Total life insurance: Logics of anticipatory control and actuarial governance in insurance technology

Abstract

Calling attention to the growing intersection between the insurance and technology sectors—or ‘insurtech’—this article is intended as a bat signal for the interdisciplinary fields that have spent recent decades studying the explosion of digitization, datafication, smartification, automation, and so on. Many of the dynamics that attract people to researching technology are exemplified, often in exaggerated ways, by emerging applications in insurance, an industry that has broad material effects. Based on in-depth mixed-methods research into insurance technology, I have identified a set of interlocking logics that underly this regime of actuarial governance in society: ubiquitous intermediation, continuous interaction, total integration, hyper-personalization, actuarial discrimination, and dynamic reaction. Together these logics describe how enduring ambitions and existing capabilities are motivating the future of how insurers engage with customers, data, time, and value. This article surveys each logic, laying out a techno-political framework for how to orient critical analysis of developments in insurtech and where to direct future research on this growing industry. Ultimately, my goal is to advance our understanding how insurance—a powerful institution that is fundamental to the operations of modern society—continues to change, and what dynamics and imperatives, whose desires and interests are steering that change. The stuff of insurance is far too important to be left to the insurance industry.

We can now reveal things that, in the past, only god knew about, thanks to technology including AI.

—Mikio Okumura, head of Sompo Holdings, Japan’s second largest insurer (quoted in Sugiura & Lewis, 2022)

Insurance has the rare distinction of being in almost everybody’s life and on almost nobody’s mind. The industry is ‘among the most pervasive and powerful institutions in society’ (Ericson et al., 2002, p. 3). At the same time, for most people ‘insurance remains in the background. It is something that people take for granted to meet exigencies they would prefer not to think about’ (Ericson et al., 2002, p. 3). Even folks in the industry will readily (even too eagerly) admit that the details of insurance policies and actuarial practices are dreadfully boring. They may be boring when treated as impenetrable texts and arcane techniques existing on their own, totally stripped of all their social contexts, connections, and consequences. Once we bring all these things back into the picture, not only does the stuff of insurance become much more interesting, its massive influence at every level of society—and thus the importance of grappling with it—also becomes very clear.

Perhaps because of insurance’s dull reputation, this subfield of research has remained relatively small. While this is an active area of study—reviewed here—it has not been well integrated into the broader streams of research on related subjects. I want to send a bat signal for the interdisciplinary fields that have spent the last decades closely studying the explosion of digitization, datafication, smartification, automation, and so on, calling attention to the interest and importance of the intersection between the insurance and technology sectors, or ‘insurtech’. Many of the major themes, dynamics, and issues that attract people to researching technology are exemplified by insurtech, often in exaggerated ways by the still emerging markets and applications of an industry that has real, material effects on everybody.

I lay out a techno-political framework for how to orient critical analysis of developments in insurtech and where to direct future research on this growing industry. This constitutes the body of the article and its main contribution to the study of power relations and governance mechanisms in insurance. Without a more in-depth understanding of how insurtech is developing, we might slip into a future where we are unable to refuse, effectively or practically, using products and services that take the logics I outline here to new heights of intensity and intimacy in our lives.

Techno-politics of insurance

Insurers have long been adept at aggregating statistics about populations, which helps them apply averages and predict probabilities. They have had much more difficulty monitoring individuals. The sources of data have been largely unavailable; at least in the volume, variety, and velocity required for small-scale anticipation and real-time reaction. The tools at their disposal—used to sort insured people into risk pools and surveil their behaviours—have been blunt. While applied to great effect, they tended to be deployed in broad ways.

That is now changing. The promise of developments in insurtech is that insurers will be equipped with sharper tools. With the capabilities offered by digital technologies like networked devices and AI analysis, insurers are gaining access to new capacities for assessing individuals, governing risk, preventing loss, and capturing value. It’s one thing for an insurer to say, ‘We anticipate that people with your demographic profile are more likely, over the course of a lifetime of coverage, to be more costly risks. But if you meet certain conditions, we will lower your premiums.’ It’s quite another for an insurtech to say, ‘We anticipate that you, based on behavioural data collected from these specific smart devices and additional data purchased from brokers, are likely to have a risk event in the next three weeks. We have adjusted your premium accordingly.’

