Transforming insurance underwriting with AI

Underwriting is the process of assessing risk and then costing it. It is underpinned by data, and it is the explosion of data sources driven by digital technology that is transforming underwriting.

Underwriters assess the degree of risk within a particular activity as a tool to help them set competitive, but still profitable, rates for things such as loans, insurance premiums and financial securities. The more information they have, the more accurately they can assess the risk. And that is why data is so critical. 

Data-driven risk assessment

Traditionally, insurance underwriting relied heavily on historical risk levels experienced by different groups. They used demographics such as age, gender and occupation to assess risk. But while this data gives a basic level of information about risks, it cannot account for the nuances of risk that are dependent on individual behaviours.

However, in a world driven by digital communication and continuous connectivity, insurance companies can access an ever-expanding set of data sources. These new data sources can be analysed using AI algorithms to provide real-time, highly accurate and personalised insights into the behaviours and environments that drive risk up or down.

Generic data

There is a considerable amount of data that is independent of individual behaviour but that can still help underwriters because it provides a more detailed understanding of risk than was possible 20 or 30 years ago.

These sources of data include geospatial data, data about a particular location on our planet. For property insurance, for example, AI is used to analyse data about population movements and weather patterns taken from satellite imagery to assess risks to property associated with natural disasters, such as floods and wildfires. Insurers can incorporate environmental data to price insurance for homes in disaster-prone areas.

Related to this is smart cities data such as traffic data, pollution levels and crime statistics. Insurers can integrate data from smart city sensors to assess risks tied to specific neighbourhoods. For instance, high-traffic areas could have higher vehicle accident risks, while pollution levels might influence health risks. 

Individual data

While generic data is helpful for baseline insurance pricing, it has little effect on the prices of individual policies. But in most cases, insurance is purchased by an individual. If policy prices can be accurately related to individual risks, then the insurance company may find a marketing advantage.

One commonly used source of data is telematics. Increasingly telematics devices are installed in vehicles to provide real-time data on driving conditions and behaviour, such as speed, braking patterns and fuel consumption. This data can be used to create personalised risk profiles for drivers; for example, young drivers who are proved to drive carefully can be rewarded with lower premiums.

UK private motorists are aware of the potential of telematics, although as yet only 8 per cent have installed telematics devices in their cars (but note that, strictly speaking, any car with a satnav device is using telematics). Telematics can even support usage-based insurance models that are based on how much drivers use their vehicles.

Health data is another valuable source of insight. Health insurers use data from wearable devices like fitness trackers to monitor policyholders’ health such as blood pressure, activity levels and sleep patterns. An individual’s health risk can be assessed dynamically, with premiums based on real-time health indicators rather than on backwards-looking data such as medical history. Some health insurance companies such as Vitality have turned health data into a marketing tool by offering rewards for activities tracked by wearable devices.

More controversially, some insurers are exploring the use of potentially sensitive data such as social media data which can be analysed to detect risk factors that might not be evident from standard data sources, such as risky behaviour or fraud indicators. Mobile device data such as location and app usage can also influence underwriting by helping insurers understand lifestyles such as how much and where a person travels, how they travel and how they interact with their surroundings.

Most controversially of all perhaps, data from virtual assistants such as voice commands, and even emotion detected from speech patterns, can be collected. Again, this data could be used to provide insights into behaviour patterns that indicate lifestyle risks, such as health habits or purchasing patterns. Many people worry about these devices because they are always listening although that doesn’t mean they are necessarily recording everything you say. However, insurers that decide to use this data need to be aware of the damaging effect its use may have on their brand.

Stronger tools for analysis

Better (but not necessarily more) data leads to better understanding of risk. Artificial intelligence is revolutionising predictive modelling in underwriting by enabling the detection of patterns and correlations in complex and unstructured datasets. Traditional underwriting models are far more limited in their ability to capture the multifaceted and interconnected nature of risk.

AI algorithms can handle vast datasets with numerous variables and with connections and correlations that might be hard for unaided humans to detect. AI tools are generally not limited to “structured” data (orderly data in rows and columns) but can also use “unstructured” data, such as social media posts, images and sensor data.

Not only that, but many (most) AI systems use machine learning which means that they are continuously adapting as new data becomes available. This means that underwriting models are always improving, leading to more accurate and profitable risk assessments over time.

New business models

With more data and better data analysis, new business models become possible. Two new ways of providing insurance, which are likely to bring considerable competitive advantage to insurers, are micro-segmentation and dynamic pricing.

Micro-segmentation divides customers into micro-segments (sometimes segments of one) based on a wide range of variables that were previously not considered. With a more precise calculation of risk insurers can offer pricing more tailored to an individual’s circumstances and predicted behaviours.

With the dynamic pricing model, customers are offered premiums that change in real-time, based on changes in risk factors. For instance, a driver’s premium could be adjusted every month based on recent driving behaviour captured through telematics, a strategy that not only supports profitability but that also incentivises safer behaviour by policyholders. Similarly, usage-based models gather data from sensors to track how much customers are using their insured assets, such as cars. This information can then be used to adjust pricing and reduce the amount of “wasted” insurance that customers have to buy, even when they don’t need it.

The risks of using more data

On the surface, the use of unconventional data sources by underwriters seems extremely promising. But there are several risks that insurers need to consider carefully.

The threat to privacy is perhaps the most obvious risk. Privacy is highly regulated around the world and any insurer using personal data needs to understand how to remain compliant with laws as well as how to use it ethically.

By its very nature, much of the new data that insurers can access will be personal, and sometimes highly sensitive data. Insurers must make sure they are acting legally. For instance, where data is already being collected for another purpose, it won’t be sufficient for insurers to rely on existing privacy protocols: they will need to explain that data collected is now being used for a new purpose.

In addition, from a branding perspective, insurers that are seen to be “spying” on their customers, for instance by examining social media posts, will soon find they become very unpopular.

Fairness is another important consideration as the use of data could lead to certain people being excluded from cover. Genomics data could be used to predict long-term health risks, allowing for more personalised premiums. But this needs careful thought, not only because it is already highly regulated in parts of the world, but also because it could result in some groups being disadvantaged.

Another important issue with data is accuracy. More data is not always better data. For instance, health data taken from fitness trackers will not always be reliable if the device isn’t always worn, or if it is attached too loosely during exercise. Wearables also present discriminatory challenges because many of them struggle to track heart rates on people with darker skin.

And the analysis may also be flawed. While huge promises are made for the effectiveness of AI in analysing unstructured data, the results are not always sufficient: even a tool that provides predictions that are 87 per cent correct could have devastating consequences for the 13 per cent who are failed by the tool.

Changing the insurance landscape

AI is fundamentally changing the landscape of insurance underwriting by enhancing risk assessment and enabling more accurate pricing. As a result, insurers are likely to see significant benefits in terms of customer satisfaction and profitability.

However, successful implementation requires careful management of data privacy concerns, ethical considerations and regulatory compliance to fully realise the potential of AI in underwriting. As the technology evolves, its role in underwriting will only continue to grow, offering insurers new ways to compete in a rapidly changing market.