With the advent of the digital age, data is becoming increasingly accessible—in terms of sheer amount, rate of generation, and types and varieties. It also exists in different shapes and forms and is obtainable through a multitude of sources and channels. The explosion of digital footprints allows for a deepened understanding of consumer behavior, which in turn allows businesses to better address the needs of their customers. Amazon and Netflix offer poster-child case studies because they exemplify the use of data to deliver convenience and customized offers to their customers. While the practice of collecting and analyzing data to forecast and predict is deeply ingrained in the insurance industry, there is much room for expansion and advancement in data scope and analytic techniques.
Most important, the industry must prioritize data security and privacy protection above all, being resourceful and skillful in mining insights and masking sensitive information while maintaining the highest degree of ethics and professionalism. The interplay is a balancing feat where actuaries strive to extract insights that could facilitate adequate, equitable, valuable and innovative product offerings and encourage positive behavior engagement while respecting the threshold of information collection and ensuring the process never contravenes any personal information boundary and privacy.
The ubiquitous nature of data might prompt more entrenched regulatory oversight and education for building a stronger ethical conscience and a safe and trusted environment for further innovation that benefits the industry and consumers alike.
COVID-19 Digital Acceleration
Living through the pandemic, many aspects of people’s day-to-day lives have shifted online thanks to virtual learning and remote working. Consumers now expect more and more in terms of digital transactions and services, which in turn generates high-frequency digital interactions and increased digital footprints. Digitalization will shift the insurance industry landscape and business model toward simplicity and personalization, as digital distribution commands easy-to-understand products. Insurance ecosystems are emerging with cross-industry partnerships to offer more all-encompassing value propositions. Nimble players are on the rise and offering embedded, on-demand, dynamic and flexible risk coverage protecting “micro-moments,” following the trend toward increasing product relevance based on contextualized insights.
The resulting plethora of data enables more sophisticated analytics and understanding of consumer behavior. Today’s data economy presents an opportunity for insurers to transform many long-established insurance practices, consumer expectations and their perceptions of the industry. Big data and artificial intelligence (AI) are reshaping the insurance industry across the board, from product development, pricing, sales and underwriting to risk analytics and management. Behavioral analytics allow for deepened understanding and more granular stratification of customers, with the vast potential to tailor products and services to specific consumer preferences, which in turn could fuel further innovations.
Big Data Innovation
As the industry embarks on the digital transformation journey, constraints around data accessibility, reliability and granularity remain. For example, data collection is often limited to a select portion of the population, such as those willing to provide their wearables data in exchange for a reduction in insurance premium. Real-time, unbiased, quality data is often controlled by third-party digital channels and not readily available to insurers. These challenges work against meeting modern-day standards around delighting customers with convenience and simplicity in purchasing products and services.
Product commoditization leaves a protection gap against more diverse, complicated health risks. The lack of a sufficiently robust and granular risk assessment framework could leave the customers of substandard risk who are most in need of protection out of affordable insurance options while limiting business opportunities for insurers.
Being more creative, intelligent and breaking through the conventional way of assessing, selecting and pricing risks could enable less obtrusive data collection and ease certain pain points today’s insurance customers experience. Data that is currently captured through traditional underwriting is often a combination of self-reported health and financial information supplemented by medical examination for higher sum assured policies. A host of auxiliary data could potentially be tapped to add breadth and depth to the assessment process, helping contextualize and enhance understanding of customer behaviors and latent risks that could be unraveled through extracting meaning and insights from a blend of structured and unstructured data.
