like, you know, execution is building ahead of the models. In this world where more and more, we see more and more automation, do you see a world where marketing, sales, individuals within this space will become like more and more irrelevant? Like, will we ever, do we need human-driven departments?

I think it's like a really hard question to answer. And I would say there's like two ways and like thoughts. If you look at like super, super logical and like you look at, hey, you know, what are these specific areas that humans will always outperform AI?

I would say there's not that many. And if you look at that, then probably you will end up with something like, oh, you know, no one is having to work anymore. Everything is run by AI, you know, whatever, right?

Um, But then you also have that, like, this is not the first time people have said, oh, we're going to be replaced. There's going to be, you know, just factories or robots or whatever, or the internet is coming. And I think humans always find a way to stay, stay relevant.

Um, so I would say those are like the two schools of thoughts. Um, I would say that the majority of the work that we do now will probably be done really quick, better and faster with, with AI and with, with agents. But then I always think there will be new room for, for other types of roles.

And I always think that, you know, you know, AI will also enable a lot more technology that maybe will create a new type of roles and so on. And we saw that happen with internet. We saw that happen with factories.

We saw that happen with like the industrial revolution, all of these things that created, created more jobs. Um, I also read this thing where like X-rays, they thought that that was the first thing that's going to go away because now AI can do X-rays so, so much quicker. Yeah, with computer vision technology and everything.

Yeah. But it just turned out that it was just because they could do more. They were so more in demand and it was so more profitable. So they just becomes more of these.

I don't know exactly the, the role of it. And maybe that will also happen, right? Um, so I don't know. I think what we have to do is just be extremely honest about like what areas will be covered by AI and will not, and then trying to make the best decisions along that way.

Um, but yeah, I think those are like how I'm thinking about it. But I think marketing will be a space where exceptional marketers will have the best time ever. But of course there will be fewer to be able to cover the same amount of work that they do now in those areas.

But then again, like the market is not limited, right? It's, it's a growing pie and you know, there could be space for more innovation and more types of companies. All of a sudden as well, there's like perhaps like more of a, uh, like it's more democratized, right?

So you don't have to have like, you know, a Coca-Cola budget or a Google budget to do cool stuff. But if you're just, you know, play, play your cards right, uh, then all of a sudden you can stretch much further. Yeah.

And I think that's like really true. I think we have like one agency that are like, I think there are like three people, small agency, and they manage like 150 clients. Um, and that wasn't possible before, right?

And I think that also enables that all of those 150 clients that couldn't have access to this type of marketing will. And maybe that will create also that, you know, new jobs for, for them because they're more profitable and so on. So I, I really agree.

I don't have all of the answers. I think we have to, you know, wait and see and do the best decisions. But yes, 100%, I think that will also happen in that sense.

So in a future world, we might have agents creating content that is kind of digested by other kind of shopping agents or whatever you call it, um, making decisions for their kind of human owner. Um, is that a future that you foresee? Yes, but I think there's like two options here.

I think like, let's say if you're thinking about like agentic buying, like every person or company has an agent that makes the decisions. There's like two options where you can do there. So first of all, if it's just going to be through an LLM or like based on relevance, it's going to be heavily like SEO.

You can call it GEO, you can call it whatever. I think like fundamentally it's not that different from SEO if you look at the basics. Um, so I think that's something you double down on if you want to be, be the first in that type of industry.

Uh, if we get to a point where like robots walk around the streets and get to see ads or browse the web or whatever, use a strawberry and be able to see the, the, the ads, then of course there's different things. And then you have to ask yourself, okay, what do they care about? Do they still care about if it's a celebrity or do they just care about the material and all of these things that are like, so, so on.

So I think that's also a really cool, cool thing. Um, but I would say that

I feel like, in general, people don't have to be that worried. If you're an individual who feel like they can stretch, that they want to learn new things, you're curious, you want to push yourself, then, like, this is, this is the time to be alive. And of course, there are going to be individuals who kind of fall out, like if you've worked in customer service your whole life, and you find it difficult to retrain or think about a new job, then that's going to be a challenge.

Yeah, I think you have to be like, the first step should be like, if you're out of this, like, let's say if you're not in AI and you feel like this is scary and so on, just like, get a foot in. Because then it's also really easy to make this like your own assessment about where things are going, right? And then I will also, like humans will always find a way.

And I think there will be like insane amount of really, really cool findings, pros with all of these things when it comes to medicine, when it comes to freeing up resources, all of that. So I think it's like, I 100% agree with you. Thank you, Elias.

It's been awesome having you here today. If someone listening wants to, like, learn more about Lucy, perhaps demo Lucy, is there a possibility to do that? Yeah, we are waitlist only, but if you sign up to epmime.

com forward slash, you know, request access, or you will get into whatever type of link you want, then you can. And then if you fit our bill where we think that we can provide a lot of value, we will reach out. So yeah, or you can just contact me on LinkedIn.

Fantastic. Thank you so much, Elias. Thank you. It's been great. Huge thank you for joining us in this episode today. Subscribe to us for more exciting episodes moving forward.