Sebastian, Director of Credit Underwriting, started working at Klarna in 2010 as a Risk Analyst fresh out of university where he’d spent the last six years. With an eagerness to learn, those six years could have been many more, but when the student grants stopped coming he thought it was time to become an adult and find a job. As a Risk Analyst he dealt with a broad range of tasks: reporting, bureau negotiation, credit policy rules, and strategic optimization. But here is the story of how he ended up writing history for Klarna when he built the first credit scorecard*. Sebastian describes his six years at Klarna as hectic, formative and challenging.
Sebastian, how did you end up working at Klarna?
The people I studied with said that Klarna was hiring. They also said it was a waste of time to apply because of their super hard hiring process. But I thought it sounded like a good challenge, so I applied. After four interviews and some tough tests and cases (I had to put in an extra snus), I got the job!
My first project was to re-design the way we did automatic credit underwriting in Sweden by using predictive modelling and making more use of our internal data.
Two colleagues and I basically locked ourselves into a small room and had three months to deliver the blueprint for a new way of approving and rejecting customers. The room was not the biggest, and we did the majority of the analysis and programming one the same computer, sitting shoulder-to-shoulder. Never leaving the room except for lunch and dinner, we eventually nicknamed it “the aquarium”. But instead of clown-fish, it had three newly minted Risk Analysts teaching themselves to make predictive algorithms with the help of only Google and some books.
In the beginning we did not have any databases, so we had to build our own from a ton of different CSV files. A third of the way into the project we found out that our BI department had hired a consultant to set up the first database with internal data. We kidnapped the consultant and placed him in The Aquarium to speed up the process of making sure that the database being designed fit our needs.
I was sent to an external training course to learn some best practises for building scorecards. That, together with Google, books, curiosity, hard work, and brains formed the building blocks of the first scorecards.
What impact did the scorecard have on Klarna’s success?
The initial impact was anything but a success. In the beginning we switched it on for all transactions for two days and then turned it off while we waited for data to mature so we could validate the performance. The three of us who had made the scorecard watched eagerly over the shoulder of another who did the validation. The results were checked and double checked: the losses had doubled! After some screaming and tears and an intense hour of trying to figure out what had gone wrong (and considering new professions), we got some positive news: the payments files were a week out of date. Every payment from the past week had mistakenly been registered as a loss! After fixing the payment files and re-running the analysis, we saw that the new scorecards kicked ass and generated tons of value for Klarna as well as our customers.
What challenges lie ahead within your department?
There are so many opportunities! We constantly challenge ourselves to automate and streamline processes and then to look to the next opportunities. Since there is so much unexplored territory, in terms of both small-scale optimization and entire new areas of research, it’s very hard to choose what to set aside for now. But we are in the fortunate position of being able to expand the team. I’m looking forward to both the giant leaps and the small fixes that will pave the way to optimal risk decisions.
It’s the ability to make a difference that makes Klarna an exciting place to work. It does not matter who I am or what I’ve done, if I come up with a good solution, it will be appreciated. This has given me the possibility to solve problems that even I thought I was too junior to handle.
And lastly, tell us about a time when you really messed up?
I have to admit that I’ve made many costly mistakes. For example, there was a model implemented incorrectly that resulted in millions in losses. But even the incorrect model was eventually able to offset its losses over time once the bug had been fixed!
* Scorecard is a way of displaying the result from a statistical model that is trained to predict the likelihood of a certain event, usually if customers will pay or not.
Do you want to be a part of Klarna’s Credit Risk team?
Great! Check out this link – we are hiring!