They’re now constantly in the media and opinion is clearly divided. So what about chatbots in insurance – pro or con?
By Prateek Didwania.
In November 2022, ChatGPT was made public by the artificial intelligence (AI) research laboratory OpenAI; by January 2023, 100 million people were using it each month. Both OpenAI and its competitors are making continuous advancements in the field. This rate of adoption and innovation could bring about a new way for people to interact with information and change customers’ relationships with companies when buying goods and services. Naturally, insurance would not be immune. But how exactly could this technology impact the industry?
Chatbots are built on large language models (LLMs): advanced machine-learning models that are trained on massive volumes of text to comprehend and generate natural language. They can then be finetuned to perform a variety of tasks, such as text classification, analysis and generation. LLMs are the reason for the rapid improvement in chatbots such as ChatGPT, which can both understand and communicate language on a par with humans.
One way chatbots could impact the insurance industry is if they become an alternative vehicle through which consumers can learn about, research and buy insurance. According to a 2016 study by market research firm Growth From Knowledge, eight in 10 people research insurance online. Insurers are aware of this fact and ensure that their products are seen on price comparison sites, search engine rankings and so on.
Chatbots could be an attractive alternative for people who are conducting this research. Instead of a consumer having to manually search for and parse a large amount of complicated information, they could use a competent chatbot to simplify the research, comparison and purchase process. Such chatbots are already being implemented in the consumer goods industry: online ecommerce platform Shopify has released an OpenAI-powered shopping assistant that can guide customers to make a pertinent purchase. It asks about a buyer’s specific needs, tailors its suggestions, answers customer queries about product specifications and compares them with alternatives.
A chatbot tailored for insurance could help customers filter through the many available insurance products. It could help them understand what product is best for their needs, based on factors such as the risks they want covered and their budget. It could also help consumers understand policy descriptions, conditions, and any legal and technical jargon. In essence, a chatbot could be a pseudo-financial adviser or insurance agent, able to help not just one consumer, but many at scale. This would be great for customers and for insurance companies. If successful, chatbots could increase efficiency by reducing costs and increasing margins, potentially leading to cheaper policies for customers. Chatbots and LLM technologies allow fewer people to help the same number of customers – decreasing staffing costs or enabling staff to efficiently reach out and help more potential customers. US insurer Lemonade already processes customers’ texts using AI during its fraud-detection segment. With the help of LLMs, this process could lead to more potential savings.
Finally, insurers could leverage their access to a large database of consumer conversations to develop better insights into what customers need and want. This could be used to develop new products, better marketing and competitive pricing.
These changes could bring about huge benefits for both consumers and insurance companies – but there are several problems that the industry and regulators will need to manage while developing and regulating the technology for the insurance industry.
The biggest problem with LLMs is their accuracy. Even Google, on launching its own chatbot, Bard, admitted that it can convey misinformation. This is a huge issue when selling financial products, which are extensively regulated: it is ethically wrong. It also brings about the important question of liability. Who would be responsible for a misselling incident? This is a difficult legal question, and companies and regulators will have to work together to ensure consumers are protected. Data protection rights will also have to be considered. How do current data protection laws affect the use of LLMs? Will companies be allowed to store chat transcripts? Who can access them and for what purposes? Can models be trained and refined, based on customer data? How should model bias be managed?
LLMs could greatly improve the customer experience when purchasing and dealing with insurance, and lead to great efficiencies in the industry. It will be exciting to see how it unfolds.
Prateek Didwania is 24 and based in Munich. He has been studying for his actuarial exams since 2021. He hopes to qualify by 2024 and go into pricing and product development
