Data: Underwriting
Data Driven

• Kate Smith
• May 2019
• 

Key Points

• Clear Picture: Underwriters are gaining more dynamic views of risk through new sources of data and advanced analytics.
• Model Client: By showing underwriters which policies will be more profitable, predictive models allow for more risk-based pricing.
• Reality Check: Overreliance on new tools is dangerous because results are only as good as the data being used, and experts say data is still poor.

Having grown up in Pennsylvania, Kassie Bryan knows well that one of the biggest road hazards in the state has four legs.

“I saw so many deer accidents growing up there,” Bryan, head of P&C Solutions for Swiss Re, said.

So it made perfect sense to Bryan when Swiss Re incorporated deer population density data into its Motor Market Analyzer, a predictive model that uses granular data sets representing accident risk factors to determine accident frequency and severity by geographic area.

“We did a project for a client who was looking to grow into the state of Pennsylvania,” Bryan said. “They had no historical loss data on auto in Pennsylvania, and they wanted to understand what the drivers of loss are and what they should be thinking about when they enter the state. One of the external data sets we used was deer population density.

“When I heard we were using deer population density, I thought, 'That makes so much sense. That's absolutely a driver of the risk.'”

Advanced analytics enable insurers to paint a dynamic, and real-time, view of risk. And at the heart of analytics lies data.

Insurers are accessing a wealth of traditional and nontraditional data sources—from credit ratings and motor vehicle records to Yelp reviews and, in Swiss Re's case, deer population density—to inform decisions, improve pricing and increase efficiency.

When combined with advanced technologies and new modeling capabilities, data becomes an exponentially powerful tool that enhances an insurer's understanding of risks.

“You're seeing a move from static information that gets updated once a year or less, to a dynamic view of businesses or consumers,” Kirstin Marr, president of Valen Analytics, said.

The implications for underwriting are significant. Not only are data and analytics enabling the automation of certain parts of the underwriting process, but as data becomes increasingly reliable experts say underwriting will become more of a science than an art.

“Today, underwriting is manual and highly experiential,” Ari Chester, a partner at McKinsey, said. “It's based on experience and hard-learned lessons from individual underwriters, who learn from apprenticeship. There are analytics and tools that are used, but they're often homemade, homegrown and inconsistent. And the data that is used is very often what is supplied in the submission, which is collected through a broker or agent from a client.

“In the future, when it's more science-based, the quality and collection of data will be more advanced. There will be more tools. There will be more rigor, more use of models and certainly less manual processing.”

While the use of advanced data analytics is in the nascent stage, experts say it's already having an impact on efficiency, as evidenced by the rise of digitally augmented underwriting.

“Parts of the underwriting process that can be automated are being automated,” Risa Ryan, head of strategy and analysis for Munich Reinsurance America, said. “And the data is being enhanced with data sources that go far beyond what we typically have used in the underwriting process.”

Predicting Profitability

Data and analytics are not new for underwriters. Actuarial analysis is grounded in those things.

“When you think about how data has informed underwriting, which is risk selection and pricing, it's largely been leveraged at the large scale by actuaries setting rates and selecting company tiers,” Marr said. “Those are things that are done once for the year and set.

“The other way underwriters have used data is by getting it from the application or by using manual reports. If it's commercial auto, they're pulling a motor vehicle report. They've done inspections. They've relied on agents to validate the business. Those are very expensive, very manual ways of acquiring and reviewing data.”

Ryan described that process as “inefficient.”

“Now information is being stored in a manner that is easily accessed and cheap,” Ryan said. “It's less expensive to access data than it used to be. And we also have the tools to access data; computers are so fast now. And we've got open-source tools that have democratized the accessibility and usability of data.”

The end result is an automated, real-time use of advanced data and analytics in the underwriting workflow.

“We have the ability to intake data in real time—transactional, primary sources of data,” Marr said.

Rather than relying on data aggregators like D&B or InfoUSA to validate a business, Marr said, underwriters can use nontraditional sources of data to build a more robust picture of a business.

“Today you can find a business on Google, you can see their Yelp reviews,” she said. “Because computing power has advanced so much, you can leverage the public databases that were really hard for companies to be able to use. You can go to NOAA and get GIS satellite data. There are free satellite data sources that are available, free weather sources that are available. There are all these primary sources that used to be managed by those large data aggregators, and they're more available now. You have the ability to manage that yourself, if you want. That's the more advanced use of data in terms of technology.

“It also allows you to access more transactional or behavioral data. By scraping Yelp and getting reviews of a business, you get a sense of sentiment. You get a sense of the quality of that business and how they serve their customers. You get things you'll never see from an InfoUSA or a D&B or one of the credit bureaus. There's a lot more you can learn that, if you're applying advanced analytics to it, can help you better assess that risk.”

Advanced data analytics are especially powerful for liability lines, Bryan said.

“Historical loss data for liability business is sparse,” she said. “Data emerges slowly and it quickly loses credibility because of the constantly changing risk environments. One of the ways we're addressing that is through forward-looking modeling, which includes an exposure-based model that anticipates and incorporates changes in the liability risk landscape into the quantitative assessment of risk. It uses external data, not directly insurance-related data, to describe the world that generates the accidents without having to wait for the accidents to happen.

