The Role of Analytics and Big Data in Insurance
Verisk Review recently spoke with industry leaders about what role insurance analytics and Big Data play, how predictive analytics is changing the actuarial function, and what areas of analytics are ripe for advancement.
G. Edward Combs, ARM, CPCU, is principal of G. Edward Combs Consulting, a California-based consulting firm. In addition to leading his own company, he serves as insurance adviser to Fractal Analytics.
Chris Early is vice president of Customer Strategy at ISO Innovative Analytics (IIA), a unit of ISO. IIA is devoted to developing innovative analytic decision support tools based on predictive modeling.
Dr. Paul Y. Mang is partner at Razor’s Edge Consulting and managing director of Avarie Capital. Previously, he was a partner at McKinsey & Company and a leader in its insurance practice.
Mo Masud is the national predictive analytics leader for Price Waterhouse Coopers’ Actuarial and Insurance Management Solutions practice, where he oversees the sales and delivery for various analytics solutions for the insurance industry.
What is the most challenging business issue for insurers today, and what role does analytics play in addressing it?
Chris Early: Premium growth is the biggest challenge. Finding an analytic solution that makes premiums better reflect the risk in a company’s current book is just the first step. Carriers also need to be able to price market segments they’re not currently reaching — geography and classes of risks — to achieve competitive rates and maintain rate adequacy. The biggest obstacle is the lack of credible internal data in these segments; therefore, insurers need an analytic solution that supplements their lack of credible data or the lack of data entirely.
Ed Combs: I would agree that delivering unit and premium growth is probably the most pressing business issue, especially when policyholders are dealing with high unemployment, flat or declining family incomes, and an aging fleet of vehicles that isn’t growing. On another front, property/casualty insurance company media spending in 2011 was up 12 percent over the previous year and now exceeds $5 billion, with virtually all spending directed to auto insurance.
Analytics plays many key roles: In advertising, carriers can use structured decision making to select the optimal way to allocate funds by location and media. Analytics also helps in identifying consumer behaviors and expectations to create a superior customer experience.
Using visualization tools, insurers can monitor and quickly spot issues with younger applicants. And telematics targets the superior loss characteristics of lower mileage and more conservative drivers, thus enabling better product development.
Are you seeing aggressive adoption of analytics, at least by larger insurers, for optimizing operational areas such as marketing, underwriting, claims, and reports?
Mo Masud: For certain applications such as underwriting, large and midsize insurers alike have aggressively adopted analytics, and there’s ever-growing adoption among large insurers in areas such as marketing and claims. Contributing to this increase is the lower cost of hardware and software computing, greater awareness among insurance executives of the business benefits of analytics, and a drive to keep up with market leaders.
The timing of the insurance industry underwriting cycle followed by the financial crisis of 2008 has also created opportunities for insurance companies to invest some of their surpluses accumulated during the hard market in more sophisticated pricing and underwriting tools in an effort to compete effectively in the current soft market.
Lastly, the professional community — both actuarial and nonactuarial and insurance and noninsurance — has done an effective job of educating insurance professionals on the merits of predictive analytics. In the era of social media, online discussion boards, and blogs, virtual analytics communities are sprouting up everywhere. These new avenues of communication coupled with numerous training and conference events dedicated solely to analytics have taken analytics out of the “back room” and into the forefront in many industries, including insurance.
Paul Mang: Clearly, more carriers are adopting analytics as a management tool. But I believe it will not be long until analytics will become pervasive in all aspects of insurance operations. To date, insurers haven’t yet changed how they view the analytics opportunity. In today’s world, data is viewed as a pond, and the focus is on (a) expanding the pool and (b) peering into the water for insights, patterns, and relationships. Tomorrow, however, data will be perceived more like a river flowing swiftly around companies and potential customers. Truly leveraging analytics in this context requires knowing how to capture data, selectively store and manipulate the right data, and, finally, draw insights. Carriers are moving in this direction. The challenge of creating the organizational context that supports this type of paradigm shift is not trivial.
How has predictive analytics changed the actuarial function both in terms of pricing and how actuaries work?
Combs: The design of contemporary auto insurance programs is an extension of long-standing actuarial classification methods. The new programs use classification methods to account for adjustments that were formerly made through underwriting judgment. Although statistics and credibility have always played an important role in rate making, predictive analytics brings these concepts to the forefront.
