Artificial Intelligence
Mar 22, 2019

AI in insurance

Demystifying the possibilities and challenges

In today's digital age, artificial intelligence (AI) is transforming practically every industry, including insurance. By leveraging AI, insurance companies can make smarter underwriting decisions, better manage risk and fraud, and create differentiated customer experiences.

While AI holds tremendous potential for insurers, there is uncertainty when it comes to where the best opportunities lie and how to build a practical framework to get started. And to take full advantage of AI, we need to first clear the air on what is possible today.

Understanding, responding, seeing, and learning with AI

We can split AI applications in insurance into four areas:

  1. Language: Using natural language processing, insurers can extract the meaning from written text, thereby turning previously unstructured data in documents (e.g. a pdf) into structured data (which can be run through computer engines). As a result, with AI, it's now possible to extract information from and automatically classify claims submissions. In underwriting, AI can automatically pull and process applicant information, streamlining the previously cumbersome approval process. If we then add in machine learning, AI can even learn and guide future decisions
  2. Conversation: Conversational AI is all about understanding intent and threading together parts of a conversation into a whole. One of the most popular applications of AI is chatbots. For example, if customers reach out to verify the coverage in their policies, the bot can intelligently understand their requests across the many sentences that are entered in a chat conversation, pull up the right policy and documentation, and reply back – all of this can be done over voice or text. It's easy to see that this benefits customers in speedy resolutions, while freeing up agents to concentrate on more complex and strategic tasks.
  3. Vision: Computer vision is the ability to see an image, recognize the various components, and make a logical decision around one or more of the components. Much in the way we visualize the damage on a car, computer vision can assess damage based on images of the car and provides an innovative way to improve the processing of automotive claims. Trained using a library of photographs from past accidents and their adjustments as labeled data, AI can “look" through photos from a new accident and estimate repair costs. This significantly cuts down the time between claim submission and payout, vastly improving the customer experience.
  4. Knowledge: AI can be used to build rich knowledge graphs, linking and understanding relationships between disparate sources of information and capturing the collective intelligence of team members with insurance expertise. Insurers creating these knowledge graphs can then bring this extraordinary intelligence to play in automating critical decision-making for different types of incidents, such as hurricanes, fires, or earthquakes. These can in turn allow insurers to solve many of their challenges using the cognitive capability of AI applied to a variety of business situations.

These four different applications are listed from most mature to least mature. Language processing is the most mature and readily available form of AI accessible to insurers today. Indeed, insurers who haven't already started using language processing are at risk of falling behind in their digital journeys.

A practical framework for deploying AI

As insurers look to deploy AI, it's important to reflect on some of the biggest challenges with AI and a practical framework to work through them.

  1. Explainable AI: One of the most exciting aspects of AI is that machines can form their own algorithms, without depending on a set of human-programmed rules, and produce a decision or answer. The advantage of this approach is that it solves many last-mile automation problems that previous waves of technology couldn't. The downside is that AI then becomes a black box in which we can't explain why it made a specific decision. This can create distrust among employees and customers, as well as create issues with regulatory bodies that seek explainable AI, especially when its decisions affect consumers. To address the black-box problem, applications need to be designed with traceability in mind – for instance using breadcrumbs so a recommendation number can be broken down into its components, in a click-and-drill fashion, all the way back to the source set of sentences in a claims submission.
  2. Low data density: Amazon Alexa and Apple's Siri work so well in understanding everyday requests and delivering responses because they are built on large data sets. The challenge for most enterprise applications is that they lack the same volume of data. As a result, it's important to think through which algorithms would best address each problem and work in low-data environments. For instance, computational linguistics is a branch of natural language processing that understands meaning based on the surrounding context of a sentence, not on vector mathematics, which can help in situations where AI does not have a lot of data to work with. Picking the right tool for the right job is vital.
  3. The need for domain expertise: While a lot of public discourse about AI centers around the replacement of people in the workforce, humans with industry and process knowledge are critical to realizing the full value of AI. A machine cannot understand the nuances of an insurance policy and how it would apply in different scenarios. Plus, AI has to be built, sustained, and maintained by human employees. These domain experts help with contextualization, goal orientation, and eliminating potential biases. Equally important are people with working AI knowledge. There is a great opportunity now to re-skill current staff with AI knowledge, as well as recruit bilinguals – i.e. people with both industry and technology knowledge.

With an understanding of the possibilities and challenges facing AI, we can use a practical framework to drive the successful deployment of the right AI capabilities in the best use cases across insurance.

Read more about AI adoption, challenges, and impact in our AI 360 study with senior executives, workers, and consumers. 

About the author

Sanjay Srivastava

Sanjay Srivastava

Chief Digital Officer

Sanjay Srivastava is Chief Digital Officer, where he runs Genpact’s growing Digital business, overseeing the Genpact Cora platform and all Digital products and services.