Insurtech 2.0 and beyond: Examining the risks and benefits of AI
The digital revolution in the insurance industry is now at a more mature state and there’s no doubt it is helping our industry become more efficient and innovative. However, new technologies such as artificial intelligence (AI) only add value if they are deployed under the keen eye of a thoughtful and knowledgeable human.
Let’s first look at the advantages that AI brings before examining its attendant risks.
The AI advantage & data integrity dangers
AI allows underwriters to bring in more information to better assess risk and price premiums more appropriately and in a more tailored way. It can be used on large datasets to automate a host of processes in a fraction of the time it would take a human to analyze the same amount of information.
AI enables insurers to look at a wider set of data than that traditionally provided by cedents, which is not always reliable. More data gives the underwriter a more detailed picture of the risk involved and allows for a more customized approach — for example, if you are a better driver, you can have better motor insurance rates, or if you work out regularly you can pay lower healthcare rates. AI’s use of larger data bases means that people who couldn’t get insurance can now be covered because the underwriter understands them better. AI is great at deciphering the complexities of risks and creating tailored solutions, offering benefits for the end user, as well as more business for brokers and carriers.
AI can also transform claims management functions such as settlement, notification and fraud detection. It can detect fraud patterns that might escape the eye of even the most watchful claims team, while algorithms can speed up the distribution chain, examining the information between the insured and the carrier more quickly and accurately, which reduces human error and manual work.
Using AI, then, makes eminent sense. But sound data hygiene is essential to prevent risk. AI doesn’t always get it right, taking information from various sources including the internet that may be inaccurate. Machine learning-based AI learns as it acquires more data, and if that data is not clean and reliable, it is possible to veer off track and quickly spiral into giving incorrect advice.
And sometimes, its assumptions are simply wrong.
Human intervention required
A glaring example of how AI assumptions can be incorrect is with compliance-related information, which is sometimes state-specific. AI may not differentiate statute information between one state and another, which means it can infer the same regulations apply to all states. It’s essential that there is a rigorous verification of the data being inputted to not only check it is accurate, but also that it is being used correctly.
You can’t put AI into place and assume it’s going to work well forever due to the inherent learning the AI does. Output needs to be constantly monitored and thoroughly tested. Human intervention needs to be in place to validate the AI’s conclusions before they go out the door, along with customer feedback measures so that you can quickly fix any problems that arise. Monitoring and testing go hand in glove with ensuring your data is as accurate and clean as possible. To mitigate against this risk, our firm only inputs data sets into AI models that we have rigorously validated, and we have an ongoing process to continually monitor the quality of the AI outputs.
When it comes to detecting fraudulent claims, AI is incredibly useful, but professional criminals keep up to date with fraud indicators and will change their behavior accordingly, meaning data scientists need to update their fraud analysis indicators on a regular basis to stay ahead.
APIs: Open exchange, for better or worse
Open application programming interfaces (APIs) are everywhere — every time you send a mobile or online payment or change the heating in your home through your phone, you’re using an API.
The software intermediaries allow two applications to talk to each other, and for different organizations to extract and share data.
Open APIs allow data to be exchanged between multiple parties in the insurance ecosystem. Carriers integrating their data this way can streamline workflows, access a wider pool of broker business, and have greater transparency in pricing and coverage, while offering an easier end-user experience. This can also help brokers with online quoting, enabling them to easily quote from more carriers.
But what happens if one of the parties involved doesn’t have good data hygiene or undergoes a cyber breach? Either of these scenarios could cause a black eye for all partners in the ecosystem. Unless all parties rigorously ensure security and check data, there is a danger that their information could contaminate the workflows of partners. It’s vital to choose partners with the same security and quality mindset.
Carriers need to use open APIs to be competitive, but some may be reluctant to share their data in a transparent way for fear of disclosing commercial secrets, so there also needs to be a great deal of trust between organizations sharing information this way.
The Internet of Things: Too easy?
Similar problems arise with interconnectivity, also known as the Internet of Things (IoT). The IoT enables multiple benefits, such as using smart phones to remotely control everything from TV sets, heating and home security devices. However, with greater interconnectivity comes the risk of wider systemic IT failure — a data breach from one connected partner could corrupt the data of many different users. There are risks for the insurance industry, such as a criminal infiltrating the system by yelling through a window to an Alexa device to unlock a door or hacking into and stealing a self-driving car.
The insurance industry relies not only on technology itself, but also that tech providers are doing a good job at providing security. In addition, they need to have partners in the insurance ecosystem that also protect themselves from data breaches. If a vulnerability in the ecosystem is discovered, it’s imperative there’s a good communication system in place to alert all parties, then rapid action to patch it and update the system.
As technology makes further strides, customers are going to be looking for coverage to protect interconnected systems from being breached. Insurers need to understand the magnitude of what could happen in the case of interconnected system failures and effectively price the risk.
