The AI-driven Insurance Company

By Øyvind Indrebø, Head of Machine Learning and AI, Fremtind

Øyvind Indrebø, Head of Machine Learning and AI, Fremtind

In 2012, Harvard Business Review published an article titled “Data Scientist: The Sexiest Job of the 21st Century”. In many ways, this marks the start of Fremtind’s and my artificial intelligence and machine-learning journey. Although it seems highly unlikely that the title will ring true 80 years from now, it had a huge impact on sleepy organizations with numerous data, analytics, and statistical professionals, i.e., your average insurance company. Suddenly we realized that we could make a similar claim: “I have the sexiest job in the 21st century.” After all, if we could not call ourselves data scientist, then who could?

Is it evident that insurance needs to become AI-driven? I certainly felt that it was, but arguing that AI is one of the most important technology and that it will have a significant impact on insurance than in many other industries felt hardback in 2014. By using insights and tools developed in Ajay Agrawal, Joshua Gans, and Avi Goldfarb’s book: Prediction Machines: The Simple Economics of Artificial Intelligence, the same cannot be said today.

Artificial Intelligence and machine learning are at its core, a means of making useful predictions based on data. Making predictions based on data is bread and butter for an insurer and in the 20th century, solving insurance problems contributed significantly to statistical research and methods.

Artificial Intelligence and machine learning are at its core, a means of making useful predictions based on data

Much in the same way as autonomous driving is pushing development in, for instance, computer vision today.

In order to understand what the insurer might lose by not keeping up with prediction technology, you have to look closely at both insurance value and growth proposition and the core of Artificial Intelligence technology. In Prediction Machines, Agrawal et.al argues that if you apply simple economic tools in thinking about AI-technology, you will easily understand its potential and its impact on your industry. Furthermore, Agrawal explains that AI serves for a single, but potentially transformative, and economic purpose; it significantly lowers the cost of prediction which in turn, leads to three types of effects:

1. Where we are already doing predictions, we will be making more and more accurate predictions.

2. The prediction will be used to solve problems, that were not previously thought to be a prediction problem, e.g., self-driving cars.

3. When the cost of prediction drops so will demand for substitutes drop; however, the demand for complements will increase.

Let us do a thought experiment on the impact of the effects above on insurance. From the first and second categories of effects, it becomes clear that there is not going to be a single digital process/service/application that will not make use of machine learning-based predictions.

If we consider the third category of effects we also have to admit that insurance can act as a substitute for lack of predictive abilities. We only have to look at the motor industry where self-driving cars are promising a lot fewer accidents. Prediction technology (the self-driving car) might significantly reduce the need for insurance.

We might already have lost our stake of the values generated from improved prediction technology in the motor industry, but if we double down on our effort to make life better for our customers and to become a truly AI-driven company, Fremtind will still be a relevant company for the next 80 years.