Imagining total life insurance

Imagine a possible insurtech product—total life insurance—amalgamated from different types of existing insurtech as they become further interconnected with each other. We can already see this happening in many other realms of smart technology where once-different networks and siloed databases become entangled and fused with each other. Total life insurance would involve insurers merging different programs, partnering with other organizations, and tapping various data streams and customer interactions to build detailed profiles and dynamic relationships with policyholders from cradle to grave. The product could come with a personalized AI life coach, synthesizing data from sources like our car telematics, smart fridge, fitness wearable, and other devices to target us with prompts for how to live healthier, less risky lives—and also take note when we do not follow recommendations. It would entice people with promises of discounts and convenience, but at the sake of submitting to totalizing surveillance and discipline.

This vision isn’t untethered from reality; it’s mostly an acceleration of already existing programs, logics, and aspirations. This is the same method used by consultancies like McKinsey or Accenture in the reports they sell to the industry on its future.

While we are discussing such scenarios in speculative terms, others are working to make them into reality. To the extent these technologies are adopted, it won’t be as found objects devoid of human influence, nor will it occur as a sudden paradigm shift. They will be adapted from things that already exist, adopted at a steady simmer. The real AI life coach could start off as a premium service subsidized by your insurer, perhaps in partnership with tech companies, before eventually becoming a required policy condition.

It seems highly unlikely that an AI life coach will arise in the way described here. However, the insurance industry is deeply interested in approximating such a system, piecing it together by connecting various devices and data, and encoding it with their own moralistic models of how the best version of yourself should live. According to insurtech advocates, there is no better method for risk management than AI powered by AI: artificial intelligence powered by actuarial intelligence.

The logics of insurtech

The insurtech sector is already beginning to construct a more pervasive and powerful regime for governing everyday life based on a set of core logics: ubiquitous intermediation, continuous interaction, total integration, hyper-personalization, actuarial discrimination, and dynamic reaction. Importantly, these logics have long been present in the insurance industry. The techniques and technologies used to enact them have changed, but their underlying motivations and aspirations are quite familiar.

We need only look at histories of insurance to see how the industry was at the forefront of investing, developing, and using large-scale information systems long before the contemporary age of data servers, cloud computing, and AI analytics (Bouk, 2015; Horan, 2021). By the 1910s, the life insurance industry had already ‘become premised on more individualized risk making’ (Bouk, 2015, p. 56, original emphasis). Major life insurers created ‘index divisions’—using index cards to record, store, and categorize detailed information about policyholders like medical assessments and accident reports—that ‘facilitated an unexpected revolution’ based on ‘new methods for personalizing risk assessments and a new centralized file (a proto-database) for storing thoroughly statisticized individuals in steel case files’ (Bouk, 2015, pp. 78–79). In 1903, a report by the Actuarial Society of America had detailed 98 ‘classes of risk’ to be used in underwriting insurance. A leading actuary at that time, Emory McClintock, said this type of fine-grain risk classification—and the detailed information, analysis methods, and risk theories it was built on—was the future of actuarial science (Bouk, 2015, p. 84). By 1981, ‘the most widely used assessment scheme’ for auto insurance was based on 234,360 different risk categories into which drivers could be sorted (Austin, 1983, p. 547). The industry’s trajectory, and the ambitions driving it forward, are quite clear as insurance embraces the capabilities of digital systems.

Contemporary innovations in insurtech are best understood as accelerating and amplifying these trends, rather than as entirely new paradigms. To be clear, my point is not that the once and future technologies of insurance are unfolding in an inexorable, linear, or any other deterministic manner. This is not a case of the tail wagging the dog, where insurance is driven by the demands of technology towards maximizing certain ends like individualization. My argument is quite the opposite: It is the interests of the insurance industry that have informed the creation and application of these technologies. This is especially clear in the emerging market for insurtech start-ups, which has largely formed to service, and attempt to shift through sales pitches, the highly capitalized needs of major (re)insurance corporations. Ultimately, any form of social governance or technological system is the result of many sources of ideological and material influence pushing in different directions until one form takes hold over the other possibilities. The history of insurance is no exception. My goal is to analyse how technologies have been created and are being used for specific purposes, what role they are playing in these developments, and place all of that in a longer history that counters narratives of sudden disruption.

