Navigating the Matrix: The Risks and Rewards of AI in the Life Insurance Industry*

Artificial Intelligence (AI) is top of mind today in the business community and elsewhere, and if estimations are correct will continue to be so in the future. It has been described in terms such as "transformative" and "next level," and promises to be just that. And since computer systems perform tasks, it is predicted to offer a myriad of efficiency, productivity, complex problem solving and other important benefits.

The uses of AI are potentially limitless, especially in the life insurance industry where business is based upon knowledge and data.

To better understand AI and what the future may hold, we've brought together three high profile Life Industry leaders to answer a few questions and get their perspectives on AI so we can better understand and adapt to this far-reaching, transformational tool.

Our panel of experts include:

Aadil Lokhandwala, Senior Vice President, Office of Strategic Planning, Lincoln Financial Distributors

Gretchen Juneau, Vice President, Traditional Underwriting, Prudential

Hezhong (Mark) Ma, Vice President and Managing Actuary, Reinsurance Group of America, Incorporated (RGA)

As it relates to Life Insurance, what do you see as advantages and potential uses of AI?  

Hezhong (Mark) Ma - RGA:

Most articles would say the biggest advantage of artificial intelligence (AI) is its ability to generate insights that are obscure to humans. Undoubtedly, this is a real and significant advantage. Yet, to me, the most important impact of AI in the life insurance space lies in its ability to deliver a digital acquisition and sales experience. Underwriting used to be a human-centered process and it routinely took weeks, or even months, to purchase a fully underwritten policy. The lengthy process has always been a top reason why people choose not to buy life insurance. By programming AI into the underwriting process, insurers can significantly cut down decision-making time and deliver an Amazon-like, seamless digital purchase experience. It is why, despite mortality slippage, accelerated underwriting is becoming the norm in the industry today.

AI is also a means to enterprise learning. In the past, I worked with many actuaries with metallic memory and superhuman analytical skills. Companies often took advantage of such intelligence, structuring the organization around it, which gradually introduced key-person risk. Learning through a properly designed and governed AI is the intellectual property of the carrier. It can continuously grow, even with personnel changes. Think of it as a bread machine that might be subpar to artisan bakers, but it will always be there, consistently and reliably getting its job done.

Every day, we see new AI technologies in creative applications. Recently, many carriers started to utilize AI-powered tools to accelerate the underwriting process or support algorithmic underwriting decisions.  At RGA, we offer a few insurance solutions that are backed by AI. For example, RGA RiskDimension RxPM and MedScore are trained to take in medical history to predict mortality risk.

I chair the Society of Actuaries Research Institute's Actuarial Innovation and Technology Strategic Research Committee (SOA AIT) and, recently, the committee sponsored a research project by Ernst & Young: Predictive Analytics and Machine Learning – Practical Applications for Actuarial Modeling (Nested Stochastic). The project included a case study of successfully using AI and machine learning to reduce the running time for the valuation of exotic index crediting strategies.

Some advanced AI techniques, such as neural network deep learning, are mostly used in handling unstructured data, including images and text. Large language models are often used to extract individuals' dates of birth from unstructured data, such as OFAC notes. Actuaries are not eager in adopting these technologies as most actuarial analyses start from tabular data. But I can see more and more creative uses of AI in the industry. For example, in 2019, a research project sponsored by SOA AIT explored different algorithms, including recurrent neural network models for time series predictions:

Aadil Lokhandwala - Lincoln:

AI isn't totally new to Lincoln – we've been utilizing AI for many years now.Our view is that AI can augment the capabilities of our existing workforce, allowing employees to offload many manual or time-consuming tasks and increase overall productivity and efficiency. This is a tremendous go-forward opportunity to unlock significant value for Lincoln!.

There are numerous tools to build artificial intelligence systems – from classical machine learning algorithms to deep learning, and of course Generative AI.All these technologies can be deployed in various forms across the insurance value chain to drive better outcomes for all stakeholders.

As an example, there are some low-hanging opportunities that exist within our sales, service, underwriting and new business teams. These include creating access to an internal knowledge base with specific underwriting guidelines, product specifications, case design concepts, files and emails. This could allow our teams to drive greater insights and be more collaborative with stakeholders.Ultimately, you would open up this knowledge-base to customers (both agents and consumers) so that they can learn on their team, ask those "first-line" questions and work with the right stakeholders on more complex and thoughtful needs.

