Earlier this month, TD Insurance released its first client-facing generative AI chatbot, the TDI Virtual Assistant, to help clients find answers through natural-language conversations.
The virtual assistant retrieves and summarizes information from the TD Insurance website for home, auto and small business insurance to answer general insurance queries. In a conversational tone, it can answer questions on topics such as accident benefits coverage and documentation required to obtain car insurance.
TD started with those three lines of business because they have the largest volumes in its insurance segment, said Kristen Gill, vice president and executive journey product owner at TD Insurance. The insurer plans to expand the chatbot to life and health insurance at a later stage.
TDI Virtual Assistant took about a year to build, said Christopher Cooney, vice president of analytics and modelling at TD Insurance. Its development involved the technology solutions team, insurance experts, lawyers, and Layer 6, TD’s AI research and development centre.
Although Layer 6 had experience developing internal tools at TD, building an external-facing AI posed unique challenges, Cooney added. To mitigate the risk of hallucinations, the chatbot uses retrieval-augmented generation technology, an AI framework that improves large language model accuracy by retrieving data only from trusted content libraries. In this case, TDI Virtual Assistant can only reference material already on TDI’s website.
Another guardrail was to get the tone of voice right when interacting with customers, Cooney said. Users can tell TDI what’s working and not working by clicking the thumbs up or down feedback icon on the chatbot’s responses.
Also in the news: What U.S. budget cuts mean for NatCat forecasts
In its early days, humans will closely monitor the chatbot’s performance to ensure it comes up with satisfactory answers, Gill said. But the goal is to move to machine-assisted monitoring, in which AI helps humans supervise the AI chatbot.
For now, the chatbot is unauthenticated, meaning users don’t need to log in, so it can only answer general queries, Gill explained. If customers need more personalized advice, such as updating their policy details, the chatbot will encourage them to call for assistance.
Eventually, TDI intends to provide an authenticated chatbot service, allowing customers to get personalized advice, Gill said. The technology is still in its infancy, and the insurer’s compliance department is exploring what kinds of changes AI would be allowed to make from a regulatory perspective.
In the future, other client-facing AI applications could include using natural language to help a customer obtain insurance quotes and report claims, Gill added. “We will want to build that capability with the right processes and guardrails and controls to make sure that customer information is safe.”
Special to Canadian Underwriter from Jonathan Got, a reporter with Advisor.ca and Investment Executive.