Insurance carriers looking to better market and manage risks should use social media as a rich component of a robust analytics platform. By augmenting existing big data projects with social media feeds, carriers can identify key information about their insureds that would otherwise be difficult to gather in a timely manner. Social media data analytics can be a competitive advantage leading to greater sales, lower claims and increased customer satisfaction. However, insurers should be careful with the data or risk crossing the “creepy line.”
With more than one billion users on Facebook and two billion total social media users across all platforms, the data shared is immense. The data that can be extracted from social media varies by platform, but in general the information goes far beyond pure text. Social graphs describe connections and relationships; profile updates highlight life change events such as marriage and the birth of children; geolocation tags highlight travel; and continuing communication can be parsed for activities and attitude.
Modern carriers looking to leverage analytics for a competitive advantage should already have a big data capability that pulls data from policy, billing and claims systems, call center logs, portal and app usage, third party enhancement tools such as Dun and Bradstreet and other sources to build a robust picture of each insured. This data can be mined using machine learning and neural networks to identify risks that should be exited, opportunities for cross-selling and best marketing opportunities to insureds and prospects. Social media is not a replacement for this data, rather a rich addition to it. By augmenting known facts with machine processing of social data, insurers can enable a more detailed and nuanced analysis that the same analytics routines can use to further refine analysis.
See also: Should Social Media Have a Place?
Examples of enhanced capabilities with this more robust analysis include:
- Prescriptive marketing: Asses the marketing mechanisms and messaging that will be most effective in converting the prospect to an insured through analysis of social graphs, profile data and language usage. By parsing the semantics of a user’s language and analyzing their social graph for the type of language they are accustomed to seeing and, importantly, that they have chosen to see, marketing can be best tailored for the prospect.
- Life event based cross-selling: Identify changes in relationship, location, job or family structure that enable marketing or sales to proactively contact the insured to recommend additional products or services. An example is increasing term life coverage for a new parent. By contacting insureds with relevant products at the moment of a life event, agents can be highly effective at converting new sales.
- Continuous risk assessment: Continuously assess insureds’ risk profiles by expanding the analysis of an insured beyond their behaviors with the carrier to their behaviors with all other parties as evidenced in their social media communications. Updates about employment, travel, family circumstances or other items can impact how a framework understands the facts of an insureds’ interactions with the carrier. By understanding this, a carrier can better tailor reserve models or reevaluate whether to renew the policy.
- Claim fraud detection: Identify potential claim fraud activity by monitoring geolocation, language and other data elements to confirm reported stories and check for telling language used in public communications. For example, a claim for workers compensation could be identified for potential challenge if a system identifies geolocation data from a golf course.
- Customer sentiment: Be proactive with alerts of customer dissatisfaction with claim handling or price adjustments through text mining, allowing for remediation prior to losing a customer. By identifying dissatisfaction, the carrier can take better next steps in communication and outreach to maintain a client’s goodwill and business.
- Language is complex data: Because social media is so dependent on written words, language analysis is a common basis for analysis. Semantic assessment is useful in identifying underlying emotions and intent. However, words have different meanings in different sub-cultures, geographies, friend groups and even in different transmission medium. As such, language parsing should often be used to augment existing analysis, not to serve as a primary source of facts.
- Usage of social media varies: In general, social media has widely different usage by age group and other demographic segments. Uptake rakes are not the same across all demographic groups, as demographic analysis of Facebook vs. Snapchat bear out and actual usage of the tools varies by group. The amount of data shared by younger users typically, but not always, dwarfs that of their parents. Analytical frameworks need to be configured to account for these differences and not draw unwarranted conclusions from different behavior patterns.
- Usage of social media starts and stops: Users of social media will start, stop and potentially resume use many times. Details of usage may also change as users’ needs or privacy concerns change. This requires analytical tools to be flexible in analysis — to understand that lack of data, limited data or infrequent posting is not necessarily an indicator of underlying behaviors of the prospect or insured.
- Security is tricky: In the post-Snowden era, concerns about data privacy and usage are increasingly spotlighted by the media. Insurers should be cautious about how they collect, how they store and how they take action based on social media information. De-identification and storing only the analysis of the underlying data are potential paths among others. This should be continuously evaluated.