Gretchen Juneau - Prudential :

At the heart of Life Insurance and the mathematics behind the science is the law of large numbers. To construct a mortality table that sits behind underwriting guidelines you need a vast amount of data. The more data, the higher the confidence interval is around the estimation of life expectancy. As underwriters, we have used data for years . . . application data, exam and lab data, APS's. Now we have Electronic Health Records as well as several other new sources of data. Data is critical for AI to work effectively. Its advantage today is that it allows us to build models and rules engines that can enhance the client experience by issuing policies quicker with less invasive data. It also allows our underwriters to focus on the higher risk and higher face amount cases.

How do you see AI impacting sales? Will it replace the Independent Agent?


We see AI as a catalyst to drive meaningful growth in incremental sales in insurance and more broadly.

AI can enable agents to focus on sales-generating activities – whether that is prospecting new clients or client engagement – saving time and allowing more top-line growth.We also believe that with this increase in capacity we will be able to drive into new segments of the markets as we can democratize certain tasks like personalized marketing campaigns and basic product education.

AI will certainly not replace the human connection and relationship building that is the core of a successful independent agent, but rather it will empower them to complete more impactful sales-generating activities.These tools and technologies, when deployed will increase productivity, allow greater work-life balance and provide better outcomes for our Agents and their customers.


Some people see AI as a threat, but emerging trends are demonstrating that AI is there to enhance and not replace people. It can do the tedious research work that no one really wants to do. It could set the groundwork for a sale, but the purchase of insurance is and should be a somewhat emotional one. AI has no emotions, but people do.


Many producers welcome the acceleration of underwriting powered by AI. Some insurtechs talk about using generative AI to create underwriting summaries for communication with agents and applicants. Other AI tools are being used to evaluate individuals' financial wellbeing, capture lifestyle data, and recommend products. The rise of direct-to-consumer products redefines the role of sales. Nevertheless, most face amounts are still sold rather than bought. The jury is out whether the future will see more commoditization of services or further innovation along the spectrum between simplified issue and fully underwritten business.

How do you see AI changing carriers organizationally and in what areas? Does it have the potential to replace New Business, Underwriting and/or Inforce staff at carriers?  


Data Science will need to be integrated into the organization and data scientists will need to work with other areas that have traditionally been responsible for Underwriting and pricing. The data scientists who build AI Models need input from the experts in the business including Actuaries, Underwriters, and Medical professionals.

AI may replace some portions of a job, but people are still at the heart of any organization. We can automate where it makes sense, utilize AI to better organize data, and use people's skills where they add the most value.


The changes are already happening. In support of the Colorado bill SB21-169, the state of Colorado exposed a draft of model governance and risk management framework. It gives broad visibility and responsibility to chief data officers, and other relevant stakeholders. Large corporations are teaming up their data strategy, data science, technology, and domain experts, such as actuarial, claims, and underwriting professionals, to tackle strategic objectives, including accelerated underwriting, fraud detection, novel product designs, and alternative distribution channels. When those goals are achieved, companies will have a core group of experts from different areas, seamlessly working together. Many young professionals are already equipped with the understanding of AI tools and skills across multiple domains, and those who can handle large and fuzzy data will move between departments until a new profession centered on AI is formally recognized.


The technologies to effectively build, deploy, manage and govern AI will require organizations to re-organize around these capabilities.It will require organizations to re-skill and up-skill existing employees and add new job categories that have not previously existed, such as Prompt Engineers or Machine-Learning Engineers.

What will be needed within all of these functions is training and skill building on how and when to best use AI-enabled capabilities.AI has its limits and it will require the workforce to understand when it is appropriate and how to continue to improve the technologies.

One of the major concepts in safely building scaled AI capabilities, especially with an innovative technology like generative AI, is the need for guardrails like having a "human-in-the-loop".When developing these capabilities, you want to make sure you have human subject matter experts reviewing the output generated by the model to ensure it meets the stated outcome and, if it does not, fine-tuning these models to get the appropriate response.These types of skills are not core to the current teams' responsibilities but can be learned and integrated into the team training.

Because AI is data driven, one of the cons is the potential for discrimination and bias due to data sets used in the decision-making process. How do you see the industry addressing these issues—especially with regulators? Can AI ever be free of bias?  


The insurance industry adheres to the federal Unfair Trade Act in order to eliminate discrimination against protected classes. However, the continuous emergence of new data sources and advanced algorithms is posing new challenges. Nowadays, attending an industry conference without a dedicated session on the ethical use of AI is nearly impossible. A fundamental belief of actuaries is that history repeats or rhymes, which is why we work hard to identify patterns and risk drivers from experience studies. Lately, more and more sociological debates and discourses suggest that historical data may have flaws. Some risk drivers, even if backed by solid experience, could be deemed unacceptable by society. Furthermore, practices that are considered acceptable today might become embarrassing for future generations. To make it worse, AI is famous for being a black box: It adapts to data without providing straightforward means for humans to understand it.

