The use of a chatbot in car insurance offers numerous benefits, according to a 2025 study conducted by David Beauchemin, co-founder and CEO of the artificial intelligence (AI) consulting firm Baseline, as part of his doctoral thesis in computer science.
According to the AI expert, the chatbot he developed does more than just provide information: it reduces inequality in access to insurance expertise and benefits for those who need it most. It minimizes the mental effort required to understand complex insurance concepts.
Beauchemin chose the car insurance sector for his research project because its forms are publicly available, allowing him to publish his results without intellectual property concerns.
He outlined his project at Université Laval during a conference presented by the Laboratoire en droit des services financiers (Financial Services Law Laboratory) entitled "Between Consumer Autonomy and Vulnerability: The Chatbot as a Substitute for Auto Insurance Advice?"
In an interview with the Insurance Portal, Beauchemin explained that he wants to contribute to the development of a chatbot in the insurance sector, with the aim of protecting consumers and reducing costs for insurers. He plans to make his chatbot available for auto insurance in 2026 and mentioned that the technology could be adapted to the home insurance sector.
Insurers have platforms that meet some of the needs, but he finds them complex and static, and believes they offer a customer journey that is unsuitable for certain situations.
A chatbot, he explains, could be personalized for each user with a tailored experience. It would explain a policy much like a human would, but without being a complete replacement for an agent. Anyway, he added, there will always be people who don't want to use a robot.
From shopping for a policy to closing the sale
What the research revealed, adds its author, is that people could use it at various stages of purchasing insurance: shopping for a policy, getting information, updating it, adding a product, and even closing the sale.
But for now, the tool developed by Beauchemin doesn't yet go so far as to recommend a specific product, policy, or even insurer. He says that will be the next step – a second solution that would be able to compare products and suggest the most suitable one for the user.
"But we could get there if we allow ourselves to dream a little," he adds.
Augmented generation through data retrieval
According to Beauchemin, consumer-grade generative AI (ChatGPT, Copilot, Gemini) is unsuitable for answering questions related to car insurance, particularly because it lacks access to all Quebec documents and it is more difficult to precisely trace the sources of information that led to an answer. He therefore chose retrieval augmented generation (RAG) as a solution, which relies on a targeted document database.
Subsequently, he developed a questionnaire, which 154 people completed, to draw initial conclusions from his study. He is aware, however, that his sample size is small, and he now wants the questionnaire to be completed by at least 3,000 people to obtain a more representative sample of users.
Key findings
More than 90% of participants felt that his AI could increase their autonomy when purchasing, renewing, or filing a claim. His tool, he observed, has the ability to make contracts easier to understand. He calls it a "cognitive equalizer." Thanks to his bot, people were able to reach a level of parity with insurance experts, with no correlation to cognitive effort.
"One in two people didn't feel embarrassed to ask difficult questions because they didn't sense any human judgment behind the process," he stated. "That was very revealing: someone said they were able to take the time to ask questions and understand what was going on."
The system, he notes, is very relevant for a certain segment of the population. Users with low financial literacy report significantly higher satisfaction and benefits. However, the gains were much lower for those with a good understanding of the insurance industry.
Furthermore, respondents expressed a strong preference for speaking to a human agent in high-stakes scenarios, such as filing a claim, or in emotionally charged situations, for example, when someone has died and the beneficiary wants to collect the life insurance.
"There needs to be a mechanism so that the user can say, 'I want to speak to a human,'" underlines Beauchemin.
The chatbot's weaknesses
But such a chatbot also has weaknesses. It can make mistakes, and there can be contradictions in its responses. Another aspect: a chatbot lacks empathy. And as Beauchemin points out, what AI delivers to the consumer is dictated by who is behind the tool.
“There are companies that can do what I’ve done, so they can sell it. This represents a huge challenge for consumer protection,” he says. “The problem for me is that these players’ incentives will be more about making money than protecting the consumer.”
Who will develop this chatbot?
Who could develop such a chatbot with the objective of protecting consumers and reducing insurers’ costs, as he envisions? All insurers have seen the potential of such a chatbot, and many have tried to develop one, says Beauchemin. To date, however, as far as he knows, the results have been mixed.
He believes the best player would be a regulator, such as the Autorité des marchés financiers (AMF) in Quebec, which wouldn't be trying to sell a specific product to users.
But ideally, he adds, it would be an association of insurers joined by the regulator. His company, Baseline, would be interested in developing this tool with other stakeholders, but not in ensuring its governance and positioning.
His thesis, he concludes, demonstrates that AI can make insurance contracts understandable, reduce cognitive inequalities, and promote customer autonomy.
