Many property and casualty (P&C) insurers are advancing their use of predictive analytics to improve business performance and leverage big data, with dramatic changes planned over the next two years, according to a Willis Towers Watson survey.
Looking beyond pricing and risk selection, P&C insurers are aspiring to increase big data use from both internal and external sources, including telematics, web clickstreams and customer/ agent interactions. The mix of modeling methodologies used
by insurers is also shifting, with much greater use of machine- learning techniques planned two years from now. And while personal lines carriers remain the market’s predictive analytics leaders, standard commercial lines and specialty lines carriers, in particular, are steadily advancing their use.
Future uses of predictive analytics
More than half of insurers surveyed (54%) currently use predictive models for underwriting and risk selection, but many additional uses are planned. Two years from now, carriers expect to significantly ramp up model use in important areas (Figure 1). Several top future uses cited by survey
respondents (such as evaluation of fraud potential, claim triage and evaluation of litigation potential) reflect the importance of effectively managing bottom-line results as carriers continue
to compete aggressively on price. Other top uses (e.g., report ordering) are directed at expense control, another key to managing profitability in an intensely competitive marketplace.
To help improve top-line growth, carriers are also using modeling to inform their marketing and advertising strategies.
Figure 1. Top predictive modeling uses beyond risk selection
As predictive analytics become more integrated across functions, particularly among larger carriers, business insights beyond risk selection will amplify.
P&C insurers are aspiring to increase big data use from both internal and external sources, including telematics, web clickstreams and customer/agent interactions.
Figure 2. Top areas where big data is expected to help
Figure 3. Top-growing big data sources
Figure 4. Top big data challenges
Big data aspirations for advanced analytics
“Big data” in our survey refers to both large volumes of data with a high level of complexity and the analytical methods applied to them, which require more advanced techniques and technologies in order to derive meaningful information and insights, often in real time. Big data is currently used by 42%
of carriers to inform pricing, underwriting and risk selection.
Two years from now, insurers expect big data use in many key business functions to more than double (Figure 2). For example, 60% say they expect to use big data for management decisions two years from now — a 41 percentage point increase from 19% of insurers using big data in this area now. Use of big data to support loss control and claim management is also expected to jump 41 percentage points — from 17% now to 58% two years from now.
In the future, data collected from usage-based insurance applications are expected to grow the most (32 percentage points). Data from agent and customer interactions, smart-home data and social media will also gain importance (Figure 3).
Big data challenges: What’s possible and practical?
While enriching models with new data sources for broader application is ripe with potential, there are some significant challenges insurers must master. Half of survey respondents are most challenged by people issues, including resource availability, training, skills and capabilities. Data capture and availability challenges rank second (44%), as many carriers struggle with legacy policy administration systems that were not designed to capture this level of data (Figure 4).
Larger carriers have been more active in exploring big data applications that use both internal and external data. This intuitively makes sense given their scale with the natural advantages of greater access to capital for investing in both talent and technology, as well as their ability to generate larger and more robust data sets. Many of the largest carriers have dedicated teams charged with this specific purpose.
Smaller carriers will need to strategically assess their options, developing big data capabilities where critically necessary, and becoming fast followers behind larger carriers when size and scale issues make using data from internal interactions unfeasible.
Trending modeling methodologies and techniques
Insurers have made great strides in increasing model sophistication, and a sizable 60% of respondents say their companies are data-driven. Insurers’ biggest challenges are mainly around data capabilities rather than aspirations. Insurers that are lagging identify data warehouse constraints and data access (74%), and data integration challenges (63%) as primary roadblocks.
Generalized linear models (GLMs) are the industry standard, used by 88% of responding carriers, followed by one-way analyses (71%) and decision trees (31%). To address data availability challenges, many carriers supplement models with competitive market analyses (73%) or industry data (68%).
Two years from now, insurers expect to increase the mix of approaches they use beyond GLMs, and 43% expect to
incorporate machine-learning techniques into modeling results. Precise application of machine-learning techniques clearly depends on implementation requirements. Modelers and data scientists are broadening the toolkit, recognizing that different methods may be suitable to analyze different data sources
and different insurance problems. Interpretability and ease of deployment are also key considerations in matching method to application.
Personal lines versus commercial lines
While personal lines carriers are the market’s predictive analytics leaders, commercial lines insurers have been making steady progress. Nearly half of commercial lines carriers surveyed (49%) currently use predictive modeling for commercial auto coverage, and 39% say they plan to use it.
Many commercial lines products are individually underwritten due to the relative uniqueness in each risk. Predictive modeling therefore plays a different role, facilitating development of benchmark pricing for the underwriter to use in their risk evaluation. Commercial insurers have gained experience using benchmarks for underwriting and pricing, and they are now seeing applications for predictive models across all lines of business and account sizes.
Personal lines carriers say the most valuable underwriting attributes are credit/financial (96%), vehicle characteristics (96%), prior claims (93%) and property characteristics (89%). Standard commercial lines carriers rely most on account experience (93%) and billing information (78%), while specialty lines carriers equally favor account experience and other non- credit attributes specific to individual/business policyholders (both 67%). These differing approaches reflect personal lines carriers’ traditional focus on individual risk characteristics for pricing versus commercial lines’ traditional reliance on individual account underwriting data for pricing.
Insurers must embrace complexity to find clarity
In spite of complex challenges, insurers that embrace modeling complexity by focusing on data enrichment, advanced analytics and technology advances can achieve a significant return on their investment. Insurers that catapult beyond their competition do this with superior data organization and analysis. Insurers aspiring to unlock the potential of big data must be strategic, persistent and consistent.
About the survey: Willis Towers Watson’s 2015 Predictive Modeling and Big Data Survey asked U.S. P&C insurance executives how they are using, or plan to use, predictive analytics and big data. The survey was fielded from September 9 to November 2, 2015. Respondents comprise 11% of U.S. personal lines carriers and 17% of commercial lines carriers.