How Klarna integrated AI in their business processes

Hi Futurist,

The revelations of the past two weeks are coming at us fast and furious. Each revelation is quickly buried by the next. Therefore, I am sending you insights, inspiration, and innovation straight to your inbox. Let’s dive into the depths of the digital future together and discover the waves of change shaping our industry.

In this post, we're dipping in:

• 📣 Byte-Sized Breakthroughs: Three (technological) developments in the digital landscape

• 🎙️ MarTech Maestros: one webinar, podcast, or interesting paper I believe you should watch, listen to, or read too

•  🧐 Last but not least: Things that I like

Do you have tips, feedback or ideas? Or just want to give your opinion? Feel free to share it at the bottom of this dip. That's what I'm looking for.

Now available as a podcast: Listen on the go!

Always on the move and no time to read? No problem! Now you can listen to our Digital Dip wherever and whenever it suits you. The link to the podcast is only available to subscribers. If you haven’t subscribe already, I recommend to do so.

📣 Byte-Sized Breakthroughs

'Byte-Sized Breakthroughs' briefly highlights (technological) developments in the digital landscape

Headstory: How Klarna integrated AI in their business processes

I can well imagine that it might still be challenging for organizations to look beyond the implementation of AI in existing processes or tasks within your organization, other than using ChatGPT for your daily work. That's why the case of Klarna, the well-known 'Buy Now, Pay Later' service provider, is so intriguing. They have launched an AI assistant, powered by Open AI. This chatbot handles queries such as refunds, returns, or cancellations—margins that are extremely important to Klarna in their revenue model. It touches the core of their service.

First some numbers. In one month, this chatbot has conducted 2.3 million conversations. That’s two-thirds of all customer service talks. The equivalent of 700 full-timers. It scores an equally high customer satisfaction score. It leads to a 25% reduction in question repetition. It resolves questions in 2 minutes, instead of the previous 11 minutes. And importantly, it's estimated to improve profit by $40 million. This is aside from being available 24/7, in 23 markets, and communicating in more than 35 languages.

This is not the future. This is happening now. And this could be for your organization too. But how and where do you start? Perhaps the following three steps will help:

1. Develop a data strategy: To optimally use the possibilities of (generative) AI, your data strategy needs to be in order. This is also the reason the chatbot works so well at Klarna. Start by consolidating customer data from various sources to create a uniform and complete customer view, so you gain insights into customer behavior and preferences.

2. Identify specific challenges and opportunities: Once your data strategy is in order, you can think about challenges or opportunities within the organization where AI can have a significant impact. This can vary from customer service chatbots, like at Klarna, to better personalization or operational efficiency.

3. Implement and experiment: Begin implementing AI solutions on a small scale, through pilot projects. This allows your organization to learn and adapt quickly, without significant risks. Focus on measuring the impact and using these insights to refine and expand the implementation.

It sounds easy. I know. But your organization needs to seriously start thinking about how you can integrate AI within your business processes. Otherwise, you will inevitably be outcompeted. Start small. Reason from use-cases to then go bigger and broader. Ultimately, it's change management. But now, everyone needs to be on board.

OpenAI introduces Sora

TL;DR

Sora is OpenAI’s text-to-video model. It can generate videos up to a minute long while maintaining visual quality and adherence to the user’s prompt. The quality is unparalleled and not yet shown elsewhere. Currently, Sora is not publicly available, mainly because OpenAI itself understands the immense impact this model could have on the creative world.

Read it yourself?

Sentiment

The internet went wild. The examples shared by OpenAI truly astonished everyone. The development path envisioned for this type of AI has been significantly accelerated by people. So much so, that Singapore is now offering retraining opportunities to residents over 40. People can't wait to start generating videos themselves.

My thoughts

This sets a new standard. Just a month ago, I was telling people in my workshop how to make videos with Midjourney and Runway, four seconds at a time. Now, it can be done with a simple sentence. A whole minute. This opens up a vast can of new creativity for everyone. At the same time, the time to value becomes shorter and shorter, budgets are reduced, and you get more time to test, try, and make mistakes. Perhaps it's already time to start counting down to your own AI Netflix?

Google revealed Gemini 1.5

TL;DR

Google has unveiled Gemini 1.5, a new AI model boasting significant improvements over its predecessor, 1.0 Ultra. This model excels in long-context understanding, handling up to 1 million tokens and showing potential for up to 10 million. This advancement enables processing vast amounts of information across different modalities, including text, audio, and video, opening new doors for developers and enterprises.