In 1988, legal scholar Jonathan Simon analysed the ‘ideological effects of actuarial practices’ and identified an already-underway shift in this system of social governance: ‘Where power once sought to manipulate the choices of rational actors, it now seeks to predict behavior and situate subjects according to the risk they pose’ (Simon, 1988, p. 772). Power, though, does not need to choose between manipulating choices, predicting behaviours, situating subjects, and doing much more. All of these capabilities are available. The logics I outline here are meant to provide a framework for understanding the techno-politics of insurance as an ongoing regime of anticipatory control and actuarial governance. This regime has been developing and advancing for well over a hundred years and continues to innovate into the future as it attempts to reach a perhaps unreachable endpoint of total life insurance. In other words, insurers are not rolling out this regime all at once. It is forming as the result of different parts snapping into place as they become more widespread, refined, and normalized over time.

This article offers a theoretical synthesis for the purpose of discerning broader patterns in the insurtech sector and its ongoing formation, relations, operations, applications, and implications, with particular focus on American, European, and Australian markets. My analysis draws from a variety of sources. First is an extensive review of literature on insurance, focused on drawing together disparate work on insurance from disciplines like law, sociology, history, and political economy, while also supplementing it with professional work from actuarial science and risk management. While much of this work does not focus on technology or predates contemporary innovations, I have sought to connect it to insurtech as a way of contextualizing current advances. Second is an in-depth market analysis of the insurtech sector, which included extensive review of relevant materials from companies, consultancies, government regulators, professional organizations, and trade media. Third is direct industry engagement based on dozens of interviews and informal conversations with people working in different parts of the insurance/insurtech sector, ethnography at major industry conferences including Insuretech Connect Vegas, and attendance at smaller insurtech networking and educational events for people within the industry. All of this research forms the basis for my theoretical synthesis. Given my goals here, I save the thicker descriptions of empirical material for later work focused on digging deeper into specific aspects.

What follows is a critical survey of each logic I have identified. Each logic deserves its own in-depth treatment with further conceptual analysis, empirical detail, and concrete examples. That work will come as my own project and the research of other scholars continue. But first, there is value in seeing how the logics all come together to constitute an important system of governance in society.

Many of the logics I outline here are, in essence, different approaches to the same goal of building better relationships between insurers and consumers. The meaning of ‘better’ is in the eye of the beholder. For consumers, ‘better’ might mean building relationships that are more trusting, more certain, more convenient, cheaper, easier. For insurers, ‘better’ might mean building relationships that are more informed, more intimate, more personalized, more pervasive, faster, fairer. The first set of features are used as promises in the marketing for new insurance programs and products (Jeanningros & McFall, 2020); while their actual design and operation may prioritize the second set of features. This is unsurprising if we consider who benefits most from asymmetries of power/knowledge in the insurance industry. Hint: It’s not the consumer.

I have ordered the logics so they flow with the ‘insurance value chain’. That chain is long, complex, and fragmented, so the tracking is not perfect, but the sequence helps provide a meta-structure and pull-out themes across the logics. For example, Logic 1 (ubiquitous intermediation) and Logic 2 (continuous interaction) both contribute to increasing the frequency and intensity of contact between insurers and consumers, as well as the datafication and automation of that contact. Logic 3 (total integration) and Logic 4 (hyper-personalization) are meant to help insurers get a more complete and intimate understanding of people’s lives, and then use that information in how they develop products and target customers for interventions. Then Logic 5 (actuarial discrimination) and Logic 6 (dynamic reaction) incorporate these data-driven capabilities into core decision systems like pricing risks, combatting fraud, and handling claims. Together these logics describe how enduring ambitions and existing capabilities are motivating the future of how insurers engage with customers, data, time, and value.

