Artificial intelligence (AI) should be used to “connect the dots” between different data silos and decisions made within an insurance company, Charles Dugas, head of insurance at Element AI, said last week.
The goal is to develop a system that builds a layer in between systems of engagement and systems of record. Systems of engagement are used to collect data from clients or other third parties (such as web apps); systems of record are where employees code the information into claims or policy systems.
“The idea is to develop this layer of intelligence in between [both systems] that processes the incoming information, automates some of the decisions, or supports the other decisions that need to be supported by your employees,” he said on Apr. 16. Dugas was the keynote speaker at the Centre for Study of Insurance Operation’s (CSIO) 2019 Member’s Meeting & Reception in Toronto.
“Just as you want to break down data silos and connect the dots between the different data silos, you want to make sure you connect the dots between different decisions that are made.”
For example, an auto claim that appears fraudulent may indicate that the company should not cross-sell other lines of business to that same client. “So, that is opportunity for you to connect the dots between different areas of the company where decisions are being made.”
For companies looking to develop a strategic approach to AI, a good starting point is to define goals. For example, a chief marketing officer may want to use AI to be able to better engage with clients and understand their needs on a digital platform. Or a chief underwriting officer may want to gain some operational efficiencies. To do this, they would need to build a list of opportunities for AI applications.
“The next question is, how do you rank them?” Dugas said. Element AI uses the “DVF framework,” which stands for desirability, viability and feasibility. “Desirability is the use case has been expressed by a key stakeholder, so something that is important to the company,” Dugas explained. “Viability refers to the return on investment – the benefit of delivering this solution. And feasibility refers to the technological feasibility of the solution.
“If you have all three… then you have a good candidate for an AI application.”
As an underwriting example, the initial portion may be to process submissions that are sent to underwriting, process the data and then segment that business and assign it to the right underwriter. A company may also want an “AI administrator,” who is able to go back to historical data. That allows for traceable AI and “observable and explainable decisions made by the AI software,” Dugas said. The AI system may even refer its decisions to a human being to review the information if it’s not fully confident in the decision it made.
“What’s important with these AI products is to bring the ‘human-in-the-loop’ into the process,” Dugas said. “So you have a human that is tasked with validating decisions that have been made, and curating the AI, and providing feedback to the AI. Over time, the software will learn through interacting with the employees, and that’s how you will capture within your system of intelligence an increasing amount of knowledge and develop your system of intelligence to become more performant.”
One audience member asked how far out until AI is commonplace in insurance. “I think everyone should be within a couple of years of implementing some use cases of AI,” Dugas said.