Inside Klarna’s Marketing ‘AI Factory,’ Which Could Save $10M In 2024

Sweden-based global payments company Klarna has had the goal of bringing generative AI to all of its functions. The company has moved swiftly, with nearly nine out of 10 Klarna employees saying they used AI in their daily work. CMO David Sandström has embraced generative AI in his department, and says it’s paid off. Overall marketing spend was down 11% in Q4 2023, but the company has increased its marketing output, thanks to AI. Cost savings from generative AI are estimated to be $10 million a year. I talked to Sandström about how AI has changed marketing at Klarna.

This conversation has been edited for length, clarity and continuity. An excerpt ran in Wednesday’s Forbes CMO newsletter.

Klarna as a company has done a lot with embracing AI throughout its operations. For marketing, how did you decide where to use AI and how far to take it?

Sandström: We decided when we first encountered the next generation of AI, we were absolutely blown away. We tested the metaverse, we tested blockchain and bitcoin, and we were never blown away. But this time around, it was just like something clicked to us, and right then and there we decided that we are going to go all in on this technology, because it’s going to be a revolution that we can’t fully fathom yet. I actually articulated already back then that my ambition—my vision, almost—is to become the most advanced AI-powered marketing organization in the world.

So we’ve been at this for 18 months, basically. And what we’ve gone basically from having no idea to having something fully produced reaching the consumers out in media. We broke that down into a couple of steps and said, ‘Hey, we’re going to create AIs that assist us through this entire journey.’ Because at the end of the day, I do think marketing is like a workflow. You start with nothing. And then you need the creation phase. Then you need the distribution phase, and then you need the analytics phase. So we’ve created an AI that helps us with ideation. We have created an AI that helps us with text, we call it the CCT—copy, content and translations. We do image creation. And we connect this entire workflow. So we don’t have this super AI, where we just click a button and everything happens. We’ve really turned marketing into a workflow and created a world-class AI for every step of that journey.

It’s also not that we have the ambition to remove all humans from this workflow. But rather, I want to have a couple or few extremely talented, smart marketeers who are assisted by the best AI in the world to guide them through this workflow. What we’re seeing now is things that two years ago cost a lot of money, took a lot of time and required a lot of people, nowadays require very few people. We do it at an extremely high pace, and we do it at a very low cost. And that’s the vision we have: Really empowering the best marketeers in the world to create at pace and with efficiency.

Klarna CMO David Sandström.

Klarna

Looking at the cost savings in your release, AI has helped Klarna save 25% on external suppliers. What does that mean? Are you using any outside contractors?

I think what AI has done, and will do more, is push humans up the value chain. If I look at a thing like translations, and I’m not saying it’s low value, but it’s basically taking text snippets and translating them into other languages. We’re live in 30+ markets. I think we communicate in 13 languages. We’ve spent tons of money on translations. What we’ve done now, together with two world-class LLMs—both DeepL and OpenAI—we’ve built a model that translates everything we need to translate. It has our personality, it has our tone of voice, it understands our products, it understands context. We’ve actually eliminated all of the translations because it’s so powerful.

When it comes to other things, we haven’t eliminated them fully, but we’ve changed the way we work with them completely. When it comes to translation, image production, text batch production, all of those agencies are more or less gone. I think this is a movement we’re going to see in the industry more and more. We’ve talked for many years about the agency world being reshaped, but I think we’re finally there. I actually just don’t think that we understood that it would be AI doing it. But now we’re actually there, where I do think a lot of agencies will have to completely reinvent themselves.

Are you creating completely new images with AI? Refining existing ones? Coming up with ideas to produce in more traditional ways?

All of the above. It started with the fact that historically, for marketeers and advertisers, the ability to experiment with high quality has been non-existent because it comes with a huge cost. Creating images, set designers, photographers, all of that has been an absolute nightmare when it comes to cost structure. We started out with just saying: ‘Hey, we now have a very powerful tool that enables us to create a lot of ideas with high quality so that we can go into the real production with more certainty.’ But fairly quickly, we understood that AI can actually produce the images fully. We saw upscalers entering the market, and we now have upscalers that are in the same league.

We go all the way from ideation in our own models, [so] the workflow is fully done. In the workflow you and I would go in and say, ‘Hey, we need an image of a blow dryer,’ and it would go through a process: Create a blow dryer, mask that blow dryer, put a themed background on that blow dryer—so if it’s Black Friday, or Mother’s Day. And it would upscale that image so that it can be used fully in production. And then we size for every single touch point that we have.

The interesting thing that we’ve built—I have yet to see anyone build the same thing—is the batch production capabilities that we have. Basically, we do not create one image at a time, but we create 100 images at a time. We basically brief the computer before we leave the office, and when we come back to the office, it has created hundreds, if not thousands, of images that we need, based on what we do. At Klarna, we have this extremely interesting organizational model, where marketeers, engineers and analysts work very closely together.

When creating batch images, is it sort of akin to what would happen at a photoshoot? Is it hundreds of images of a blow dryer, and you pick the right one?

Batch production is basically just my way of saying that instead of having a human being doing every single prompt for every single image, we’ve created almost like what I call the “AI factory.” We feed information into the batch production. The batch production is then trained according to workflow to put out a specific image.

We have a lot of data. We do 3 million transactions a day. We understand what people like, what people want to buy. What we do every day, we take the 30 most-sold categories in each market. This is all automatic. We load that data into the batch production. The batch production then produces images for all of these categories, and the next day people can find these categories with nice images in emails and in our app. We are extremely relevant at scale.

There’s a huge trend going on with weight loss supplements for pets. That was one of the 20 most trending categories in our system: weight loss for pets. So that goes into the system. It happens automatically. So when I come to the office every morning, I have a beautiful image that is on-brand, works in all of our touch points that can be used for promoting: ‘Hey, did you know that you can buy weight loss pills for pets with Klarna?’

