At least twice in 2016, well-known publications have questioned the longevity of social media analytics. Forbes published an article entitled “Is the era of social media analytics coming to an end?” and Entrepreneur posted an article called “This is why social media analytics are essentially worthless.” While this was happening, social media data was reportedly being used to help facilitate voter targeting operations during the 2016 U.S. presidential campaigns, and some of the world’s largest banks were exploring social media in lieu of traditional credit histories.
Given these conflicting developments, property/casualty professionals who analyze social media may rightfully wonder if their work has already ended or just begun. In recent months, news feeds for many insurance professionals have probably turned up reports involving social media data being used in property/casualty insurance. In one case, a large global reinsurer reportedly invested in a “personal data marketplace” start-up. The premise of the start-up is, in part, to help solve a perceived problem in which individuals share personal data online in exchange for free services. Depending on the agreement by which the data is shared, recipients may later use data for crafting relevant advertisements or even for resale to third parties.
Data originators arguably have limited ability to participate in the upside of these transactions. The start-up attempts to address this perceived asymmetry by providing a one-stop platform where consumers may decide, a la carte, with whom to share their social media data and for what potential benefits. While the investing reinsurer reportedly does not have immediate designs on using exchanged data in underwriting, long-term possibilities include a redefinition of the insurer/policyholder relationship and, in behavioral analysis, to help policyholders improve their health and safety.
Elsewhere in the industry, early attempts at redefining insurer/policyholder relationships have been met with frustration. Recently, a primary insurer announced a trial whereby prospective policyholders could voluntarily share selected social media data during the auto insurance quoting process. The insurer reportedly planned to analyze this data to discern personality traits associated with safe driving and then award discounts where deserved. In reporting the launch, the insurer posited that one social benefit from the program would be that young drivers, who typically pay higher rates (in part, due to their inexperience), could potentially lower their costs by having social media data considered in lieu of an extended driving record.
Before the program could fully launch, however, a leading social media provider reportedly blocked the insurer from using data communicated over its platform in the manner described. While the program apparently was relaunched in a more limited fashion acceptable to both parties, the reinsurer’s and primary insurer’s differing experiences reaffirm our original question. The age of social media analytics for reinsurers is in its promising beginnings, while the primary insurer’s era—at least as originally constituted—has come to a sudden end.
As insurers ponder various possibilities, it may be useful to define the term “social media.” When many hear this term, it brings to mind Facebook, Twitter, and other leading social networking sites. In fact, Forbes cites one harbinger of doom for social media analytics: the fact that an increasing portion of interactions over these platforms is labeled “private and ephemeral,” rendering resultant data “out of reach for public data mining.” One example would be direct messages that self-delete after a short time. The Entrepreneur article argues that even public content may contain limited, potentially misrepresentative feedback and that businesses may be better served relying on higher-quality sources of insight. For insurers, such sources may include classic auto rating variables, including age, gender, marital status, or accident history, as well as relatively newer sources such as credit histories or telematics.
Further consideration of social media may reveal sufficient quality and availability at the disposal of insurers if they look in the right places. Merriam Webster defines social media as “forms of electronic communication (such as Web sites) through which people create online communities to share information, ideas, personal messages, etc.” Beyond the handful of platforms mentioned above, Hootsuite, a social media company, identifies at least eight different types, including personal and interest-based networks, discussion forums and media sharing, and e-commerce and online reviews. In this way, the burgeoning “sharing economy” may be a powerful instigator of social media activity, as participants tout and rate each other’s services and patronage. Essentially, anything communicated over the web by or about an individual or business may loosely be considered social media, and it’s difficult to conceive of many entities that have not interacted with social media in some form, however peripheral.
Within the property/casualty space, the Insurance Research Council estimates that more than 10 percent of American drivers do not purchase auto insurance, even though auto policies are required in most jurisdictions. Princeton Survey Research Associates International estimates approximately 1 in 3 adults between ages 18 and 29 (the group called millennials) does not purchase auto insurance. The percentage of uninsured motorists resembles the percentage of adults with no credit history, which the Consumer Financial Protection Bureau estimates at approximately 10 percent overall and 25 percent for millennials. The exact extent of overlap between the uninsured and those without credit histories hasn’t been studied in depth. Given that many insurers use credit as part of their rate calculations, drivers without robust histories may not qualify for related discounts—and this may be one reason some forgo coverage. So, can social media data present a remedy?
According to Pew Research Center, approximately two-thirds of all adults—and 90 percent of millennials—communicate on social media. This may be one reason why global banks are exploring how to use social media data as a supplement or replacement for analysis of traditional credit histories. For some time now, online lenders have reportedly (on an opt-in basis) been using social media data to qualify small businesses for credit-line increases. The factors considered include a business’s activity on online marketplaces and auctioneers. A leading credit rating agency recently began exploring a similar approach for consumers, although a social media provider reportedly stymied such efforts by restricting data access. Ultimately, important considerations of privacy and access must be managed, even as the use of social media in assessing credit risk may become routine.
Whether we’re talking about insurance discounts or next-generation credit scores, an important assumption likely lies at the heart of recent market activity: Social media data is somehow predictive of risk behavior. To validate this hypothesis in insurance, we performed an experiment using a sample of 1,000 social media users nationwide who publicly indicated that they purchase insurance and, in some cases, filed a claim. For each user, simple attributes were listed, such as the number and types of different social media accounts they maintained, hobby and interest categories, and perceived influence and number of friends or connections. (We found the last category to be quite predictive.) Data for 500 users was used to develop a “random forest” algorithm that related the social media–based attributes to the probability of a claim. Finally, data for the remaining 500 users was used for validation as shown in Figure 1. Verisk’s experiment had no motive other than to test whether social media data could be used to predict anything. This simple experiment was able to display differences in claim likelihood of greater than 50 percent between risk quintiles.
The extent to which insurers and other financial institutions will use social media in their risk-related practices going forward is unknown. Certainly, data derived from social media has been a valuable source of insight for claims investigators in combating fraud.
Although cracking down on fraud benefits everyone by helping keep costs down, many of the recent developments presented suggest there may be a larger place for social media data along various links of the insurance value chain. To make constructive use of this data, insurers need to better understand its availability and rights of use, establish predictive value (as in this simple experiment), and design innovative programs that are respected by policyholders, regulators, and social media providers. For insurers that are able to penetrate markets in this manner, the era of social media analytics may just be beginning. The rest may find their era has ended—and the rich opportunity presented by social media seemingly over before it started.
Figure 1. Sample predictive power of social media data for insurance claims.
We used attributes created from 500 users’ social media to predict their likelihood of experiencing an auto claim. We then validated the models on a holdout data set of 500 additional users. In the chart below, we sorted the 500 users from lowest claim likelihood to highest claim likelihood and divided them into five equally sized groups (horizontal axis). We then calculated an average claim frequency for each group (vertical axis). Even our relatively simple approach was able to identify one in five users who were more than 70 percent likely to experience a claim.