Klarna saved $40M in customer support and spawned Superbowl style campaigns a core CS company. 5 lessons on what I think they did differently and that everyone can learn from.

Clean up data "really" well:

On a podcast with Sequoia, CEO Sebastian Siemiatkowski mentions great instructions/ documentation was a key secret ingredient for their success. This is not easy - documentation always has a few gaps. My guess is that Klarna improved its CS documentation through GenAI itself - questioned the documentation through AI, asked AI to revise it, review drafts through humans and got much better documentation!

Know state of the art (SOTA):

Klarna did well by partnering with OpenAI to understand what is available. While not everyone has Sam Altman's number (see the awesome commercial Voiceflow made on this, link in first comment)- today, there are leaderboards, playgrounds, custom evals to test the toolset and get a sense of SOTA before you take a plunge.

Human in the loop:

For any mission critical flow like refund (which is one of the major actions that Klarna can take), there is a human fallback.

Focus on easy to change stuff:

No on-roll jobs were lost through this and this was used for reducing outsourced spend. This would have helped rally the internal stakeholders and kept change mgmt to a minimum

Don't over-spec:

My guess looking at Klarna's chatbot's responses, feels that this is a RAG implementation. Squeezing all that you can from less intensive tech (vs training/ finetuning) is the cost efficient way to go about using AI.

While the 700 person years saved might have some creative mathematics behind it, fair to say that there is some creative thinking that all enterprises can imbibe as they think of agentic solutions!

PS: I am sure Bredan Ream has lots to say to this. Inviting him through this shoutout!

https://www.linkedin.com/posts/braden-ream_keepitupkim-activity-7172939389289787392-9ZOE/