What’s next for AI in life underwriting?

People toss around buzzwords like "revolutionary" and "transformative" so much they often lose their impact. So what do we really mean when we say that AI is revolutionizing the future of life underwriting? 

To start, I’m going to challenge you not to limit your perspective. With AI, the pace of innovation is astonishing—we launch a new model every two weeks. What was impossible six months ago is now part of our daily operations. This rapid evolution in AI in the insurance industry brings capabilities that were once thought unachievable.

One of the key drivers behind these significant technological advancements is the power of combining different AI technologies. It's not just about enhancing the AI’s capabilities; it's about creating a system that's significantly more powerful than the sum of its parts. This approach is essential for tackling common AI issues like hallucinations and the opacity of so-called "black box" systems. By utilizing a combination of AI tools, we gain the ability to provide click-to-evidence access to the source documents directly from the AI’s output and can reduce the risk of hallucinations down to nearly zero. 

So let’s talk about accuracy—it’s often the elephant in the room with AI discussions. Everyone wants to know the percentages—95% accuracy, 97% accuracy, etc. But in the realm of life underwriting and claim analysis, we need to think beyond just numbers.

The real question is: Was the decision made using the AI as good as, or better than, the one that would have been made without it?

That's the benchmark we set for ourselves, and our solutions enable users to measure it effortlessly. It’s about creating AI that isn’t just correct, but is useful to the underwriters. It’s the same reason it’s so critical to use large language models (LLMs) that have been specifically trained for the insurance industry instead of off-the-shelf models like ChatGPT and Google AI. If you think about everything that off-the-shelf models can do—from writing emails and poetry to planning trips—the training required for those models is very different from the training required for AI in insurance underwriting. Our models are not just about regurgitating correct information; they interact in a human-like manner, suggesting logical next steps, and are well-equipped to handle the complexities of life underwriting and claims review. 

There’s one more thing that’s critical to understand when discussing the differences between different AI models, and that’s the “context window.” It’s another buzzword that gets thrown around often, though few people truly understand its significance. For underwriters, using an AI solution that isn’t limited by a context window is critical to success. 

In simple terms, here’s how it works:

1. The model will read the document and segment it into topics.
2. Each segment will be assigned a code.
3. When a user asks a question, the model will assign a code to that question.
4. The LLM will review the source document and search for a code that matches the code of the question.
5. The LLM will then answer the question based on what it read in the matching section of the source document.

Sounds simple enough, right? The problem arises when there is a limit to the number of paragraphs or words the LLM can review to answer a question, resulting in a sharp decline in accuracy as the volume of text expands. Many times, there will be more sections of text needed to answer the question than the number of paragraphs the LLM can process. That’s why many engines have a limit on the number of pages they can read. It’s also why, with a limited context window, you may see a dramatic reduction in accuracy with longer documents.

In the context of life underwriting, the significance of this can’t be overstated. Can you imagine how many paragraphs of a medical record the LLM has to read to tell you when diabetes was first diagnosed? Or how long a patient has been taking heart medication? These are the questions that underwriters ask every day. If you ask a question like that to an LLM with a limited context window, that’s where you’re likely to get a wrong answer, and wrong answers can get you in a lot of trouble in the underwriting world. So when you consider using generative AI in insurance, make sure you’re using a solution provider that has no limitations on context windows, like DigitalOwl.

For life insurance, selecting AI technology that you can trust is paramount, especially within the tightly regulated insurance industry. As AI governance committees become more influential, the choice of AI service providers will demand even greater attention. Insurers must prioritize providers that adhere to ethical, transparent, and accountable practices and demonstrate flexibility to adapt to future governance frameworks. Choosing an AI provider that meets these dynamic standards is essential for successfully navigating the evolving AI landscape and ensuring sustainable success.

What's next for AI at DigitalOwl? Honestly, the possibilities are limitless. As fast as AI evolves, we continue to adapt and innovate, ensuring that every step we take is grounded in enhancing the services we provide, improving the accuracy of our processes, and maintaining the trust of our clients.

In embracing AI, we are not just adopting new technology; we are advocating for a transformative tool that will redefine the future of medical underwriting and claims management. This is the revolution I speak of, and it’s happening right here, right now, at DigitalOwl.

Contact us to learn more about how we’re revolutionizing the life insurance industry. 

‍