Artificial intelligence, and more specifically machine learning, is being deployed in the insurance space in some very exciting ways — from assessing underwriting risks to determining pricing to evaluating claims. But with these advances come sizable risks, some of which are already surfacing. Insurers need to take a proactive approach to mitigate risks so that they don’t wind up experiencing the same financial and reputational difficulties that other industries have seen.
Where is the harm?
In 2019, Apple launched a branded credit card in partnership with Goldman Sachs. Before long, users noticed that women were generally being offered lower preapproved credit lines than men. Many immediately took to social media to decry what they saw as blatant sexism on Apple’s part. Even the company’s co-founder Steve Wozniak joined in on the discussion, pressing the company to respond to the allegations.
Unfortunately, neither Apple nor Goldman Sachs were able to explain how the machine learning models that determined who was offered a specific line of credit worked. Their systems were not auditable, so they couldn’t explain the differences in preapproved credit levels except to say that gender was never explicitly part of their predictive model. Neither organization had a plan in place to respond to charges of proxy discrimination, so the result was a public relations disaster. The reputational damage to Apple and Goldman is evidence that companies deploying artificial intelligence need to get ahead of such risks by having governance structures in place in advance.
It’s fair to say that no insurer wants to be called out for discriminatory practices based on their models, nor do they want their pricing to be adversely affected if models fail to account for new information. There are a whole host of potential risks machine learning can expose a company to because of the technology’s present limitations. Does that mean machine learning shouldn’t be used? Not at all. Artificial intelligence and machine learning have the potential to be transformative for insurers. They just need mechanisms in place for governance and assurance to protect themselves over the short and long term.
Fortunately, the nascent fields of artificial intelligence governance and machine learning assurance aim to address risks and ensure that the technology is deployed successfully, in a manner that maximizes the overall business value of these technologies.
Preparation for artificial intelligence governance
What does good governance of artificial intelligence and machine learning systems look like? For insurers, that means developing cross-functional awareness of the potential exposure created by these technologies, especially among those responsible for compliance, audit and internal controls. It means adopting an intentional approach to AI governance led from the top of the org chart.
To get there though, it’s worth pausing to consider questions of organizational readiness with respect to artificial intelligence/machine learning. Does your company have the right people in place? Are executives aware of what it will take to make artificial intelligence/machine learning initiatives successful? Are your technology professionals well-versed in the needs that compliance will have as they develop their models?
There are a range of questions that touch on the technical domain as well: Which specific business problems lend themselves to artificial intelligence/machine learning applications? Does the company have the necessary technical resources and infrastructure in place to implement and manage artificial intelligence/machine learning initiatives? Is the right data available to feed machine learning models, and is it of sufficient quality?
You may not have all the answers right now, but you can certainly take concrete steps today toward many of these areas by empowering your cross-functional teams to take a proactive approach.
Best practices for managing risk
- Insurers that are considering or building with artificial intelligence/machine learning are positioning themselves well to get their products to market faster than their competitors and grow their share of market. To ensure that you can develop and deploy with confidence, consider the following best practices.
- Proactively implement a machine learning assurance program. Companies must be prepared to show and explain their work, that is, to demonstrate how their artificial intelligence/machine learning models are designed, how the organization went about selecting data and ensuring its quality, and what steps they took to proactively eliminate potential sources of bias. They must demonstrate that they have well-considered controls in place to protect against internal and external threats, to audit systems for potential violations, and to take prompt corrective action when necessary.
- Add oversight of artificial intelligence/machine learning into your existing governance structure. Insurers already have long experience with governance as a key to risk mitigation and unlocking value. They can take advantage of the existing institutional knowledge and processes to develop accountability and alignment across their organizations for ownership, management and problem mitigation. This starts with building a baseline competency in artificial intelligence/machine learning throughout the company, with strong cross-functional communication and collaboration at the center. Compliance officers, data scientists and line-of-business owners must work together to ensure that artificial intelligence/machine learning is implemented and managed with an eye to traditional risk areas like compliance, operational, strategic, reputational and legal risk.
- Stay up to date on statutory, legal and regulatory developments related to artificial intelligence/machine learning, both within the insurance industry and horizontally. Foster relationships with regulators to ensure that your efforts to mitigate risk are clearly understood, and work proactively to ensure regulators leverage your investment in accountable, intentional approaches to assurance and governance of these advanced systems.
These practices will help you get the most out of all that artificial intelligence and machine learning have to offer insurers, while still protecting your company from the limitations of these emerging technologies.
Was this article valuable?
Here are more articles you may enjoy.