Read it yourself? 

Sentiment

The response to Gemini 1.5's capabilities has been overwhelming. People are astounded by its ability to handle an extensive context window of up to 10 million tokens, which enables remarkable feats like full tutorials, large codebases understanding, and multimodal operations. This leap forward has ignited excitement for the potential applications in various fields, showcasing the AI's prowess in retrieval, zero-shot learning, and cross-modality operations.

My thoughts

The introduction of Gemini 1.5 Pro is a game-changer, not just for AI developers but for society at large. The possibility of processing and interacting with up to an hour of video or texts as lengthy as 700,000 words is mind-boggling. Consider the implications if this technology were integrated with platforms like YouTube. It could transform the site into the largest educational, engagement, and entertainment platform ever conceived, allowing for interactive experiences with long-form content that surpasses anything currently possible. The ability for AI to understand and interact with such vast amounts of data opens up unprecedented opportunities for learning, creativity, and communication.

More byte-sized breakthroughs:

• Groq is the fastest AI chatbot you know
  ChatGPT is criticized for slow responses, while Gemini is praised for its speed. Yet, Groq surpasses both with instant replies upon pressing enter, thanks to its innovative Language Processing Unit (LPU). Unlike others relying on GPUs, Groq's LPU is a new kind of end-to-end system, excelling in fast inference for AI language tasks.

• Apple is entering the sports industry with its free Sports app
  Apple steps into the sports arena with its free Apple Sports app for iPhone, offering comprehensive sports stats, standings, and play-by-play action for leagues like MLS, NBA, and Premier League. Unlike other platforms, Apple Sports centralizes data for easy access, marking Apple's ambitious move into sports market engagement.

• Finally: The European counterpart to GPT-4 has arrived
  Mistral partners with Microsoft, launching AI model Mistral Large. This alliance brings Mistral's advanced language models to Azure AI, marking Microsoft's progressive AI venture. Mistral Large, equipped with a chat interface similar to Chat GPT, excels in benchmarks. Representing a significant European contender, it promises to reshape the American-led AI field.

🎙️ MarTech Maestros

In 'MarTech Maestros,' I share one webinar, podcast, or interesting paper I've encountered, which I believe you should watch, listen to, or read too.

Navigating the future of marketing: AI takes over

This comprehensive guide is crucial for every marketing professional aiming to stay ahead in an increasingly AI-driven industry, offering strategic insights, best practices, and a roadmap for navigating the complexities of modern marketing. A must read.

 🧐 Last but not least

• Describe a sound and generate it: De video’s van Open AI’s Sora kwamen zonder geluid. ElevenLabs dacht daar anders over.

• A new era of digital content for the sports industry: With Move AI, you can create 3D animations from 2D images. Now you can easily capture athlete movements to create engaging digital experiences for fans.

• OpenAI discontinues plugins and fully commits to custom GPTs: Due to the success of the custom GPTs and the GPT store, OpenAI will discontinue the plugins in April. At the same time, they are adding new options to the GPT store, such as rating and feedback for custom GPTs. Additionally, there is now an 'About Us' section from the creator of a custom GPT.

• OpenAI has been developing a web search service: OpenAI is possibly developing a search engine to compete with Google. Many users use ChatGPT as a search engine. OpenAI already has its own web crawler, so a step towards a search engine is small. For OpenAI, this is a new addition to tip the competition against Google in their favor.

• Open source model with 1M tokens: There exists an open source AI model, built on Meta's Llama 2 model, with 1 million tokens. The current best OpenAI model has 'only' 128k tokens. The open source model, for instance, can provide extremely accurate answers to questions from an hour-long movie, something ChatGPT cannot do.

• Spatial Personas for FaceTime with Apple Vision Pro: Apple is working on a 'multiplayer' mode, with the ultimate goal of allowing multiple people to be in the same space. To 'truly' look someone in the eyes.

• Two users share space in mixed reality: Not only is Apple working on a multiplayer mode for their AR/VR headset, but Meta is also engaged in this. They launched a demo called ‘Cryptic Cabinet’, a sort of digital escape room that transforms your living room into a play space where you and another person must complete tasks.