For example, the use of wearables and Internet of Things (IoT) sensors (e.g., devices measuring biometrics as well as connected cars and smart home telematics) can track one’s fitness conditions and support usage-based solutions, managed care, risk prevention and a range of assistance services. GPS and mobile data can be used to develop activity detection models and draw an activity map leveraging location and call detail records. Studies have been conducted to explore the use of trajectory data to link one’s mobility patterns to lifestyle behaviors, potential health conditions1 and mental state.2 Spatial-temporal analysis can reveal activities and events and the evolution of their patterns over a sufficiently long observation period, which is key to understanding individual behavior consistency and habits.3
The combination of spatial and internet data with an added time dimension is powerful in informing various human behavior routines and inclinations. For instance, geostatistical information, internet data and time can be evaluated together to inform distraction level, nighttime activity patterns and social connectivity. Research linking consistent exposure to light at night and excessive screen time to potential impact on coronary health and risk of cancer and obesity could offer fresh insights around emerging lifestyle risk factors, such as circadian disruption and sedentary routine and their correlation with morbidity and mortality risks. Analyzing one’s usage of mobile apps could offer potential indication of lifestyle preferences and interests, leveraging the concepts of “app usage fingerprints”4 and “unique app signatures.”5
Advanced measures, such as radius of gyration and location entropy, aim to capture one’s moving range (i.e., distance in terms of both magnitude and diversity) and the dispersion or randomness of one’s time spent at various points of interest. Research observations note that a larger moving range could imply less time available for a workout,6 and higher levels of social anxiety and depression could be related to more time spent at home.7 Combining this with alternative data with which insurers are already familiar or using (e.g., electronic health/medical records and financial records) on top of public domain information relating to credit history, home value and map information presents myriad opportunities for uncovering complex behavioral dynamics and reducing the cost of information asymmetries.
Structuring of Behavioral Insights
Turning unstructured data into structured information requires a thoughtful and systematic process of dissecting and reconstructing data in different forms from various sources. A flexible, modular data block structure can serve as a foundation for more advanced and adaptive analytics. Time blocks of data contain inherent event-based properties and attributes that can be systematically analyzed. The continuity and consistency of how these blocks evolve manifests into human behavior patterns and clusters. These building blocks of data can be further expanded and rearranged to incorporate real-time information updates and allow for dynamic readjustment and refinement of evolving perspectives.
An ensemble of supervised and unsupervised techniques can be used to unlock behavioral insights, enabling smarter risk assessment, personalized products and services, and an enhanced customer experience. For example, on top of using regression to associate indicative behavioral patterns and extract insights with actual claim results, a block-based decision tree approach could be applied to organize behavioral indicators for personas mapping and segmentation. Latent behavioral function also could be created through measurable behavioral attributes and metrics with demonstrated and dominating predictive power in risk classification, leveraging existing underwriting guidelines to obtain a deeper understanding of behavioral and risk propensities. Being able to translate traditional risk rating perspective into a systematic, quantitative framework by incorporating measurable, nontraditional rating factors will augment insurers’ a priori risk segmentation mechanisms and experience upon unraveling of new behavioral and other types of risk factors.
Potential Value Propositions
Streamlining insurance underwriting is becoming more pressing as uniformity, simplification and commoditization of products and their distribution via digital channels increases. Information enrichment through tapping emerging data sources could help strengthen and supplement traditional underwriting judgment around applicants’ socioeconomic status and potential health or medical-related circumstances; sum assured limits might be raised while requesting less information from customers to simplify the overall insurance application process.
Product features and marketing strategies could be augmented and customized through a deeper understanding of customer behaviors and their inherent risk profile and characteristics. More personalized and efficient coverage could be extended based on the latent risks of individuals, while clustering and segmenting by various risk dimensions and behavioral criteria could enable more targeted customer reach and effective distribution. Insurers could select and upsell to preferred risk segments or filter through an apparent substandard pool of risks to “skim the top” for targeted offers.
Being able to track behavioral patterns and shifts along key rating dimensions (e.g., health, financial, lifestyle) over an extended time frame might lend predictive power to the assessment of one’s behavioral health through identifying consistent patterns and detecting deviations around one’s activities and habits, indicating level of discipline, variety-seeking tendencies and other behavioral science attributes relevant for risk analytics and consumer insights. Events-driven and/or location-influenced changes in behavioral patterns (e.g., starting a new job, more frequent travel, moving to suburban area) could spark novel risk assessment considerations and business potential at the same time. For instance, a conditional personal accident offer could be extended to a group of preferred customers who exhibit relatively consistent behavior and do not travel much (discerning by location-based statistics measures, such as entropy and radius of gyration).
From a risk management standpoint, being able to dissect individual behavioral patterns could support claims adjudication, help control for moral hazards and anti-selection risks, and provide a basis for ensuring fair, adequate and reasonable premium pricing. A more holistic suite of incentives and motivations could be designed to promote insurance risk reduction; for example, via managed care and other feedback mechanisms, encouraging positive behavioral change and shifting focus to proactive risk prevention.