Logic 1: Ubiquitous intermediation

Ubiquitous intermediation describes how insurers are trying to insert themselves into more and more interactions. For instance ‘embedded insurance’ products integrate the option to buy coverage into the purchase process for insurable goods or services. These forms of ‘add-on insurance products’ have been around for a long time (Baker & Siegelman, 2013)—extended warranties for electronics, damage waivers for car rentals, or travel insurance for trips—but now the mechanisms for creating, targeting, and selling them has changed.

Currently, most embedded insurers are focused on the e-commerce checkout process, such as by including a checkbox to add a ‘protection plan’ on an online payment page. The insurance is thus embedded into the purchase. However, the industry has begun gauging consumer interest in (and tolerance for) banks offering embedded insurance based on account-holders’ transactions (PYMNTS and Cover Genius, 2021). If your bank sees that you bought a couch or booked a holiday, they could contact you directly to offer a targeted and personalized coverage plan (sold by a third-party insurtech firm, and underwritten by yet another firm that carries the risk). As the industry becomes increasingly digitized and fragmented, the number of intermediaries operating—and the opportunities for intermediation—in the ‘(re)insurance value chain’ have proliferated (EIOPA, 2020).

Insurance market analysts often frame embedded insurance as a solution to what’s variably called the ‘insurance gap,’ ‘coverage gap,’ or ‘protection gap’—‘the difference between the level of coverage that is economically and socially beneficial and what’s actually bought’ (Torrance, 2020). Like the term ‘better’, the real question is ‘beneficial’ for whom. For insurers, the answer is obvious: embedded insurance ‘could account for over $700 billion in gross written premiums by 2030, or 25% of the total market worldwide,’ and ‘could create over $3 trillion in market value—for those who enable it’ (Torrance, 2020). At the same time, the form of add-on insurance that embedded insurance evolved from has been, for decades, (1) recognized by behavioural economists as an irrational purchase in almost all cases due to the cost vs. loss ratio being so high and (2) routinely investigated by regulators due to excess profits generated by such products (Baker & Siegelman, 2013).

Thanks to e-commerce distribution and digital intermediation, the market for embedded insurance has scaled exponentially. For example, Cover Genius, an Australian embedded insurance platform, was founded in 2014 and is now valued at a billion dollars, operates in 60 countries, has sold more than 10 million policies and reported a 667% increase in annual sales from 2021–2022 (Cover Genius, 2022). Extend, a start-up that aims to be the ‘Apple Care for everything that’s not an actual Apple product’ (Lunden, 2021), was founded in 2019, recently valued at $1.6 billion, has sold 300,000 ‘protection plans’, and formed partnerships with ‘hundreds’ of manufacturers and retailers. The huge potential for value capture via ubiquitous intermediation has caused many start-ups and incumbents alike to rush into embedded insurance while the market analysts and consultants sing its praises as the future of insurance.

The goal for embedded insurers is twofold: to gain what’s called a ‘situational monopoly in which the seller has a captive market for that purchase’ (Baker & Siegelman, 2013, p. 39), and to turn insurance into an everyday product rather than occasional service used to secure big things like cars, homes, health, life. There are stark similarities with another rising fintech trend of ‘buy now, pay later’ (BNPL). In this case, tech companies offer easy loans and payment plans on everyday purchases at the point-of-sale, thus also taking advantage of a situational monopolies and turning consumer credit into an everyday product, rather than an occasional service for large purchases. Indeed, embedded insurance and BNPL aim to operate side-by-side. As the CEO of Extend stated in an interview with TechCrunch, ‘Same day delivery, buy-now-pay-later, and other tools: we are now all part of that core e-commerce toolset’ (Lunden, 2021). So, for example, you buy new headphones through a payment processor like Stripe, using a BNPL like Afterpay, while bundling in a warranty with Extend.

The past two decades have seen an explosion of this kind of hyper-intermediation under the rubric of platform capitalism as (third-party) tech companies seek to insert themselves into every social interaction and economic transaction (Langley & Leyshon, 2017; Sadowski, 2020a; Srnicek, 2017). Now the insurance industry wants in on the action—especially reaching those who may be unable to afford traditional forms of insurance and thus represent uncaptured value for the industry—by being yet another ubiquitous (and unavoidable) intermediary in society.

