Being a Data Scientist at If
The title “data scientist” has become hot and sexy. What you actually do as a data scientist, however, will vary a lot from company to company. What does it mean at If Insurance? It can be hard to know from the outside (and sometimes even from the inside).
At If, the persons with this title have many different skills and varied business focus. Interesting topics like NLP, A/B testing for marketing, insurance pricing and deep learning for image recognition are handled by our Data Scientists.
In this article we talk to Kalle Lindblad, data scientist at If.
What do you do and how did you end up in the field of data science?
I work with insurance claims. Most of our projects have been geared toward automation and improving customer experience. My background is originally in engineering physics. I have always liked the field of data science and I think that it can have a huge impact on any business if done right. I started out as a data analyst but have steadily moved towards more advanced topics that interest me more. This has been made possible by the increased focus within If on data and analytics during recent years.
Take us through a day at as a Data Scientist.
As I pour my first cup of coffee for the day, I log in to the daily standup meeting. The mood is usually cheerful and everybody is excited about the day to come. At the daily standup, the squad discusses any news and the plan for the day. The standup is usually followed by sessions of deep work or collaborative pair-programming sessions. Both are rewarding and productive in their own way. In today’s standup we wrote a couple of user stories, discussed how to move forward and then got to work. Two of us paired up to help each other solve a specific issue while the other two started on their own. I spent the day by adding a feature to a dataset and retraining a model that we are working on. After changing a bit of sql code and uploading the data to cloud storage, I pulled the data into a python notebook and retrained a model with the new feature. I then evaluated the model and saw that the new feature had improved the accuracy of the model.
Tell us about a recent project.
In recent weeks, our squad have been building machine learning models that try to predict whether a certain vehicle-panel can be repaired or if it needs replacing. In a wider context this model will be used to improve automation, cost and environmental impact when dealing with vehicle claims.
This is a fun project! Starting out we sketched the idea together with our stakeholders that have domain knowledge about vehicle repairs and know what is going on in a car workshop. After the initial phase, the project turned into a “traditional” machine learning endeavor. Finally, after training, deciding on the best model and getting buy in for moving forward we will build the API that hosts the model in production. This API will be a plug in solution to an existing automation solution.
We get to plan and build almost all of the moving parts! This is something that I really appreciate. It gives satisfaction to know that the squad have taken the project from beginning to end. It also gives great learning opportunities for the squad. Some might be more skilled data wranglers while others are more comfortable with modelling or building APIs. Having a varied range of tasks means that you get the opportunity to both teach others and learn.
What skills (technical and non-technical) do you draw on the most?
The non-technical skill I draw on the most is my urge to get things done and making things work. I have a hard time letting things go unsolved. I also get along quite well with our stakeholders and can be a bridge between the non-tech people and the tech people. As for the technical, I have good skills in data handling and a good eye for programming. Even though I am always trying to improve.
How will you develop as a data scientist in the near future?
The next project we are moving into involves working with image-processing and deep learning. Building an application that automatically detects damaged windscreens in images. This has been a new area for me and I have had to learn and test many new things. I look forward to widening my skillset into this area.
In insurance, the areas where data science can be useful are endless. The list of interesting problems to solve will only grow going forward. If has a need for competent people and the community of data scientists is steadily growing. So is the culture for continuous feedback, learning from each other and being a bit nerdy!
