Everyday in insurance is a journey of discovery
Feedback loops, data quality, ethics and ‘reinventing insurance’ were all of the topics discussed at the closing panel Getting practice with AI – Laying the foundations of the intelligent insurer at the close of day two of InsurTech Rising in London.
Right from the start the panel, which included Nikhil Kaithwal, group chief analytics officer at Tryg, Michael Natusch, global head of artificial intelligence at Prudential and Babak Ahmad, CEO at insurers.ai, outlined the complex and labour intensive journey most insurers need to take to start the process of including machine learning into an AI analytical layer.
“We are on a journey with AI,” said Kaithwal, “a lot of our resources are dedicated to getting the data flow right from our legacy systems. There is a lot of validating.”
Not only must the data be validated or ‘cleaned’ in order to populate the analytical engines, each set of data must be timely and properly labelled with the event a business wants to predict, adds Natusch.
Each process of cleaning and labelling needs to be paired with the correct set of questions before the “feedback loop” of machine learning on top of AI can be complete, agrees Ahman. “Even if you have good data – it may not be the right data to answer your questions.”
But what is the purpose of AI and machine learning in insurance? Moderator Chris Sandilands, partner at Oxbow Partners pressed the panel to give relevant examples. “What does it unlock,” he asked?
Kaithwal pointed to work at Tryg where the insurer is piloting the use of AI in conjunction with its Salesforce-based sales pipeline. Not only did the pilot signpost ‘hot leads’ for the sales people to follow, the system also paired certain leads with specific sales people who would be best placed to make the deal. The results of the pilot saw conversation rates increase by “three to four times”, he said.
However, questions around inherent biases in AI were also discussed focusing on the ethical implications. If AI were to target certain groups of people, for example young people when marketing car insurance, as too expensive – what is the ethical implications of pricing certain people out of the market, asked Sandilands?
Natusch veered away from questions around ethics and instead focused on creativity. He said:
The reason we are pricing people out of market is because we are lacking the creativity to find a solution to include those people.
Questions about inherent biases around sex or race, which may find themselves embedded in AI, was also given a practical answer. Both Natusch and Kaitwal confirmed that an AI algorithm is not built and then set to run without monitoring. Algos are constantly monitored to look for and correct any inherent biases. At Tryg there is an independent body that conducts these regular audits, says Kaithwal.
“As soon as we start trading with historical data we will be susceptible to historical biases,” said Natusch. Because of this a team running an AI-based model need to be active to combat those historical biases. “No model is just build and then runs. All models need to be reviewed – not just for ethical reasons, but for business reasons.”
The next few years the panel predicts a world where ‘AI’ is no longer on a job title, but is instead viewed as a commodity, like electricity. The end goal, commented Natusch is to ultimately “reinventing insurance.”