Two years ago, someone would have thought about that category, spun up an agency, created an idea, having set designers, photographers, post-production. We are talking about two, three months and loads of money. And after those two, three months and loads of money, you’re still not sure if it’s going to work.

A couple of years ago, the big challenge for marketeers and CMOs, in my opinion, was the choice between quality and quantity. Either you had loads of these categories and were just going all in. You had thousands of images, but all of them were either some sort of iconography or something like that. Or you would have brands that have like five to 10 assets, but they’re of the highest quality. What we can do now is have thousands of assets of the highest quality, and that obviously opens up a lot of opportunities.

What about the people who you used to work with to do this stuff? Do you still have humans doing the actual image production?

I don’t think we’re going to replace humans. I don’t think we have a future within [times] soon where I will press a button and say, ‘Hey, I need a campaign,’ and the AI will produce a campaign. We need smart and very capable marketeers along with the entire workflow. Every single piece in this workflow—say it’s 10 to 16 steps to create a campaign—needs a couple of very capable humans. I don’t have an ambition to replace all humans. I have an ambition to create things at scale with extreme quality.

However, when it comes to our agencies, my feeling has been that a lot of agencies have not fully embraced the power of AI, and that makes it hard for us to collaborate with them. The agencies that we still do collaborate with, the baseline is that they understand AI and that they’ve embraced AI. One concrete example is we work with agencies now on a new big campaign for the U.S. In the pre-AI world, they will obviously write guidelines for tonality and things like that. And I’ve told them that I don’t want that. I want prompts that the AI understands. I want copy prompts. When we have a copy prompt that informs our AI so well that it writes exactly how we want it to write, that is what we want. We still want to employ extremely talented marketeers, so this is not a man versus machine kind of race.

Have you run into some of the issues that everybody has heard about with AI-generated images—like the images don’t make sense, or they’re oddly rendered, or they bring in incorrect branding for what you are highlighting?

That is still a risk. We do a lot of product images. A huge risk is IP rights. Creating images that really look like a Burberry bag, but it doesn’t say Burberry. That’s a risk we’re running. We’ve run into all of those risks. I treat the risks as just an opportunity to tweak the AI even more.

AI has a fantastic ability to assess images. So we’ve built an AI-driven creative director. I call it ‘acid’ because AICD looks like ‘acid’—to which we feed the AI-created images. So we have one workflow that creates images, and then we have the AICD that actually assesses the images, and it does so from a creative perspective, from a legal perspective, and from an efficiency perspective. So, will this ad work? And it does it really well. What we’re seeing is more and more AI assessing AI images, rather than humans doing so, and it is absolutely mind-blowing. We’ve trained the legal assessment on all of the guidelines, all of the IP rights, all of the compliance needed in different markets, and it then makes that assessment very, very well. And the AI is never tired, never has a bad day, all of that, so the consistency in that assessment is fantastic.

Klarna has been saving a significant amount of money on marketing by bringing in AI. What does your marketing headcount look like right now, versus before AI and what are you doing with the cost savings?

Cost savings are great. Investors obviously like that. I like that because I hope that we’re reinvesting those cost savings into media, into initiatives, into campaigns. But to me, that is not the overarching goal. To me, I want to realize a dream that we’ve had at Klarna. We have 150 million consumers. We know that very personalized communication and messaging works three to four times better than general messaging. Is there a world where every [one] of our 150 million consumers can get a very personalized message, both from a text perspective, from an image perspective, and from a timing perspective based on our data? That is where I want to go. I don’t want to do these big general things. I want to make sure that every single consumer gets unique communication, because I think that’s the future of brands: fully personalized, fully unique communications. And that is enabled by AI. You are going to save money along that journey, but I think that vision is way more powerful.

Editor’s Note: We reached out to Klarna’s press team to confirm marketing headcount changes due to AI, and they responded with the following statement: “We’re seeing across our whole business that things that previously took people a lot of time can be done much faster and much shorter with the help of AI. For that reason, we have paused actively recruiting. The majority of open roles are engineering, and others may be backfilling. In the tech industry where on average people move companies around every 5 years, and due to the implications of AI, Klarna sees a natural churn of around 20% per year.”

And that’s because of AI?

It’s to some extent a pivot in strategy, moving from hyper growth and scale into something else. But to a large extent, it is due to the fact that we are starting to understand what AI can do for us.

What would you say to a CMO who is on the fence about using AI in their business, or not quite sure where it will and will not work?

It is easy to sit down and talk about the risks with AI, and what might or might not happen. I think that this is one of those pivotal moments in history, where you just have to decide that you want to lean in. It’s an infinite kind of moment where it’s instinctively feel[ing] that we need to be on top of this as an organization. AI, to me, is as much a cultural thing as it is a technology thing. It is about organizational culture, being curious, leaning in, trying. Making it part of your every day, even if it doesn’t solve anything huge for you from day one, but just get started.

The biggest win for us has been to automate boring things. Translations is a boring thing. You get a text string in English, and you need to translate it to 13 languages. It’s a very mundane, very basic, very non-value-add task. You’d be surprised how much of the shit work that AI can do for you. And once you realize that, you start uncovering that AI can actually do more and more and more.

What are your next plans for using AI in marketing?

I do think we’ve done a lot of work on the creation, production, assembly, ideation side. I think the next big frontier will be on the data side. We have loads of consumer data, which means that we have a very thorough understanding of what consumers want, when, what buttons to push. AI can process an unprecedented amount of data in real time. So applying it not to image production and text creation, but to segmentation, targeting, finding trends, finding moments. Those are the kinds of things we want to apply AI even more to.

Originally Appeared Here