Meet your new underwriting assistant: GenAI

Generative AI (GenAI) won’t replace underwriters, but you can train it to be your underwriting assistant, one AI expert shared at an industry event Tuesday. 

“In North America, there are quite a few insurers that are going through proof of concepts right now where they’re positioning [GenAI] as an underwriting assistant,” Martina Conlon, head of insurance insights at Datos Insights told Insurance Canada’s ICTF2024 Conference in Toronto. 

“Are they having ChatGPT or any of the large language models actually doing the underwriting? No. But that is certainly in the future for when these technologies are more stable and predictable.” 

Essentially, large language models (LLMs) are a type of GenAI that does text-generation (think: ChatGPT or Microsoft Copilot). And insurers are increasingly finding creative ways to use these LLMs in their operations. 

Conlon warned insurers should stick to completing more specialized tasks, like policy wording, while delegating more menial tasks to the AI. For example, AI is capable of prioritizing new business based on rules set by an insurer. 

Or, when trained in an insurer’s underwriting rules, appetite guidelines and product definitions, GenAI can be used to conduct loss runs and minor risk analysis and help complete insurance submissions. 

“For text processing, we gave it one of our clients’ loss runs,” Conlon said, “and we gave it instructions to create a JSON data structure…It mapped the data from the loss run into the JSON structure without any guidance.” 

As an example of minor risk analysis, Conlon said, “You can ask those questions of ‘what form should I attach for Uber drivers in Quebec?’…or ‘What is the riskiest location on this policy?’” 

Some AI can also be deployed to browse the websites of your clients businesses and gather information or make product recommendations. 

Conlon shared an example of a colleague who’d snapped a picture of a basement HVAC unit and asked ChatGPT to evaluate the image. The AI was able to read both hand-written and printed text in the photo and determine the HVAC’s installation date, for example.  

“One of the most interesting things is it noticed that there was no maintenance sticker on [the HVAC unit]. So it not just interpreted what was in the picture, but it interpreted what wasn’t in the picture.” 

There are other ways to apply GenAI across the P&C ecosystem. Claims is another insurance function that can reap benefits in areas such as: 

• Claims triage 
• Categorization and severity 
• Adjudication assistance 
• Insurance coverage guidance 
• Fraud detection 
• Notes analysis and summarization. 

GenAI is also capable of demonstrating significant insurance knowledge. In fact, one U.S.-based insurance software provider’s proprietary GenAI passed its insurance appraiser exams in New York, California, Texas and Florida.  

The AI even began correcting the tutors hired to train it, proving the tech is capable of making informed decisions on “regulatory guidelines, legal frameworks and workflows,” in the P&C space.  

And because these LLMs are constantly trained on new data, they develop new capabilities every month, Conlon said. 

That means there are more use cases to be seen on how GenAI might be applied in the P&C ecosystem in the near future.  

Feature image by iStock.com/Mikhail Seleznev