Importantly, the goal of ubiquitous intermediation is not only about creating new opportunities to engage consumers and new insurance products to sell them. It’s also about the insurer being able to better do what it has always tried to: directly and indirectly mediate our habits, behaviours, choices, lifestyles, and environments, like an actuarial angel on our shoulders telling us how to reduce risk—and taking note when we don’t.

Logic 2: Continuous interaction

The second logic could also be called everyday insurtech. I use the term ‘everyday’ to indicate a core design principle motivating new products and relations, which is premised on insurers’ interest in using technology to multiply their points of contact with consumers. Rather than existing interactions that only occur occasionally—initiated by specific events such as signing up for a new contract, renewing an annual policy, or filing a loss claim—the goal here is to create an ongoing relationship based on everyday monitoring, reporting, managing, and nudging. Of course, this relationship is not one based on parity of position. Insurers aim to be the ones doing most of that active engagement in the form of, for example, smart devices used to collect data about consumers behavioural lifestyle and push notifications that recommend behavioural changes (Barry & Charpentier, 2020; McFall, 2019; Röschmann et al., 2022).

This is a massive change from the standard insurance relationship in which you might interact with your insurer only a couple times a year, and when they thought about you mostly in abstract and aggregated terms. Now insurance companies are keen to build relationships with customers based on constant contact. This type of relationship provides far richer veins of data for insurers to tap: more concrete, more granular, more detailed, more direct, more varied, and more of it. At the same time, it also provides insurers greater opportunities to engage in, and further refine their methods for, risk management and loss prevention. Every interaction with an insurer can become a potential intervention point for changing people’s behaviours—through information, recommendations, and policy conditions—so they more closely align with actuarial models of a ‘good risk.’

For example, the insurtech firm HiRoad (a wholly owned subsidiary of State Farm) requires policyholders to use an app that measures their driving via the smartphone’s sensors. According to an explanatory video on the HiRoad website:

The HiRoad app provides feedback in the form of four driving scores. Distraction Free is about keeping your hands off your phone so you can focus on the road. Driving Patterns looks at where, when, and how much you drive. Safe Speeds measures if you’re courteous on the road and avoid reckless driving. Smooth Driving shows how well you accelerate, brake, and take corners. These scores are based on a holistic view of how you drive over time. (HiRoad, n.d.)

The app shows a driver’s current scores, trends in how their driving has affected scores, the discounts earned by having good scores, and driving tips for how to improve scores. The explicit purpose of the app is to help customers ‘build better driving habits’ and ‘drive more mindfully’ through multiple channels of feedback based on continuous interactions and regular interventions. HiRoad is somewhat unique in terms of the number of scores it creates for drivers and the moralistic language in how it frames better driving (i.e., taking the high road, promoting mindfulness, being conscientious). But it is far from alone in the growing focus on behavioural interventions powered by big data analytics in insurance (EIOPA, 2019; Meyers & Van Hoyweghen, 2018).

Consider another example focused on home insurance. Jon-Michael Kowall, then the Assistant Vice-President of Innovation at USAA, said in 2016 he wants to create a suite of tech that acts like a ‘check-engine light for the home’ (Higginbotham, 2016). The idea is to fill each customer’s home with sensors that oversee everything, from leaky pipes to daily routines, and send status reports to the insurance company. This data can then be used to notify customers about potential issues, such as maintenance tasks—it’s time to replace your pipes (on pain of denial of coverage)—and even ‘whether or not a child made it home from school on time’. As Kowall goes on to say, ‘In the near future, you’ll give us a mailing address and we will send a box of technology to you. What’s in the box will prevent claims and also offer a better service to policy holders’ (Higginbotham, 2016).

