There are lots of ways generative AI is being used by companies. In case of Klarna they launched an AI customer support assistant that is available 24*7 and supports 30+ languages. They launched the assistant using Open AI as their LLM provider. If we have to guess they have finetuned it using their own customer support interaction data set that they have collected over the years. We often underestimate it when we say a customer support AI assistant, but if you look at what it takes to create a support team that works 24*7 and supports 30+ languages, it requires a lot of resources for any company.
What features does the Klarna AI assistant support?
- The assistant is designed to handle wide range of questions related to refunds, returns, payment-related issues, cancellations, disputes, and invoice inaccuracies, ensuring swift and effective solutions.
- Like any tech feature, its available 24*7 across all geographies.
- For customers who want to know what they can afford and not afford, it also provides info related to purchase power, spending limits and the reasons why it exists.
- All the info of a customer account is with the assistant to help any questions related to it.
Is the assistant effective?
Its cool to say that we have used gen AI and see the stock pop up, but the real question is, does the feature deliver value to the company. And from the numbers it looks like it is beating expectations. Here are some of the numbers.
- Two-thirds of all customer service chats are handled by the assistant, which is an impressive 2.3M conversations.
- Customer satisfaction same as human conversations.
- 25% drop in repeat inquiries
- Estimated to drive $40M in profit improvements to the company.
Multilingual coverage has also improved the experience for immigrant and expatriate shoppers, who can now converse in their native language without waiting for specialized staff.
How it is built?
Klarna combines OpenAI’s GPT-4-class model with a proprietary retrieval-augmented layer that injects fresh order data and policy documents. Fine-tuning on years of support transcripts helps the bot mirror Klarna’s brand voice and comply with local regulations. A confidence-scoring system automatically hands uncertain cases to human agents, keeping quality assurance intact.
Should you build one too?
If your firm has a sizable archive of labeled support data, start by clustering tickets to find the top 20 intents—typically 80 percent of volume. Spin up a retrieval-augmented bot in a sandbox, keep humans on standby for low-confidence cases, and track three metrics: time-to-resolution, repeat-contact rate, and CSAT. Double-digit gains in two of those usually justify rollout.
A rollout this sweeping touches more than technology. Klarna ran “shadow agent” weeks where the AI answered in parallel with humans so supervisors could audit every response. Once accuracy held above 90 percent, traffic was gradually shifted to the bot. Daily dashboards track hallucination rate, hand-off frequency, and compliance flags that regulators can audit after the fact.
What are the takeaways from Klarna’s gen AI assistant?
- Its clear that generative AI is being adopted and creating value for companies at all stages.
- Customer support will get more effective and cheaper.
- We will see a lot of growth stage startups follow Klarna to boost their profitability and reduce expenses.
- Open AI’s board drama didn’t effect its trust among companies to use them for production scale deployments.
- Expect more companies to announce improvements because of using gen AI technology in fields outside customer support.
- If you are a company with large data set related to customer support, you should explore creating an AI assistant for support interactions.
Follow me on Twitter, LinkedIn for latest updates on 100 days of AI. If you are in tech you might be interested in joining my community of tech professionals here.