What's the promise of large language models for life insurance?

By now you’re likely familiar with the prospect of large language models (LLMs). These augmented intelligence tools, with user-friendly interfaces, combine artificial intelligence (AI) with human intelligence, enhancing and amplifying human abilities: generating content, delivering answers quickly, decision making, problem solving, and beyond. LLMs will drive digital transformation for organizations that embrace it, prompting reflection on the value of and talent requirements of human staff. 

Largely brought to light by the launch of ChatGPT by OpenAI and Microsoft in November 2022 and the subsequent launch of the AI chatbot Google Bard, news of this transformative technology captured attention globally—with good reason. LLMs represent the next generation of natural language processing (NLP) and natural language generation (NLG). By tapping into billions of parameters and diverse public data sets, the deep learning and transformer-based models of LLMs deliver flexible, diverse, publicly-available output, with guardrails established by the provider.

What impact may LLMs have in the life insurance industry, specifically? 

Value

Current LLMs offer two main categories of value in the field of life insurance, aiding varied end users and use cases:

“Tell me” (descriptive) provides functionality for customers and employees, alike. At this fundamental level, LLMs may provide training and guidance. For example, they may offer policy information (giving policyholders quick, accurate information about their coverage, deductibles, and other policy details) or text synthesis and analysis (pulling from varied documents and trained with an organization’s information to identify specific items). This can be useful with the unstructured information commonly found in the life domain.”Do it for me” may refer to customer-facing smartbots that provide information and troubleshooting, or employee-facing services, such as risk mitigation tools. LLMs can help drive everything from marketing (content generation, social and email marketing, data analysis, and A/B testing) to insurance-specific processes related to underwriting (gathering applicant information to determine risk profile; analyzing integrated health, insurance, and alternate data for straight-through accelerated underwriting; and claims processing, with automated initial stages to gather policyholder information, complete data entry and document verification, and determine eligibility.)

LLMs offer the potential to deliver even greater value for life insurance in two additional categories. Together, these have the potential to analyze large amounts of data (e.g., for identification of fraudulent data; consolidation of health data with policy data to predict potential fraud; pattern and anomaly detection), strengthen customer service (through integration into advisers’ apps and websites to deliver instant responses to customer inquiries, reducing workloads of advisers), and improve operational efficiency for varied insurance functions (including claims processing, fraud detection, underwriting, and premium calculation).

“Tell me” (predictive), differing from the descriptive functionality identified above, this category draws on the potential of fine-tuned generative pretrained transformer (GPT) models. Pretraining on large amounts of text data enables the model to learn patterns and relationships in the data, fine-tuning that information for specific language tasks (e.g., text generation, question answering, and sentiment analysis).”Advise me” taps into the realm of machine learning (ML) and the subfield of deep learning models, such as decision transformers. 

Economics

LLMs offer the potential to facilitate a new wave of automation, risk/loss mitigation, and data-driven decision-making for insurers. The returns are made possible through improved market data analysis to better inform growth strategy and the ability to leverage predictive modeling of customer behavior. The cost of these initiatives are for maximizing automation and minimizing risk/losses. 

Revenues may be improved through direct (product sales) and indirect (share of wallet gains) opportunities. LLMs may drive direct revenues by optimizing pricing; guiding current customers to optimal products; improving the product selection experience for new customers; and increasing the strategic and creative work of insurance staff, removing repetitive tasks, and delivering improved customer insights. Indirect revenue may come through improving the customer experience, providing personalized recommendations, and generating upsell and cross-sell opportunities for current customers, while increasing engagement and amplifying brand awareness for new customers. 

Celent anticipates the rapid adoption of LLMs by enterprises in the coming years. In 2023–2024, the early adopters (~20%) will be innovation-driven companies that are experimenting, fine-tuning, and building/testing LLM prototypes. 2025–2026 will be a critical proving period (re: economics, compliance), during which the early majority (21–50%) will invest in LLMs, with the late majority (51–75%) adopting and integrating LLMs around 2027–2028. LLMs are likely to reach maturity by 2029, when laggards (76%+) have adopted the technology and when the growth in use case scope will likely taper.

Implementing strategy

Where on the curve will you be? Insurers that aim to adopt and implement an LLM strategy must be farsighted, bold, and responsible. They must evaluate their short-, medium-, and long-term goals; the need for dedicated resources; and the cost implication for platforms and services. They must be willing to embrace disruption and paradigm shifts. Their development and deployment of AI must be accountable, compliant, ethical, unbiased, and transparent. This requires steps across the organization, as implementing an LLM strategy may impact the insurer’s tech infrastructure, business model, operating model, and culture. 

Because LLMs are relatively new and evolving rapidly (including, at time of writing, the launch of OpenAI’s newest model, GPT-4), insurers will need to consider their policies and practices for deploying these technologies. For example, filters or other measures to limit the use of LLMs for certain initiatives may be appropriate. Careful consideration of LLMs is also important in order to navigate ethical considerations about bias, for example. 

The pace of LLM innovation can be head spinning, but not taking action comes with its own risks. Competitors that adopt LLMs may gain a lead that life insurers who choose to wait will find difficult to close. Customers who have used LLMs, perhaps in another industry, will have higher service expectations, such as for AI-powered chatbots. Employees who value AI/ML tools—particularly because of how they enable staff to focus on value-adding work—may leave for other, more advanced firms. Last, but certainly not least, life insurers that continue to rely on manual processes may suffer from inefficiencies that LLMs would be able to automate, reducing operational efficiency and increasing costs. Though each of these is a potential risk, each also presents the opportunity for innovation and success for those life insurance companies that choose to invest in unproven technologies and dictate the pace of change.