Unleashing the Power of Generative AI in Insurance – expert.ai

Key Takeaways from Our Webinar with Forrester

In the insurance world, AI, and generative AI specifically, is rapidly becoming a C-suite topic. In order to navigate the complexities of these new capabilities while capturing their benefits, it’s important for teams to understand the playing field so that they can move safely and effectively.

We recently hosted an exclusive conversation with Forrester Research insurance expert Indranil Bandyopadhyay, Principal Analyst, Insurance, to get the analyst’s perspective on generative AI in the insurance industry. In this excerpt, we are sharing some of the most important takeaways of the discussion, from current industry trends to the practical considerations for insurers in moving from experimentation to implementation.

Insurance Industry Dynamics

Before we delve into the intricacies of generative AI itself, it’s important to understand the macro trends shaping the industry.

Forrester reports that both revenue and costs are under pressure, making efficiency and effectiveness, which can positively impact profitability, a critical area of focus for insurers.

These aren’t the only factors affecting profitability more broadly. Forrester Research cites six main factors that will impact insurer profitability over the next decade: unpredictable geopolitics, economic challenges, regulatory pressures, climate change, higher customer expectations for technology, and an estimated $3 trillion protection gap. These areas will influence the business model, products and processes for insurers for many years to come.

The State of Generative AI in Insurance

Today, the insurance industry is at a pivotal point in its relationship with generative AI (GenAI). While a recent Forrester survey acknowledges the revenue growth and operational efficiency potential of GenAI, in reality, technology functionality is leading the GenAI journey in most cases, where it is providing value for functions like coding and testing.

Despite recognizing the potential gains, only a fraction of insurance companies have specific policies for generative AI, with just over half of them planning to update those policies. This underscores a critical imperative for alignment between generative AI initiatives and overall organizational strategies. Currently, most companies are in the exploration and experimentation stages with GenAI, and deployment is largely confined to departmental domains and not yet pervasive across the organization.

Commercially available foundational models and embedded AI are the primary tools being used by insurers, and the main use cases driving adoption over the next year will be those for knowledge management and self-service capabilities for data and analytics. While spending on GenAI is expected to increase, the investment currently ranges between $100k to $500k USD.

Finally, the business concerns around GenAI are significant. The leading issues—privacy and data protection, job loss and trust in GenAI outputs—are at odds with leadership who are eager to launch customer-facing solutions. And while companies are familiar with the use cases for GenAI, nearly all of them depend on external capabilities, which underscores the importance of partnerships for supporting emerging technologies.

Use Cases: Where GenAI Adds Value

While the use cases span the enterprise, from marketing and IT to operations, the true litmus test for GenAI lies in its ability to address concrete business challenges. Bandyopadhyay says that it’s important that companies recognize that GenAI cannot solve a problem on its own, but it can augment a solution, depending on the use case. This is why it’s critical to clearly understand the problem you are trying to solve.

Forrester identifies four general categories of use cases for insurers:

1. Data use cases can help generate, translate or correct messaging, in multiple languages, provide summarization or remove manual overhead.
2. Conversational tasks benefit from more human-like interactions for internal processes, and once tested, rolled out to external applications.
3. In knowledge management, Gen AI can be used to institutionalize knowledge and make it easy to understand. This is particularly useful when agents need quick and accurate information on the policy guidelines for a certain state or for synthesizing extensive research, for example.
4. Across various lifecycle areas, Generative AI enhances tasks like claims management, loss assessment, report generation, and customer communications.

For companies beginning their exploration, Bandyopadhyay suggested three initial areas:

• Claims operations and customer experience: This is where several insurers are using GenAI to optimize worker compensation claims through medical record digestion and summarization and bring it all together to make claims operations faster.
• Customer service: Augment current capabilities internally for employees to help customers in a much more targeted, faster and human way.
• Agents, brokers and underwriting: Link with OCR and other connected systems so that agents and underwriters have a much better understanding of claim and risks.

Build vs. Buy Dilemma: Charting the Path Forward

The age-old question of build vs. buy is even more critical in the context of generative AI capabilities.

Why? For one, generative AI is a relatively new technology and one that is evolving quickly. It’s well known that creating a large model language (LLM) is expensive, both in terms of the upfront costs and for the ongoing investments in training and fine tuning that will be necessary to get results you can count on. Specialized in-house staff will be needed for both. Finally, as regulations continue to evolve at the global and industry level, companies cannot risk non compliance.

It’s worth noting that the expertise needed to put an LLM in place is often underestimated. In addition to the categories mentioned above, further advanced skill sets for developing and integrating the applications will be required. In all, the build path quickly becomes both a cost- and labor-intensive approach. And, while a point approach may seem like an easy solution to a very specific problem right now, the siloed approach will make integration, data sharing and management difficult down the road.

Instead of investing in building, the path forward is clear: opt for the buy route. By entrusting major areas like infrastructure development and model refinement aspects to trusted, experienced partners, insurers can concentrate on honing the problems and use cases for their unique business requirements.

The Last Word

As an AI company doing business with insurers and a wide range of industry sectors, we see this latest cycle as just another innovation in the evolution of AI technology.

One of the most exciting developments around LLMs and GenAI is the focus that this has brought to a very important challenge: artificial intelligence applied to language data. This is a particular challenge that our company has focused on for more than 30 years.

While using an LLM or employing GenAI are not mandatory for success in AI, there are methods to utilize any AI technology securely.

We take an engineered approach, working alongside our customers as partners to configure the specific capabilities necessary to meet their technical and business requirements. This means that our solutions are configured to support the specific tasks that make up a typical workflow pipeline. We choose the best AI method fit for the data, the task or the technical success metrics required around explainability, scalability and accuracy. Explainability is core to our methodology, and it’s increasingly an area that regulators are looking into to make sure you have visibility around how your AI capabilities are making decisions.

It’s important to remember that GenAI is still a relatively new technology that is evolving quickly, and there is a still a huge amount of complexity in getting it right and deployed correctly the first time. We give insurers the ability to access generative AI capabilities in a safe, secure and cost-effective way, with the domain-specific insurance knowledge that most complex insurance processes require.