From the frontier Lessons from Klarna's ambitious AI rollout

50.

100% maybe. I hope it's 100.

I had to pay a lot for saying that. Yes. That was a very expensive thing I said. Yeah.

No, but people were very angry with me for saying that.

I mean, I think media has a tendency to simplify the story a little bit. First and foremost, what happened in our case was that we were not the best customer service, meaning that the most simple questions that normal companies would have dealt with in an FAQ or like press one or press two in an IVR system, we had humans deal with them. So it was easier for us to automate and give this to AI because the question would be like, 'Did I pay?' 'Yes, you paid.' 'Okay, good.' So it wasn't the most sophisticated errands. But the benefit of this was that when I first experienced ChatGPT in 2022, I was just like, 'Wow, this technology is going to change everything.' And I remember I was like, 'Jesus, I need to know who's doing this.' And then I found out there's this guy called Sam Altman building this stuff. And so I called Sequoia, which is a shareholder of ours as well, and said, 'Can you make an intro? Because soon this guy will not make meetings anymore with somebody like me. He will mostly talk to governments and stuff.' So I jumped on this plane to San Francisco and I had a two-hour session with Sam booked, and by the time I arrived it was only 30 minutes, which already was telling me like, okay, this is going fast.

Yeah. I got in there and I talked to him very quickly and I was trying to convince him. And then funny enough, he said after that, he was like, 'So what are you doing in San Francisco?' I was like, 'I kind of come here for you.' So he's like, 'Do you have any dinner plans?' And I was like, 'No.' So anyways, but the point is I told him I want us to be your guinea pig. I want us to be the ones that try everything that you do, all the new stuff. And we started doing that with OpenAI, which was very helpful. Today I actually personally work more with Anthropic, but we did apply that in customer service and very early on saw, and it was a scary moment to be honest, because we kind of built the first AI-powered customer service at that point in time. And the scary thing was when we pushed a button and literally 600 agents were out of jobs.

And that was scary because I've never seen that. I mean, to some degree you've always done this, right? You try to improve your product so less people call you because something's broken, it doesn't work. So there's nothing new that you try to optimize and make your product function better so that you have less people complaints or issues. But seeing a single thing take away 600 agents, that was scary. Now in our case, we mostly worked with outsourcing. So a lot of these would be working agents for big customer service companies. And these agents would go on and work for other clients who still had a lot of errands. So nobody lost their job there and then. Teleperformance, one of the big companies, lost a billion dollars in market cap when we announced it. But nobody lost a job there and then, but it was to me just like, wow, this was a shocking moment to see. And since then we've kind of improved on it. But to be entirely fair, if you try our AI customer service today, I'm not going to be proud about it because what happened internally is we started realizing, wow, if we want to bring this to the next level, it's not going to be enough to build it and do it the way we do it. We're going to have to do it in a totally different way. And since then, we've had a lot of our smartest people work on how do we bring it to the next generation, which I hope we will release this year, but let's see in a few months or something.

Yeah. So that's another thing. So a lot of times people get it, it's funny nowadays because sometimes people use us as an example of AI success and then they use us as an example of failure.

Yeah. It could be both. But no, no, no. There was a few things that we realized and one of them was first and foremost, the big thing when we launched AI, which we had most chat not phone. So the big thing was actually I think all of you experienced, you chat with the customer service representative and they obviously have five other chats going on at the same point of time which makes sense because you may also be busy doing other stuff. So you're going to be a little bit annoyed, you're not going to focus on this and so the agent does. But the problem with that is when you then chat with your agent they are slow in responding because they may be busy with four others at the same amount of time. And the difference with AI was that the medium time for a customer to resolve its errand went from 12 minutes with humans to 2 minutes with AI, which was just that basically the AI responded immediately, right? So it just took shorter time so the quality was higher. Now what we also ended up thinking about was like, okay, but in the end, what happened when you know, whatever we got factories making shoes or furniture or whatever, we start appreciating what is crafted, we start putting a higher price at what's crafted. So we kind of felt like, well, there's a counter trend here as well which means that our AI customer support will be the cheap stuff, it will be the one that everyone gets, but then there will be the VIP one which is like, I got to talk to a human, right? I got somebody who had an emotion and actually talked to me and cared. So we then tried to send, which is very difficult, like media is not very good with two things can be true at the same point of time.