• Google introduces an open-source model called Gemma:
  Google has announced Gemma, an open-source model based on Gemini. It is available in two variants, one of which is also suitable for mobile devices.

• Gemini available in Google producten: Google is bringing Gemini to Gmail, Docs, Slides, and other Google products. Generate emails. Have documents written. Create presentations. You need a paid Google One Premium account for this.

• Generate and directly test Python code within Gemini: Within Gemini Advanced, it is possible to test the Python code you just generated. In this way, you can directly see what the output is and whether the code is correct. You also need a paid account for this, but you can then use it directly for the capabilities described above.

• Gemini image generation capabilities halted: Google temporarily pauses the ability to generate images in Gemini. What's happening? The model generated inaccuracies in some historical image generation depictions.

• Google's deal with Reddit for AI training data:: Social media platforms sit on a gold mine. Data. Not just for advertising opportunities, but also for the potential it offers as input to train your AI models. Google is willing to pay 60 million dollars for it. Per year. Interesting, because on Reddit, users share their real opinions.

• Google launches their latest model 'Genie': It’s an AI model that can generate interactive video games based on a simple prompt. Naturally, this is super interesting, but unfortunately, it's not publicly available. This tool is, however, and claims to be able to do almost the same.

• DomoAI Video-to-Video 2.0: With a press of a button, you can change your video into various styles through Domo AI. See the examples here.

• Possible partnership between Midjourney and X (formerly Twitter): Elon Musk hints at a partnership between X and Midjourney. There are a few reasons. One might be to recognize AI-generated content on the platform. Another is to generate content within X. Final clarification is still pending.

• Text-to-3D method that can generate complex scenes: This model can generate complex scenes in 3D with multiple objects based on text input. So far, one of the better models.

• Adobe introduces CAVA research team (Co-Creation for Audio, Video, & Animation): Adobe has established a team of 50 researchers working on all aspects of video creation - video, music, speech, sound effects, animation, and interaction design. It wouldn't surprise me if Adobe launches their own Sora this year.

• $100 million in investment due to claims of achieving AGI: This company is still somewhat under the radar. But their goal is to develop artificial general intelligence. And they claim to have made a significant move towards this.

• Disney includes Elevenlabs in its accelerator program: The accelerator program is a program where Disney invests in companies shaping the future of storytelling. They have now also included Elevenlabs in this program. Should we say goodbye to voice actors?

• Snapchat launches their own text-to-video model: Snapchat is also working on a text-to-video model. The quality is far from comparable to Sora but is a competitor for Runway or Pika Labs. The crux is, Snapchat is about sharing stories. And they see this as an addition to that.

• Stable Diffusion 3 launched: Stable Diffusion is one of the leading text-to-image models out there. They recently launched their latest model. This model is even better at generating images with multiple objects, image quality, and spelling abilities.

• Alibaba is launching a model that can generate a video based on audio and an image: Have you ever seen the Mona Lisa talk? Or an AI Mona Lisa sing a song by Miley Cyrus? Well, it's possible here.

• This robot is now completing real-world tasks: Really, everything about this is autonomous. Crazy to see a robot doing a human job.

• Deutsche Telekom wants to free smartphones from apps: For entertainment, shopping, or travel. Currently, you have to switch between different apps. If Deutsche Telekom has its way, not anymore.

• Goodbye privacy with the transparent screen: Lenovo has shown a prototype of a laptop with a transparent screen. Seriously. Why would you want this? I'm already scared when I'm on the train and someone glances at my screen.

• Parody variant of Terminator 2 with AI: 50 AI artists will show a feature-length parody remake of Terminator 2, fully with AI images, on March 6th. Check out the trailer below. Or if you have the chance to be there live in Hollywood, buy your tickets here.

• Pika is launching a new function called Lip Syncs: Yes, Sora is astonishing. But imagine if you could combine that with this feature from Pika?

This was it. Our sixth digital dip together. It might seem like a lot, but remember; this wasn't even everything that happened in the past few weeks. This was just a fraction.

I can imagine that this might be quite overwhelming. Understanding all developments, overseeing them, and being able to reason about their impact. I would like to have a conversation with you. Even if it's just to discuss your data strategy or to identify the first use cases together for incorporating AI into your business processes. Click reply and let’s have a chat.

Looking forward what tomorrow brings! ▽

-Wesley