AI and insurance: From data to decisions

By Jane Rheem, Chief Data and Analytics Officer, Zurich North America

Artificial intelligence (AI) has been around for decades, but you’d be forgiven if you thought it was invented in 2023. That’s because several technology companies unveiled generative AI tools this year that allow even novices to create digital content with a few simple prompts.

The insurance industry, which is one of the most data-hungry sectors of the economy, has been aware of the potential of AI for a long time and, as such, has been at the forefront when it comes to balancing the risks and rewards of this game-changing technology.

The recent emergence of generative AI has, for many, revealed the promise of AI, but it has also unleashed concern among those who believe an unregulated technological explosion could lead to disaster. Indeed, the prospect of sentient computers overtaking their human creators is frightening. But that is pure science fiction — for now.

We are a long way from being subjugated by computers. In fact, despite the long history of AI, we’re just at the very beginning of what’s possible.

Evolution of AI

Looking at the evolutionary scale of AI is important to understand where we are today and where we’re headed. Generally speaking, there are three phases of AI: narrow AI, general AI and super AI.

Narrow AI is programmed to perform only a limited range of predefined tasks. General AI is self-aware and capable of carrying out any tasks its “mind” can conjure. Super AI will be able to outperform human intelligence.

Despite the widespread and misguided belief that we’re in the throes of an AI revolution where machines are thinking for themselves, the technology that we’re working with today is narrow AI. The new online tools that allow us to create content in a matter of seconds has merely lowered the barrier to access AI.

For the insurance industry, that means our primary focus today is to transform unstructured data into structured data.

Unstructured data has no predefined format or organization, making it much more difficult to collect, process and analyze. Examples of unstructured data run the gamut, including text, video files, audio files, mobile activity, social media posts, satellite images, etc.

Structured data is highly organized and formatted so that it's easily searchable in relational databases. Examples include dates and times, phone numbers, banking transactions, customer names and addresses, product prices, etc.

Use of AI in claims and underwriting

The promise of AI for insurers is to help make more informed underwriting, pricing and claims-handling decisions. But the transformative power of AI goes beyond claims and underwriting — it could impact the entire insurance value chain. It may also play a big role in designing new products that are more intuitive and offer even more accurate insurance coverage.

Here are some ways we’re using AI at Zurich today:

• We apply AI through our business process, driven by business use cases, to reduce expenses, enhance customer satisfaction and improve business efficiencies. Our Claims organization, for example, has gained operational efficiencies by applying AI which otherwise wouldn’t have been possible, eliminating the need to spend days and weeks to go through extremely large claims documents.
• On the underwriting side, we’re extracting information from unstructured data sets to help underwriters gain additional insights which are otherwise difficult to derive from the massive number of unstructured documents.
• Through machine-learning models, we are improving both pricing and risk selection. Underwriters are consequently able to gain insights into the right pricing and make appropriate selections through data points.
• By applying AI, we are also able to create new insights for our customers so we can enhance risk management.

AI is a transformative tool that will be able to help the insurance industry and its customers manage and transfer risk. We do not, however, see it as a threat to the jobs of those who work in commercial insurance. We see AI evolving job profiles rather than replacing them. Large language models will help remove repetitive, time-consuming tasks to allow people to be more creative and perform higher-value work. The insurance industry is nothing if not a relationship business, and only humans can develop and strengthen those relationships.

Challenges of integrating AI

As the insurance industry further adopts AI and other new technologies, challenges will arise. For example, how do you integrate data from various technologies and disparate systems into one place for effective analytical use? How do you address the cultural changes that will accompany your transformation into a data-driven organization? How do you promote data literacy so everyone is aware of what data sets and insights are available for them and how best to utilize those insights into their day-to-day business operations?

Successful organizations will meet these challenges by fostering a culture of innovation and learning. Training modules will be created to keep employees up to date on the latest tools. Taking risks will be encouraged and, therefore, failure will be expected.

AI offers the promise of a more efficient and more effective insurance industry. This cannot be accomplished by algorithms alone. Constant learning — by people — and clear business objectives will be the keys to success.