Marius Blaesing, founder and CTO Getsafe
Their names are Carlo and Allie (Allianz), novomind iAGENT (AXA) or simply travel assistent (ARAG): more and more insurers are using interactive chatbots to automate customer service. They often talk about neural networks, machine learning and natural language processing. But how intelligent are the systems? A closer look quickly reveals that most insurers are still miles away from artificial intelligence (AI). They fail because of a fundamental problem.
It could be so simple: you are shopping when your smartphone sends you a message that the washing machine is leaking. A sensor on the device detected that water was running out of the machine and immediately stopped the water supply. Identifying risks and preventing them instead of paying for damages – this would be a win-win situation for both customers and insurers.
But what is possible from a technological point of view fails in practice due to outdated IT systems. Artificial intelligence cannot be realized with programs from the last century and promising words won’t change that.
Insurtechs want to close this gap. They have the technological knowledge, the lean structures, the short decision-making paths and often the necessary risk capital to implement innovative ideas quickly. The insurtechs do not lack ideas or full-bodied promises – on the contrary: here, too, the founders brag with marketing jargon and superlatives. But what about the AI-truth?
Let’s start with a definition of terms. The problem begins with the fact that everyone understands artificial intelligence to be something else. This makes it much more difficult to compare allegedly smart solutions – in case of doubt, even tape announcements in waiting loops on the phone are celebrated as sophisticated technical innovations.
Put simply, AI refers to computers that are capable of more or less providing assistance and more or less solving problems and making decisions on their own. Artificial intelligence is therefore an attempt to create systems that can provide human-like intelligence services. But what does that mean? After all, there is a good argument about when a person is intelligent – and when he or she is not. And even supposedly intelligent people cannot solve all decisions and problems. AI demands more than pure computing power and means that algorithms can learn from “experience”.
The prerequisite for this is an infrastructure that allows customer data to be bundled over the entire contract term and all interfaces. Almost all providers are still confronted with countless data silos. The marketing department knows how many customers read the newsletter, the sales department knows the number of customers from Berlin or Cologne, and the customer service department has a good idea of which amounts of money are paid for which damages – but the companies cannot connect the individual data points with each other. Instead, each department works with its own IT system, with its own data and their own third parties, and in case of doubt, customers from different departments are bombarded with conflicting information by mail, email and phone – preferably on the same day.
At the same time, almost all insurance companies have heard the wake-up call of the “young savages” and are working intensively on digital solutions. The declared aim is to relieve employees of time-consuming routine tasks so that they have more time for more complex tasks. The potential for the use of artificial intelligence is high: many processes are data- intensive and repetitive – good prerequisites for automating processes.
It can therefore be assumed that artificial intelligence will fundamentally change four areas in particular: Firstly, it will accelerate numerous insurance processes such as the conclusion of policies or the settlement of claims. This will make insurance products more flexible and customer-friendly. Secondly, artificial intelligence will enable insurance fraud to be detected more quickly and risks to be assessed more accurately. As a result, insurance companies will be able to refine their customers’ risk profiles and reduce their loss ratios. “Good”, trustworthy customers could then benefit from lower premiums or repayments. Third, AI can be used to create personalized products. If the insurer can better assess the customer’s needs based on historical or behavioural data, he is in a position to put together tailor-made insurance packages. And last but not least, AI, in combination with sensors and intelligent devices, helps with prevention. In industry, machines can already be repaired before expensive downtimes occur in production and manufacturing. Similar concepts are also conceivable in the home, in the car or when it comes to health.
However, AI is still in its infancy in many places. For example, the ERGO Group is experimenting with language assistants that can recognise and process natural language. The Versicherungskammer Bayern uses IBM’s cognitive system, which became famous under the name Watson when it defeated people in Jeopardy. Watson is supposed to recognize anger or irony in customer letters. Basler Versicherung has automated the processing of glass damage according to its own specifications. Carlo”, a chat bot that is to be offered via the Facebook messenger, is still in the test phase. The robot is to calculate offers for car insurance. And the ERGO subsidiary Europäische Reiseversicherung (ERV) and the Deutsche Familienversicherung (DFV) offer international travel insurance in purely digital form via Amazon’s language assistant.
What these approaches have in common is that they are usually based on rule-compliant algorithms that obey firm rules. In addition, according to a study by Bain, many insurers concentrate primarily on product development and sales when using artificial intelligence. However, the greatest potential lies in the downstream processes.
What most insurers lack to exploit this potential is data. Only if algorithms are fed with data can they make better decisions and learn from experience.
As one of the first insurtechs, Getsafe has now laid the foundation stone for machine learning: A unified insurance platform makes it possible to record and evaluate data across the entire value chain in a structured manner. Algorithms for automatic claims processing or fraud prevention are currently being trained. One, Lemonade, Coya and other insurtechs are working on similar solutions.
One thing is clear: the future of insurance is digital. And it is also clear that the backlog of modernization of traditional insurance companies offers a decisive competitive advantage for young insurtechs. It remains to be seen whether they can exploit this advantage for themselves.