How LLMs Create New Possibilities For The Insurance Industry

Kannan is SVP, Global Head – Insurance and Industry Head – Financial Services at Infosys.

In my last article on the topic of generative AI adoption in the insurance industry, I touched upon the efficiency and cost savings it can bring to underwriting, claims processes and customer services.

I would like to delve deeper into the industry applications of large language models, which can be considered the roots of all generative AI. To do this, we will have to start from a familiar place: understanding how the open-source LLMs we use on a daily basis operate and why they respond the way they do.

A Basic Understanding Of LLMs

It is well known that LLMs are capable of processing vast amounts of information. This information has to be acquired and intentionally fed into the model for it to act as a source for query resolution.

Aside from the quantity of information that is provided to the model, the quality matters a great deal. Since the information is based on humans and the systems they construct, the possibility is both real and high that LLMs can get trained with the same biases and errors in judgment to which we all are prone. This is why at present, most open-source platforms are not reliable enough to be used for critical tasks that have political, socioeconomic, corporate or other repercussions.

How LLMs Can Be Trained To Suit Industry Needs

Contrastingly, adapting LLMs to a vertical need—for example, in insurance—is done with the express purpose of deploying them in high-stakes real-world scenarios such as underwriting, claims settlements and fraud detection. This means that the expected output from the LLM should be of high quality, unbiased and factually accurate.

Acquiring pristine data and skilled personnel to train these models, therefore, becomes a time- and resource-intensive endeavor for any enterprise. This is why I strongly advise that companies equip themselves with a strategic road map of budgets, turnaround timelines and projected ROI before starting the LLM trek.

Where To Begin And How To Proceed With Enterprise LLMs

In my view, the first step in charting an LLM plan is to decide whether to use a public, hybrid or private LLM. If seen on a spectrum, the degree of control, security and efficiency increases in proportion to the privacy level of the model.

As expected, public platforms pose several risks and disadvantages to enterprises that are looking to scale up their LLM usage and develop exclusive or proprietary applications. For example, feeding classified data into open-source platforms means that anyone can access it, which vastly shrinks their scope for integration into sensitive operations. Furthermore, due to a lack of customization options, the training potential for these platforms is limited.

This is why I believe domain-specific models become a covetable and potentially lucrative option for organizations that wish to gain industry recognition and strengthen customer trust. In data-rich, regulation-heavy fields such as legal, insurance and finance, the avenues for creating valuable, business-enhancing applications are many. Some prominent use cases of LLMs in the insurance value chain are insight generation and submission pack summarization for underwriters, improved claim estimation for adjusters and simplified semantic search functions for accessing knowledge repositories.

Training LLMs In Domain-Based Skills

Since we have already established that training LLMs is both strenuous and expensive for enterprises, it would be wise for companies within an industry to consider joining forces to achieve their goals. Using anonymized data, sharing findings and helping improve the fine-tuning of LLMs can be mutually beneficial for the partners involved.

Since LLM development is a nascent and fast-evolving arena, it requires a comprehensive set of rules and regulations. Like-minded organizations can work with national and international legal entities and certification boards to create ethical frameworks for LLM innovation and monetization.

At the same time, they can release verified case studies, manuals and other enterprise knowledge assets to help their industry peers begin their LLM journeys.

Possible Approaches For Enterprise LLM Integration

Given that the financial sector has so many varied functions—internal, peer-facing and customer-facing—it would be practical to test the efficacy of different models on various tasks before locking down any of them for long-term use. In my view, it is also advisable to rely on public/hybrid models for generic, nonsensitive company functions while allowing core tech teams to work with private models and confidential data.

Such an approach can help long-term budget allocations and shorten the time frame within which companies begin significantly benefitting from AI implementation (also known as a return horizon). The difference in time and energy investment depends on low-level and high-level training of public/hybrid and private models. While chatbots and internal data-fetching applications can be optimized with relatively less training and cost, precious funds can be directed toward critical research in the areas of fraud detection, claims processing, pricing calculations, detailed underwriting and more.

How Leaders Can Prepare Their Employees For LLM Adoption

When introducing LLMs into an organization, there is bound to be an acclimatization period during which employees must adjust to changes in workflows and technological upgrades. I strongly advocate that leaders be keenly observant of their employees’ reception of overarching and department-specific changes.

It is crucial to acknowledge that every functionality within the company will be impacted in a different manner by integrating LLMs. Therefore, leaders must be prepared not only to answer wide-ranging questions about the value of these integrations, but also to offer tailored technological orientation programs to benefit each department.

When employees at every level understand the relevance of their skills and experience to the LLM development process, they are more likely to come forth with valuable inputs and suggestions. This broadens the scope for enterprise-wide innovation and industry disruption.

In conclusion, I see a bright future for organizations that use domain-specific LLMs to better extract and leverage the existing wealth of industry knowledge among their employees, while also creating challenging new tech roles and upskilling opportunities.

Forbes Finance Council is an invitation-only organization for executives in successful accounting, financial planning and wealth management firms. Do I qualify?

Source link

About The Author

Scroll to Top