Brave New Ideas
There have been promising developments in the insurance market with applying data and analytics to business use cases. One insurer is using wearable technology and real-time data to develop a “biological age model”8 not only to underwrite, but also coach and motivate customers to lead a healthier lifestyle. Another player partners with a health technology company to promote a “personal activity intelligence”9 score for reducing mortality risk from cardiovascular and other lifestyle diseases through biometric sensing. These market examples demonstrate the immense potential available to insurers, making use of not only digital health data, but a variety of alternative data, such as mobility, internet, social media and other exploding amounts of unstructured data. But the noise must be tuned out in the process of extracting meaningful signals.
More advanced and mature big data-enabled analytics and applications could emerge as the industry continues to test and learn through experiments and pilots, partnerships with the technology sector and expanding the innovative mindset. In particular, behavioral metrics could be further developed with migration measures tracked across time to enable advanced persona delineation. Multifaceted event and behavioral detections through a combination of location data (e.g., GPS tracking), mobile usage patterns (i.e., apps and calls), contextual information (e.g., point-of-interest mapping) and research learnings could lend informative insights and enable powerful risk and customer segmentation capabilities. Furthermore, studying cross-variable interactions—for dimension reduction, control variates and behavioral inferences—and integrating with external observations (e.g., claims, take-up and persistency experience data) could boost portfolio performance.
Case Study: Information Enrichment Tool
Combining information collected through a conventional application questionnaire with alternative and external data sources can help insurers paint a more complete and unbiased picture of applicants, enabling a fuller risk profiling assessment. For example, one global reinsurer10 is developing a rating tool that leverages real estate transaction and lifestyle-based indicators to strengthen the evaluation of one’s financial standing. Taking an applicant’s residence information and supplementing this information with a repertoire of real estate transaction data, rank ordered by the property value of each carved out area locally as well as regionally, the tool can assign a home value-based score as one predicting component of the applicant’s level of financial affluence. On top of that, certain activities, such as golfing and international travel, and mobile app usage, such as apps for high-end investment or micro-loan users, could serve as potential indicators of one’s socioeconomic status.
An information enrichment framework can be designed in a way that each piece of information is evaluated and incorporated by assigning greater weight to more illuminating information. Here is an example: Person A resides in the highest-value ranked area and person B the lowest. Both are identified as golf club members. While the tool will adjust both individuals’ ratings upward given golf club membership represents a positive indicator of financial affluence, the scale of adjustment will be smaller for person A and larger for person B. There is information overlap in discovering someone from an upscale neighborhood (person A) plays golf, which merely affirms the wealth factor for person A. Whereas the same new discovery reveals more divergent insights around person B’s financial profile and warrants a more meaningful adjustment to the previous home value-based perspective.
As mentioned previously, the explosion of ubiquitous data in terms of both volume and diversity could be deployed to unlock insights previously unattainable. In addition to self-disclosure questionnaires, wearables and IoT devices, auxiliary public information and data from various industry partners (the likes of golf clubs and travel agencies) could be tapped into for elucidating mobility- and activity-based behavioral patterns, persona characteristics and socioeconomic attributes, enabling a range of applications for augmenting the value of insurance and the overall customer experience.
Care and Responsibility
Appropriate uses of data yield significant benefits to both consumers and insurers, including more customized products and value-added services, efficiency and convenience, enhanced risk insights and controls, personalized care, proactive prevention and early intervention. Nevertheless, caution must be taken to ensure consumer consent to data use, protection of personal sensitive information and contextual integrity. At the same time, unduly pricing out risks, potential discrimination and data leakage must be avoided.
As an industry and a profession, actuaries must uphold themselves to rigorous standards; follow the guidance of internationally recognized laws and regulations, such as General Data Protection Regulation (GDPR), Health Insurance Portability and Accountability Act (HIPAA) and Actuarial Standard of Practice (ASOP) No. 23; and stay abreast of the latest regulatory updates and requirements. Precautionary measures related to data access and processing should be considered to protect data security and consumer privacy.
- Obtain all necessary consent and preauthorization from customers and working partners before data extraction, and encode all sensitive data to ensure security.
- Establish strict processes and controls where only qualified and authorized personnel are allowed access to sensitive data systems, and conduct regular training on data protection and privacy to ensure involved personnel can demonstrate sufficient proficiency in understanding and practicing the necessary data protection policy rules.
