Navigating AI Ethics and Uncovering Actionable Data for Accurate Risk Assessment: NeuralMetrics Is Helping Carriers Grow Their Commercial Business
The commercial insurance market offers lucrative opportunities for carriers — if they can find the right solutions to quickly assess risk, quote, and bind policies. Our technical co-founder, Marcus Daley, discusses how AI is significantly improving commercial underwriting, how to find the right data, and how to navigate the ethical considerations of AI.
What opportunity does NeuralMetrics see in commercial insurance?
Our feeling is that commercial insurance, including small/medium businesses (SMBs), is an underserved market for risk intelligence. It’s especially difficult for SMBs to get insured. Insurance is not something that business owners are familiar with and they don’t have the time to learn. They view it as an expense.
So, from the carrier perspective, we wanted to turn a high-volume market into a high-margin opportunity. And if we could create an open, unbiased data solution for risk assessment to facilitate precise coverages and fair premium pricing for policyholders, then carriers could more readily expand into the small/medium business market, and businesses that previously struggled to get adequate insurance would be able to get the coverage they need more efficiently.
On a larger scale, AI-powered classification and risk assessment data solutions are an opportunity to change the culture of insurance transactions in the SMB market segment. Our goal is for business owners not to perceive their insurance needs as just an expense, but as a strategic tool to de-risk and better position the enterprise for growth.
How does NeuralMetrics ensure the delivery of accurate data for risk assessment?
We focus on disciplined data extraction from a variety of public sources and generally assume whatever a company conveys to its customers is a solid indication of what they do. Regardless of product or service scope, we believe customer communications are an important element for identifying the risks associated with the business, and how the business should be classified. The business might have different motives for communications to non-customers, so that level of information might not be as relevant to their risk profile.
We also look at any compliance information about the business, including a range of licenses, permits, and policies, as well as government requirements to operate in compliance with regulations.
Additionally, we typically scrutinize social media. If the business has a substantial social media following, the data can be very useful in helping to determine risks and exposures. But we also know some businesses have a limited social media footprint. So this data track can have a bearing on risk assessment but usually not as much as other categories above.
While AI and data are improving the underwriting process, how can insurers make sure they are using the information ethically and complying with regulations?
In evaluating data on companies as entities, it’s best to steer clear of issues that can raise ethical red flags. For example, demographic data should be avoided, because making assumptions from that information can be murky and deficient.
Avoid using anything that could fall into the category of personal identifiable information. For example, insurers should not deploy AI to crawl the internet and pull information about the business owner, management, and the board of directors. Factoring information on individuals into the risk assessment can result in breeches of privacy.
Be cognizant of potential risks with AI. For example, image recognition can be a powerful tool. But depending on how the AI tool is programmed, bias could be introduced into risk assessment. Image recognition AI learns from humans. It uses how humans interpret and associate value to an object in making its inferences. So the biased view of human “teachers” towards a particular object can get programmed into the AI tool.
Here’s a useful benchmark for averting bias: Can the information be reused across industries and can it be linked back to a source document? This is a good way to avoid the introduction of bias, but still gain valuable data insights for risk assessment.