Artificial intelligence in insurance (Part 1): AI in action

AI is entering the mainstream—and it has some powerful applications for the insurance sector.

By Eileen S. Burns, FSA, MAAA, Consulting Actuary; James Dodge, Practice Leader, Advanced Analytics & Data Solutions; and Tony Huang, Data Scientist, Advanced Analytics & Data Solutions, Milliman.

The chief scientist at Baidu, which is China’s answer to Google, has described artificial intelligence (AI) as “the new electricity.” His point is that AI has the potential to revolutionize how all industries manufacture goods and deliver services.

AI clearly has enormous potential to improve insurance operations. But implementing it is far from straightforward.

In essence, AI allows computers to do what humans do—but better, faster, and cheaper. The concept is nothing new; it has been around since the 1940s. But in the last few years, developments have taken AI out of the laboratory and into the real world.

The trend for digital business models has been accompanied by the declining cost of processing power and data storage—which has made access to large data sets and high-performance computing cheaper and easier. Meanwhile, there have been considerable improvements in the algorithms and platforms on which AI systems are built.

The resulting hype surrounding AI has led commentators to envisage all sorts of applications that could revolutionize the way we live, and how businesses work.

Practical applications
For insurers, there are two clear, practical, and highly beneficial applications in the immediate term.

Firstly, AI can enable insurers to use data and mathematical models to improve predictions of future events (predictive modeling). And secondly, it offers techniques that use past experience to detect relationships, and projects how they may play out in the future (predictive analytics—which can include AI, machine learning, and deep learning).

Giovanni Marchetti, Program Manager, Machine Learning and Data Science Ecosystem at Microsoft, notes that, “Machine learning techniques can be used to help insurers quickly and efficiently forecast behavior and identify anomalies in a vast data set.”

Improving predictive modelling and analytics has the potential to:

• Enrich the customer experience—e.g., by using robo-advisors, or analyzing click patterns, to help customers find what they are looking for
• Enhance risk selection and pricing—by improving insurers’ understanding of customer behavior
• Provide feedback to customers that helps them minimize their risk—e.g., via apps linked to auto and life insurance products that aim to make drivers safer, or encourage healthier lifestyles
• Improve fraud detection—machine learning is already being used to identify patterns in false applications and dishonest claims
• Enable better claims processing

On the ground
At Milliman, we’re employing machine learning techniques in several ways. They’re being used to improve our policyholder behavior models, and to discover new drivers of risk, nonlinearities in relationships, and complex interconnections. This enables insurers to make better assumptions, which in turn helps them to price, reserve, and capitalize more accurately.

The most important of these techniques is the discovery of new drivers of risk. As we follow the property and casualty (P&C) industry into adopting new sources of data such as credit, lifestyle, and demographics, we find ourselves with an ever-expanding universe of possible drivers. Investigating each driver in turn, when there are hundreds or thousands of them, is an impossible task without machine learning. Machine learning, and some creative thinking on model form, allows us to quickly identify which drivers contain the most helpful information about how policyholders behave.

Part 2 in this series on artificial intelligence in insurance will look at how to implement AI projects, and the risks associated with them.


no comments

Thank you for your submission.