“We all know that what happened in the past is not necessarily what will happen in the future,” Bryan added. “So advanced analytics lets us take advantage of real-time data and advanced computing power to model the exposures of the future and quantify the risks they will produce. We're forming a picture of the world that is causing the losses.”

On the property side, publicly available data sources give more information about particular risks.

“The industry has now got information about square footage of buildings, flights of stairs in buildings, what the roofs are made of, how many windows,” Ryan said. “From an aerial imagery standpoint, we can see the amount of open space around a building. We're actually looking at that information for underwriting on the liability side. What obstacles or hazards are around the perimeter of a building? Those are types of information, especially the pictures, that we didn't typically access in the underwriting process.”

Ryan said census data, credit scores, LexisNexis data, demographic data and financial information also can be valuable in building models.

“Those data sources have been in existence for a long time, but they have not been as easily accessible as they are now,” Ryan said. “We are constantly updating our models with the newest information available. That helps us inform our underwriting decisions here internally, but it also helps us build models that influence our clients' decisions.

“We've built a risk score model that helps our clients understand which policyholders will be more profitable than others.”

Marr said that's the beauty of predictive analytics.

“Instead of having to wait 18 months for a policy to mature in order to know if it was profitable or not, you can know in advance where your profitability is heading,” she said. “It allows carriers to align price to risk in a way that's really hard to do without predictive analytics. It allows you to do real risk-based pricing.”

Future of Underwriting

Another consequence of data and technology advancements is improved underwriting efficiency. Carriers are using digitally augmented underwriting to automate parts of the process.

This combination of human underwriters and artificial intelligence-based programs can take several forms, Jeff Heaton, vice president and data scientist at Reinsurance Group of America, said.

“On one end of the spectrum, it is AI programs performing the entire underwriting task and referring cases to human underwriters when the program is unsure,” Heaton said. “On the other end of the spectrum, it is simply using the AI program as an assistant to the human underwriter.”

Last October, RGA launched its “AI-Augmented Underwriting System” with the goal of creating greater efficiency.

“An underwriting file can contain a great deal of information—much of it duplicated,” Heaton said. “The augmented underwriting system being produced scans this underwriting file and identifies key components, such as individual health records, motor vehicle records, and all places that critical information about the insurance applicant are presented. These sections are compared and checked for consistency. This allows the underwriter to quickly navigate to the appropriate parts of the underwriting file.”

While increased profitability is the primary goal of advanced data and analytics, certain underwriting lines, such as small commercial, need the efficiency gains just as much.

“In the small and midsize commercial lines, you may have a loss ratio that's generally OK, but the challenge is someone is touching the policy too much,” Chester said. “You have a human doing a lot of underwriting and administration on a policy that's $2,000 or $5,000. So even if the accounts are profitable, the level of effort going into it is way too high.

“Of course there's relatively more value in the loss ratio. But really the analytics and the data are enabling a big expense improvement by allowing the underwriter to reduce manually underwriting the $5,000 account or the $1,000 account. Those can be mostly underwritten by algorithms. And then the underwriter of course, will review the portfolio, look for trends, monitor adequacy, and continuously contribute to recalibrating and updating the algorithms. But they're not actually sitting there and manually trying to assess and underwrite thousands of small accounts.”

Chester does not see automation or artificial intelligence replacing the human underwriter any time soon, particularly in commercial lines. But he does see the underwriter's role shifting in several ways.

“There will be less administrative and manual work and more thinking,” he said. “We like to compare it to a physician who has a nurse or a physician's assistant come in and check blood pressure, take your weight, capture your medical history. Doctors have become more focused on patient assessment and less on the prep. Underwriting is going to be similar to that.”

Chester also anticipates primary insurance underwriting will look more like reinsurance underwriting.

“Reinsurance is based on portfolio-based underwriting. It's looking at accounts in the aggregate,” Chester said. “Primary and individual risk insurance will become more like treaty reinsurance, where you're looking at every individual risk and underwriting it, but you're also looking, almost in real time and in an integrated way, across the portfolio. You're taking a dual approach to look at the risk and its place in the portfolio at the same time.”

The day will come when underwriting will be fully automated and computers, through machine learning, will be able to notice trends and update algorithms in real time. But that is still far off.

“As data gets better and as the algorithms are tested, there will be a time when decisions can be mostly, or even exclusively, data driven,” Chester said. “Whether it's a decade or two decades away, or more, it's certainly no sooner.”

Chester said commercial underwriters, at this stage, shouldn't be overly reliant on new tools and models. The art of underwriting, he said, is still at least as important as the science.

“It's important to not have too much faith and too much hope in what the tools and data can provide,” Chester said. “If there is a tool or model built that gives you an answer, some companies fall into the trap of giving that answer more credibility than it should have. They've taken the emotion and human judgment out of the equation. But the answer provided by the algorithm is not necessarily accurate. It may be that the human judgment and the imperfect, emotionally driven decision might in fact still be superior to what the algorithm can produce. The reason is, the output is only as good as the data, and the data quality is poor.

“Second, risks change and they change pretty frequently. Algorithms today do not have the sophistication to proactively learn and adapt until new loss experience is already apparent. Third, it's not like personal lines where there's a very high volume of mostly homogenous risks. Even in small commercial, you still have a lot of heterogeneous risk. So it's easier said than done to build these algorithms that can provide insight around a particular risk in a seamless way.”