AI’s Still a Work in Progress for Property Underwriters. Here’s Why the Human Touch Won’t Go Away

Collisions with insurance company legacy systems is just one reason why the use of artificial intelligence in property underwriting has not advanced as far as some would like.
By: | July 18, 2020

Not very long ago it seemed as if artificial intelligence (AI) was going to take over property underwriting.

Big carriers were buying commercial systems from service providers that were proliferating, or even investing directly in such insuretech firms. Those practices continue, but there is new understanding that the process is more evolution than revolution.

“Technology needs to be useful, not just novel,” said Tony Fenton, vice president of commercial lines digitization at Nationwide, which has dedicated $100 million to AI, and so far has three partners in property underwriting.

“We consider AI to be an enabler,” Fenton continued. “That is a large and complex effort … We have been working at tech transformation for several years and are using Guidewire to get some of these systems to consume some of the same data. It is a multi-year process to optimize and customize.”

AI is Becoming More Effective and Efficient

Senior executives at several carriers said their AI efforts are improving efficiency and effectiveness. And, for each implementation, new questions are raised about the processes themselves.

“AI is clearly an enhancement for underwriting,” said Michael LaRocca, head of general property for North America at Swiss Re Corporate Solutions (SRCS).

“We see it all over the lines we write, more or less in different uses. In property we take in so much data, that everyone started with ways to process unstructured data – documents, spreadsheets, portable document format files. We can only advance if we understand data; what do we have, and what do we need.”

Given the sheer volume of data in property underwriting, LaRocca said the first benefit of AI is triage. “The more efficient we can be the faster we can find information that is missing or suspicious. We can also review contracts with machine learning for block finding of keywords. That can help with claims processing,” he said.

Some carriers have made equity investments in insuretech startups, but Swiss Re has followed a more conventional partnership approach. “When we believe in an idea, we prefer to use the R&D capabilities of the Swiss Re Institute and form a partnership with a tech firm or a university,” LaRocca said.

“These collaborations deliver greater long-term benefit for us and our customers.”

In March SRCS formed a strategic alliance with Microsoft. The focus of the alliance is Swiss Re’s Digital Market Center, which will help develop large-scale predictive and risk-management tools. The center will draw n Microsoft’s Azure cloud technologies, Internet of Things and AI capabilities. The first areas of application are planned to be connected vehicles and mobility, industrial manufacturing (Industry 4.0), and natural catastrophe resilience.

Harish Neelamana
Co-founder and President

Swiss Re’s work in this area will go beyond new coverage to provide broader risk insights for complex, interconnected systems. For example, risk managers can get a greater understanding of how the loss of a ship’s cargo may affect supply chains, or how natural catastrophes will affect a government’s infrastructure investments.

In May 2019 SRCS formed a collaboration with Airbus Aerial today announced a collaboration to help companies identify flood risks and predict weather-related flood damage. Under the collaboration, SRCS’s flood assessment tool, Float, will use Airbus Aerial’s drone imagery and data.

Among Swiss Re’s more innovative internal work is the insurability of AI. “We look at algorithmic risk,” said LaRocca. “As our clients operate and make decisions based on AI, we are looking into the insurable risk there. There is still a tangible piece there, with humans making the decisions but we are looking at faster time frames. This is a big initiative, and there is more to come.”

Investing in Companies That Provide AI Services

Direct investment by carriers in companies providing AI services has yielded mixed results. Early in June Nationwide made an investment in Planck, a company that uses artificial intelligence (AI) to help insurers increase premiums while reducing loss and expense ratios.

In early May, AIG disclosed as part of its first-quarter report that at the end of March it decided to place Blackboard U.S. Holdings, its technology-driven subsidiary, into run-off, and took a pre-tax loss of $210 million.  The company declined to provide any further information about the decision or disposition of Blackboard.

In July 2019 FM Global invested $1 million in RiskGenius, an insurtech startup that applies AI and machine learning to automate underwriting. It is the largest investment FM Global has made in a startup to date, and followed $250,000 that FM Global provided to AirWorks, an aerial mapping and surveying startup.

In November 2019 National American Insurance Co. selected Convr, known as DataCubes at the time, as its AI service provider for P&C underwriting. “From a business perspective it is fair to say that AI in frontline underwriting is still a work in progress,” said Harish Neelamana, co-founder and president of Convr.

Human Touch Isn’t Going Away

In the early days of AI there was an eagerness to adopt the technology, but not always a clear understanding of what it could do, and how it worked.

There is more convergence these days, but some service providers still say that on occasion they have to modify or rarely even decline projects because upon closer examination, what the underwriter requested what different from what it actually needed.

“From a business perspective it is fair to say that AI in frontline underwriting is still a work in progress.”— Harish Neelamana, co-founder and president, Convr

As AI evolves, “the role it can play in different segments of commercial underwriting will evolve with it,” said Neelamana.

“At its current maturity, AI is best suited for decision-making based on quantifiable inputs in problems that are complicated yet discretely defined. That would include tasks such as digitizing submission documents from agents, complimenting that information with third party data, automatically answering underwriting questions and prequalifying risks using a score.”

In cases where the situation involves problems that are less defined or are multi-dimensional, “the human mind continues to excel,” Neelamana added.

“Underwriters are best suited to use AI for mundane tasks while using most of their time to evaluate more complicated risks, manage agent relationships, and juggle multiple priorities to achieve the carrier’s underwriting goals.”

Tony Fenton
V.P. of Commercial Digitization

Michael Lebovitz, senior vice president of innovation at FM Global, concurred that “established insurance companies have legacy systems and procedures long in place, while startups have a high level of focus and fresh energy. We are learning together” how to adopt and adapt AI most effectively for property underwriting.

At the most basic level, the FM Global research team is collaborating with the tech partners in their development.

“We are also looking for practical applications for our processes,”  Lebovitz said. “That is the goal of a partner service providers. And we are also helping them to shape their processes.”

“We view AI as augmenting, not replacing, our underwriting process,” said Jonathan Salter, head of risk engineering for AXA-XL.

“We have an expert team of risk and account consultants who would never be replaced.” Quite to the contrary Salter stressed, “you need to take your best people and dedicate them to the project. AI is like a trainee. The better you train it, the better it will work.”

Such training requires time and effort from people who are closest to the process. Invariably it becomes something of a burden for them to do their jobs, and also to train an assistant at the same time. Beneath that there may be some resistance to technology or even the fear of replacement.

“We got some pushback,” noted Scott A. Ewing, Americas property risk engineering leader at AXA XL. “People are busy. But we explained that it’s like your 401(k), you have to invest a little now and a little later to have something to withdraw at the end.”

As an illustration he noted that once a document had been processed by AI, the tool has hyperlinks in the summary report “so that a risk engineer can dig back and agree or not,” said Ewing. “As our people do that, the tool will continue to improve.”

Salter added, “recently I uploaded 50 location reports into our Spotlight system on a Friday for a presentation on Monday. The results were ready in seven minutes. That is something that would have taken most of the day, if not more without AI. Accuracy is about 80-90% for normal risks. It can flag sites that we need to go back to.” &

Gregory DL Morris is an independent business journalist currently based in New York with 25 years’ experience in industry, energy, finance and transportation. He can be reached at [email protected].

More from Risk & Insurance

More from Risk & Insurance