So then we kind of tried to come out to Bloomberg and say look we actually are rethinking this, that we're going to use AI to support humans and also support the simplest errands. But we will also offer premium support. But that also means that our customer service to be honest was more of like, hey, let's worry about cost and make sure those agents deal with a lot of errands very quickly. So the costs are low. And we now have to flip that into these has to be our most passionate, engaged employees that love the Klarna product and support our customers in a way that's VIP, right? So it's also a transformation of who do we hire for customer service, who works there, what kind of salaries do we offer them. See what I mean? It's a whole transformation of the thing. And actually the most exciting project we have currently is we're hiring our own customers. So we're actually onboarding our own customers, the most passionate customers. We're saying to them just like they can go today, some people of that generation will go down and drive an Uber.

Because it's a nice extra gig, right? And we're doing kind of the same. We're saying if you're a Klarna user and you love the product, you can come and work for our customer service for a few hours. And we see people work remotely and the difference is these people, they love Klarna, right? So they know how it works, they know all the, and their customer service is just so different, right? But again, these more nuanced messages are difficult to get out in a Bloomberg article.

Yeah, I'm happy to.

So because we are predominantly, most of our employees were in Europe not in the US, labor laws are different, meaning also that in the US there is a different culture where people simply say okay you're out because you know, and that's not how people do things in Europe and people can have different political opinions about this. I'm not going to get into that debate, but and you're based in Sweden. Yeah, most of our employees, the bigger base, like we have a few, I think half a thousand of people here in the US, but the rest of our now 3,000 people, 2,500 is in Europe. And so that obviously has some implications for this, but the other thing we said is okay, what's going to happen and it's to me very clear AI will take jobs. I'm not going to pretend anything else. I think that's going to happen. So, but rather than pretending to my employees that that's not the case, we said let's be frank and honest about this. But then we asked ourselves, how can we do that so that's not negative, right? Or at least make it less negative for our employees. I'm not saying for society as a whole. This is going to be a major problem, but I'll talk about my company now that I need to deal with. And what we decided to do is we said look we know already that Klarna just like other tech companies have a natural attrition of 20%. So that's just like typical tech company, people stay five years they move on to do something else. So we said if we stop hiring, if we really commit and manage to stop hiring, we will naturally shrink by 20% every year without having to do layoffs. And then but that obviously means that all of our employees needs to adopt these AI tools to drive the efficiencies because otherwise we couldn't operate the business because it's a growing business. You have to remember from 2022 we used to be 6,500 but we are now less than 3,000 but our revenue has grown and exploded since then. So the revenue per employee has gone from $300,000 to $1.3 million in that time frame.

And then what I said to our employees also said, 'Guys, I'm going to be honest. This is what's going to happen. But as a consequence of that, we're going to do a lot of savings because we're going to become less and we're going to do more. And I promise you, and I've talked to a board, we're going to reinvest a big portion of that in acceleration of your compensation.' So the average compensation at Klarna per employee has grown 60% or something. I can't remember the exact number at the same point of time. So my employees know that they're driving efficiencies but they are also participating in getting the benefit of that. It's not just more work. It's also more compensation.

One thing, like I am, again it comes back to the same I said about customer service, relationships are not going to be replaced by AI anytime soon. And for example one of Klarna's, a big part of our employees are today teams that work with our partners, could be like Macy's, Sephora, etc. And I have amazing teams that are distributed across the world. So I have people across 50 offices and they sit in, you know, they are in Seattle and deal with local companies there or they're in Portland dealing with Nike or they're in China dealing with Shein, right? So they're across the world. Those people, AI is not going to touch their jobs. So there are obviously pockets but there are other pockets where it's having very significant implications for the amount of people you need to do something.

Yeah, first the interesting thing is that first we said okay let's not hire at all non-engineers and only hire engineers and that was kind of the first two years but now it's switched actually, it's almost the opposite. Because what we find, and now I'm generalizing slightly, but engineers tend to be very technical but don't understand the business as well. While non-engineers tend to understand the business really well but don't have the capabilities. But now what's happened because of AI and everyone can code, suddenly the business people, like the engineers first they're like okay I coded this feature what do I do next I don't understand the business so I need to go and ask a business person to tell me what the next thing we should do. I'm obviously generalizing, not obviously some engineers are better at this and so forth but generally speaking while the business person knows exactly what they would like that dashboard to look like or that feature to look like and so forth and if they can talk to the computer directly to do that, then suddenly they are creating amazing things very fast. So I think there's a shift right now where I feel that we ended up at 50% engineers and now I think the business people again will take over. But it's more like a business person that actually understands the customer, that loves speaking to the customer, that enjoys that conversation and then thinking about using AI to bring more value to our customers.