The transition picked up steam in the late ’90s with the introduction of auto insurance programs that combined rating and underwriting into a single, discrete module; underwriter intervention and exception granting were, for the most part, eliminated. Following a path seen in other businesses, more powerful computers, ever-larger databases, and predictive analytics have resulted in more sophisticated and accurate rating models. More sophisticated generalized linear model (GLM) methods have made possible the analysis of more complicated data sets. For many insurers, nonlinear techniques are supplementing GLMs. Because of their expertise, actuaries are drawn to analyses that extend beyond their traditional rating and pricing roles to address process optimization and operational efficiency. Although the impact will be greatest on actuaries who work on rating and product design issues, predictive analytics along with expanded data will affect reserving as well.
Early: I believe analytics has added precision to actuarial work, which in turn has resulted in more accurate pricing overall and by risk. Predictive analytics has made the actuary’s job a little easier in that analytics can help the actuary evaluate extremely large amounts of data more quickly and perhaps incorporate data elements that previously were not available or were not considered to be relevant in the assessment of loss cost. Additionally, as departments of insurance become more familiar with predictive analytics, these tools could help streamline the approval process. All told, advancements in predictive analytics could mean instantly leveling the playing field — simply by adopting an off-the-shelf analytic solution.
There could be some danger, though, in that actuaries may become too reliant on model output and forgo their own judgments. Regardless of the sophistication of a predictive model, I think human judgment will always be of value in determining the best solution for a given company and its specific situation in a given time period.
How might an insurer cultivate deep analytical expertise and an analytics-based culture?
Mang: The CEO and senior leadership set the tone here. They have to serve as role models who exhibit the type of behaviors and fact-based decision making that will change the culture and ultimately make the most of their investments in analytics.
Any insurer that wants to cultivate an analytics-based culture should think hard about how uncomfortable the marketplace for ideas really is. Will executives with significant organizational influence today want to operate in a truly idea-based environment? It can be threatening. Strong leaders will take on the challenge, and they will invest in analytic capabilities — or outsource the skills if necessary — to support the long-term performance of their companies. The senior leadership in these carriers will spend less time reviewing highly positioned reports for internal consumption and will spend more time exploring the data.
Masud: Creating an analytics or fact-based culture is no easy task and requires a formal change management process. To cultivate an analytics-based culture, insurers must follow the direction of senior leadership, provide employees with the appropriate technical training, have the necessary performance metrics in place to measure results, and align compensation and an overall rewards strategy to encourage employees to apply analytics consistently in their day-to-day decision making.
Despite advances in data and modeling, insurers can still be exposed to the unexpected. What areas of predictive analytics are ripe for advancement?
Masud:The changing global climate and economic uncertainty will drive the need for more analytics applications in risk management, particularly catastrophe modeling and regulatory compliance. The current economic downturn has made it difficult for insurance companies to grow organically; therefore, there will be greater emphasis on analytics for target marketing, customer acquisition, and agency appointment applications.
Combs: Advances in data and modeling depend on statistical methods, which are based on the law of large numbers. On the surface, it seems such methods would have little to say about major, unexpected individual events that affect the company as a whole. Natural disasters, for example, appear sufficiently random and infrequent to make statistics inapplicable. But sophisticated statistical models can match a company’s individual loss exposures with probability distributions of infrequent events. The key point is that these models don’t predict next year’s specific weather events; instead, they use a historical probability distribution and match it to the company’s current distribution of exposures.
Looking to the future, expect analytics to play an even more prominent role. Insurers may use crowd-sourcing techniques to query big data sets developed from social media, blogs, and other publications to identify significant new exposures. For example, if one could have been performed, it is easy to imagine a crowdsourcing analysis in the early part of the last decade providing a warning on toxic mold exposures to property insurers in time for them to develop risk mitigation strategies.
How are the most advanced carriers pushing the analytics frontier?
Mang: The carriers I would watch for future breakout performance are those that have done more than incorporate new data sources and analytic tools. While there is much recent focus on customer centricity, the real source of sustainable advantage for carriers as they look for profitable growth is to develop a ruthless economic perspective as part of their analytics capabilities. The insights gained from all the ways carriers can now observe and predict behaviors need to be translated into business models that create economic value.
Early: Recently, I’ve seen insurers looking into price optimization over a multiyear time horizon. Critical to the success of this technique will be validation of actual realized results versus expectations and then using that information to continue to refine the price optimization model.
Where could the industry push the frontier? The next “silver bullet” in pricing is the complete risk assessment of an individual or policy applied across various insurance products (personal lines and commercial lines). For example, the auto-related characteristics of an individual or policy might be predictive for homeowners and vice versa, or identifying a nontraditional rating factor could be predictive for one or more lines of insurance.
To accomplish that, a company will need to merge data for auto, homeowners, and other lines of business; perhaps append nontraditional data; and run predictive models to determine the complete risk assessment of an individual or policy for all lines of business. I haven’t seen anyone do this in a significant way, but think of what opportunities those analytic applications could bring.