Sompo Holdings’s new head of digital for Asia, Patrick Chin, says the Japanese insurance company is poised to revamp its entire operating model around artificial intelligence-based processes.
He is pushing internally to bring to Asia ex-Japan the level of tech-led change that is already taking place in Sompo’s domestic market.
He told DigFin on the sidelines of a recent conference in Hong Kong that this could ultimately redesign the entire process, even automating claims, by putting the customer at the center of its business model, while orienting staff to create new services.
But to make this happen, Sompo Asia has to go beyond cherry-picking A.I. tools for niche activities.
“Our aspiration is to be a top digitally competitive insurance company, able to respond continuously to customer needs, through technology,” Chin said during a conference presentation. “To be digitally competitive means to be able to collaborate seamlessly with [companies like Alibaba, Amazon or Facebook].”
Big in Japan
Chin joined Sompo in Singapore earlier this year from MetLife. Sompo is the largest non-life insurer in Japan, with $25.8 billion in net written premiums, $112.5 billion in total assets, and 20 million customers. The business is almost entirely Japanese. Therefore in the rest of Asia, Sompo has less to lose, and more to gain on rivals, by adopting a more radical digital strategy.
Sompo Holdings is already a stakeholder in insurtech companies such as Trov and Slice, respectively San Francisco- and New York-based providers of on-demand insurance. The insurer will launch its own on-demand products soon, including for travel delays and for cyber crimes, in which the claims are also automated.
In Japan, Sompo has already created a data lake and opened internally a sandbox for various A.I. projects. Chin is now doing the same for the rest of Asia, so the firm can begin pilots.
“There are many A.I. solutions, and if you pick the wrong one, you can’t end it so easily,” Chin said. “So we have to be clear about what A.I. is doing.”
The focus is on analytics that predict and learn. These provide a variety of insights: into customer behavior, agents’ needs, product and policy development, and into the claims process. In claims alone, A.I. tools can help insurers improve their reserves forecasting and fraud detection, which can have an immediate impact on the bottom line, especially in the short-term world of general insurance.
Chin says for any of this to be effective, however, the insurer needs access to data: around pricing, sales, distribution, claims, policy administration, and customers. This last one is the most difficult to get, because intermediaries such as agents or brokers own the customer relationship – hence the need for the data lake first.
But the firm has already launched its first A.I. pilots in Japan: automated underwriting, and fraud detection.
For underwriting, Sompo Japan built an app for agents to scan a customer’s insurance history to generate pricing offers on policy renewals. The idea is to give agents digital tools to help them sell, while ensuring they find it most convenient to sell Sompo products. Chin says this has been a success in Japan, helping grow Sompo’s agency force.
For fraud detection, Sompo Japan is in a pilot using both internal and third-party technology, to try to have an industry-wide view of fraudulent claims. The company has found that it is easier to spot fraud using supervised A.I. (when data scientists feed machines both data inputs and expected outputs, thus “teaching” the algorithm how to conduct the function in question).
But, says Chin, this can only take insurers so far: “If you don’t know which claims are fraudulent, you need unsupervised modeling” coupled with an experienced human expert to give the machine hints. For example, no machine would, on its own, flag a perfectly completed claims form. But in the real world, it turns out that perfectly completed forms are a hallmark of fraud.
Tasks like fraud detection get more complicated. This is the sort of area where insurers would do better if they collaborate. For example, a person or a group of conspirators may commit fraud among several insurers, none of which will flag it until it’s too late – but an industry-wide view would detect such activity.
But this takes technology back into the world of business politics. This is just one example where digitization can help save costs or grow revenue, but it needs to understood in the context of the entire insurance value chain.
“We need to learn as we go,” Chin said. “It’s not an exact science; it’s intuitive. But ultimately we can redesign the entire process, and use A.I. to create new products and services.”