- Put in place firewalls and checkpoints to prevent unauthorized access or improper use of any personal sensitive information; maintain separate development, testing and production environments to ensure proper segregation of duties and access controls.
- Set up proper governance around information extraction procedure; for example, clearly document data fields being used along with the “why” (purpose) and “how” (extraction method/analytic technique) that will be used, and report any addition or update for review and approval by the appropriate governing committee.
- Maintain clear line of sight with working partners and increase accountability through formal information security and data protection agreements with close monitoring of data activities.
- Extract only event-driven insights and contextualized information, employ analytic techniques to generate aggregate distributional statistics and categorical cluster variables that are agnostic to individual identity, and prioritize privacy protection.
Ensuring data integrity is not only part of an actuary’s professional remit, but it also is essential to delivering reliable results and relevant insights. ASOP No. 23 covers the appropriate steps to take, including reviewing data for reasonableness and suitability for purpose; sense-checking against alternative sources and external references to ensure data used is current and reasonably consistent; documenting and disclosing data sources, any judgmental adjustments and assumptions applied; evaluation of data accuracy and completeness; any constraints and limitations identified; and the corresponding techniques used to fill in the gaps and address questionable observations, such as interpolating missing values and smoothing oscillating data points.
After all, as an industry, our ultimate point of existence is to serve as a risk solution provider to society. One clear path forward is to take advantage of the current data revolution and advances in computational analytics to help individuals and communities manage their risks and enhance their insurability.
Statements of fact and opinions expressed herein are those of the individual authors and are not necessarily those of the Society of Actuaries or the respective authors’ employers.
- 1. Lin, Zongyu, Shiqing Lyu, Hancheng Cao, Fengli Xu, Yuqiong Wei, Hanan Samet, and Yong Li. 2020. HealthWalks: Sensing Fine-Grained Individual Health Condition via Mobility Data. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies 4:1-26. ↩
- 2. Müller, S. R., H. Peters, S. C. Matz, W. Wang, and G. M. Harari. 2020. Investigating the Relationships Between Mobility Behaviours and Indicators of Subjective Well–Being Using Smartphone-Based Experience Sampling and GPS Tracking. European Journal of Personality 34, no. 5:714–732. ↩
- 3. Wang, Yingzi, Xiao Zhou, Anastasios Noulas, Cecilia Mascolo, Xing Xie, and Enhong Chen. 2018. Predicting the Spatio-Temporal Evolution of Chronic Diseases in Population With Human Mobility Data. Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence, 3,578–3,584. ↩
- 4. Tu, Zhen, Runtong Li, Yong Li, Gang Wang, Di Wu, Pan Hui, Li Su, and Depeng Jin. 2018. Your Apps Give You Away: Distinguishing Mobile Users by Their App Usage Fingerprints. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies 2, no. 3:138. ↩
- 5. Welke, Pascal, Ionut Andone, Konrad Błaszkiewicz, and Alexander Markowetz. 2016. Differentiating Smartphone Users by App Usage. Proceedings of the 2016 ACM International Joint Conference on Pervasive and Ubiquitous Computing, 519–523. ↩
- 6. Chen, Xinlei, Zheqi Zhu, Min Chen, and Yong Li. 2018. Large-Scale Mobile Fitness App Usage Analysis for Smart Health. IEEE Communications Magazine 56, no. 4:46–52. ↩
- 7. Saeb, Sohrab, Mi Zhang, Christopher Karr, Stephen Schueller, Marya Corden, Konrad Kording, and David Mohr. 2015. Mobile Phone Sensor Correlates of Depressive Symptom Severity in Daily-Life Behavior: An Exploratory Study. Journal of Medical Internet Research 17, no. 7:e175. ↩
- 8. Biological Age Model (BAM): Using Wearable Data to Empower Healthier Lives. SCOR. ↩
- 9. Ashley, Thomas, Jean-Marc Fix, Ammon J. Dixon, and Yvonne Ren. Life Underwriting: Personal Activity Intelligence (PAI) and Cardiorespiratory Fitness—Validated for Mortality Risk Assessment. Gen Re, 2020. ↩
- 10. Pacific Life Re Partners With Sapientus in Cutting-Edge Behavioural Risk Analytics for Insurance. Pacific Life Re, January 25, 2021. ↩
Copyright © 2021 by the Society of Actuaries, Schaumburg, Illinois.