Forrester analyst says insurers should wait for Gen AI competition to play out

Adoption of generative AI by insurance companies in 2024 may not happen as fast as some think, according to Indranil Bandyopadhyay, principal analyst at Forrester, and author of "Generative AI: What It Means For Insurance," a trends report from the consulting firm.

Carriers should wait out the competition between different Gen AI initiatives including OpenAI's ChatGPT and Google's Gemini, Bandyopadhyay said. "There's a huge competition among the players. I would rather let that settle down, give it another 12 to 18 months," he said. "That would be a more pragmatic approach, given where the insurance industry is."

While Bandyopadhyay sees potential for Gen AI to improve claims operations, customer service capabilities and insurance companies' own internal productivity, he says there will still need to be human oversight. Insurance brokers who work directly with insureds could use Gen AI to augment their work but not replace it.

"If a generative AI powered solution can quickly give answers, it is not going to change the broker," he said. "The customer is sitting across the table for a reason. That reason is to talk to a human being. Generative AI is not going to change the equation in this context. Generative AI is going to make the broker much more productive, with much more depth in understanding the customer."

Insurers are conscious of issues with Gen AI "hallucinations." Bandyopadhyay points to retrieval-augmented generation (RAG) as a means to solve them. RAG sets parameters or limits for a large language model's (LLM) response to questions or commands. If Gen AI is asked to compare two companies profitability, for example, it would retrieve their annual reports and summarize the relevant elements, Bandyopadhyay explained.

"RAG is actually controlling more around the variability of the content of the answer, and using LLM to make it," he said. 

Bandyopadhyay's report says that Gen AI does have potential to improve claims operations and experience by "summarizing documents, extracting keywords, responding to specific inquiries, and alerting loss adjusters to overseen items." LLMs can help customer service agents resolve issues more efficiently, he adds, and help agents, brokers and underwriters be more productive. LLMs do this by summarizing knowledge and synthesizing data to produce quick and accurate answers to professionals' questions, according to Bandyopadhyay.

Similarly, Gen AI can summarize information, produce contextual insights and make it easier to manage knowledge and information through its conversational capabilities, according to Bandyopadhyay.

Before proceeding with adopting Gen AI technology, insurers should first identify appropriate use cases, use publicly available Gen AI models, and prepare their data architecture, Bandyopadhyay wrote in the report. 

Not every insurer will need Gen AI, he said. "It can augment or make some use cases better," such as customer service chatbots, he said. "Generative AI can now curate those responses in a meaningful way. Now, suddenly, chatbots become usable and a more augmented solution."

Insurers should think of Gen AI as the "cherry on top of dessert," Bandyopadhyay said. "If you haven't baked the cake, just the cherry is not going to do the job for you. … Look at your problem statement. Look at your opportunity statement. Don't force feed a generative AI because everybody's talking about it."