Insurance is a heavily regulated industry. Companies rely on regulators to define terms like bias and proxy discrimination, and to spell out how to test and mitigate bias. Those definitions and regulations may not align precisely with how outsiders think. Instead of debating whether AI can ever be free of bias, we could focus on developing processes to define and detect unfair treatment of consumers resulting from AI use. This will certainly require years of collaboration among regulators, carriers, and industry/consumer advocacy groups. Nevertheless, the industry can take immediate actions. RGA, along with several other companies and consulting firms, has published its own guidelines in ethical use of data and AI. Improving model governance and risk management would be an ideal first step. Are you satisfied with the transparency of the model and the model-building process? By providing more people with the under-the-hood details, especially by engaging subject matter experts in the area where the model is intended to be applied, companies are more likely to identify unintended consequences of AI, even though it does not guarantee full detection or elimination.


Having a broad governance framework will be imperative to scaling AI across any organization.Having policies and risk management practices designed to tackle AI will be paramount to building trust and adoption.Making sure you crawl before you walk is important, so developing internal experiments that are not being used in isolation to make decisions is an important way to understand the gaps and opportunities as you develop models.As you mature through the use-cases, having a rigorous process in which you continuously refine and test the outputs of the model will help remove any unintended consequences due to bad input.

As it relates to the regulatory environment, we fully expect frameworks to be established at the federal and state levels as well as by industry-specific entities which we will fully comply with and make sure we advocate for the best practices going forward.Like with anything new, we must expect some roadblocks but need to ensure we are not doing any harm to our stakeholders along this journey.


Underwriting and Life Insurance pricing by its very nature is selective. Better risks get better rates. Regulators' concerns are around unfair discrimination via models and non-traditional underwriting data that may not have a direct link to life expectancy. The concern is that this data will make a correlation without causation.

We, as carriers, and with the help of our vendors, need to be proactive in addressing the issues. We need to understand the regulators' concerns and work with them to show that we are not using data inappropriately. We also need to show regulators that what we are trying to achieve is a win for consumers – providing a product that provides financial security in the fastest, most cost effective and least invasive way possible.

Looking forward and in terms of years, when do you believe the majority AI impact will be felt by the life insurance industry? How close are we to this change?  


I expect to see some interesting use-cases and deployment strategies over the next 18-36 months in our industry.This may seem like a long time, but in a highly regulated industry deploying capabilities to our customers' needs to be heavily thought through and evaluated.I do believe we will benefit from industries like retail and consumer technology to understand the "art of the possible" which will benefit our industry and our stakeholders.At the same time, insurance carriers will be testing out AI in a variety of experiments behind the scenes to understand how to best deploy and manage these capabilities in a responsible manner.


Depending on your definition of AI, we are already there. Recent NAIC bulletins describe AI as follows: "Artificial Intelligence" is a term used to describe machine-based systems designed to simulate human intelligence to perform tasks, such as analysis and decision-making, given a set of human-defined objectives. This definition treats machine learning as a subset of artificial intelligence.

A rules-based underwriting system is on the spectrum of AI. More and more reinsurers and data vendors are using AI models and rules engines. The question is how far will we take it? What other forms of AI, including the recently popular ChatGPT, should insurers be looking at? How will regulations impact how fast we move? There are a lot of unknowns at this point, so it's hard to predict how things will look in two to five years but we do know that AI is here to stay and we should all be embracing the opportunities.


As previously illustrated, AI is already here and will continue to grow in both the technology and applications. The adoption of AI can lead us to higher efficiency, more precision, increased automation, and ultimately more financial protection for consumers. Yet, the path will not be linear. In addition to the social and regulatory concerns surrounding AI, as mentioned earlier, its black box nature can limit its usefulness. A variety of techniques are already available to assist with tasks, such as data visualization, SHAP, LIME, and more. Those in charge of applying algorithms should also focus on producing case-level reason codes and establishing a monitoring program, among other steps.

Thanks everyone!  Any closing comments or thoughts?


Early in my career, I had a mentor. His first job was to punch cards for computers, a tedious job of translating commands into a bunch of holes on a piece of paper for computers to interpret. When more modern computers stopped needing those punch cards, some actuaries initially felt lost, realizing that the skills they had developed were no longer relevant. However, many of them quickly recognized that they were liberated from mundane tasks, allowing them to engage in more intellectually stimulating work. This is happening again. Those who embrace the changes of today will be empowered by new tools to make a more meaningful impact.

[* Blog title courtesy of ChatGPT]

Making a Statement and Setting the Pace in (Very) ...

Related Posts



No comments made yet. Be the first to submit a comment
Already Registered? Login Here
Saturday, 13 April 2024

Captcha Image