Seven years later, insurers are making these plans into reality. It’s increasingly common to offer free or discounted smart devices to customers in exchange for the continual oversight and information they provide. The Australian home insurtech start-up Honey recently partnered with a tech company Notion (owned by telecom conglomerate Comcast) to provide its customers with a free ‘smart home sensor kit’. In return for promised discounts and faster claims, customers install the sensors all over their house and give Honey access to the data streams.

Recently, State Farm made a $1.2 billion investment in ADT, making State Farm the second largest shareholder of a major American home and business security company. This is after Google made a $450 million investment in ADT for a 6.6% stake in 2020. As State Farm’s Chief Operating Officer said, ‘This partnership with ADT gives State Farm the opportunity to provide Smart Home technology that takes us from our “repair and replace” model to a “predict and prevent” mindset’. In addition to acquiring ownership stakes, there is also another $300 million from State Farm and $150 million from Google being invested into ADT to fund product development, technology innovation, and customer growth (Ben-Hutta, 2022).

Tech, security, and property companies now partner with insurers on everything from smart locks and smart appliances to security systems and environmental sensors. The data-for-discounts deals make sense as straightforward transactions wherein consumers trade intangibles like data in return for money and technology. It’s not hard to imagine that soon (if not already) one of the biggest drivers of adoption for stuff like smart home devices, vehicular telematics, or fitness wearables could be subsidies from insurance companies.

All of these sensors, devices, and gadgets are designed to transform the insurer-consumer relationship so that it’s no longer based on discrete events (e.g., annual medical check-up) but on continuous interactions (e.g., personal health monitor). A review of big data analytics in motor and health insurance by the European Insurance and Occupational Pension Authority (EIOPA) highlights how insurers have long provided ‘risk prevention and mitigation services’ like annual medical check-ups. ‘However, while these services were traditionally based on a static set of information and medical criteria,’ with data from devices like vehicular telematics and health monitors, ‘it is possible to process more granular information on a continuous basis and therefore offer more tailored and timely services to consumers’ (EIOPA, 2019, p. 20).

Logic 3: Total integration

‘All data is credit data’ is the old slogan of ZestFinance, a start-up focused on creating alternative credit scores by analysing a wide variety of data about people, beyond the parameters of more traditional credit score. The start-up has since changed that name and slogan, now going by Zest AI and promising ‘more approvals, less risk’ thanks to its accelerated, automated, ‘machine learning underwriting’ for lenders (Zest AI, n.d.; see also Aitken, 2017). The problem is cast not as an epistemic one, but a technical one: Do we have the systems needed to make sense of the data, to make the data talk? That technical framing then provides support for a normative conclusion that has gained influence in the industry: Since any ‘data might tell the insurer something about risk’, insurers have the right, even the obligation, ‘to collect any and all data’ (Minty, 2022). We can see similar claims that ‘all data is X data’ applied in other areas governed by regimes of risk management and predictive modelling, such as health, policing, investing, and insurance (Sadowski, 2020b).

Insurers have long thought that all data could be or should be insurance data (Bouk, 2015; Horan, 2021; Yates, 2009). Datafication, as both a socio-technical process of making data and a political economic regime of capitalizing data (Sadowski, 2019), provides a basis for the translation. The technical capabilities are now available as sprawling networked systems designed to create and process a ‘universe of data’ have become pervasive (Sadowski, 2019). Enough data is captured about each individual, from a such variety of sources, that every person can be a universe of data unto themselves. Data can be supplied knowingly and voluntarily through traditional sources (e.g., a questionnaire about demographics, medical history) and alternative sources (e.g., fitness wearables or vehicle telematics). Data can also be captured from a vast range of ‘extrinsic’ sources from which ‘the average consumer would not legitimately expect insurers to collect and process to evaluate risk and price their policy’ (Bednarz & Manwaring, 2022, p. 10). These include, for example, ‘social media, Internet browsing history, websites’ cookies, retail loyalty schemes, credit cards, smartphone applications, wearables, connected cars, smart home devices, other private and public spaces with embedded sensors (such as workplaces, shopping centres and Internet-connected bus shelters)’ (Bednarz & Manwaring, 2022, p. 10).

All of this data can then be combined to provide a fuller, deeper, more granular profile of the insurance subject than was possible until recently. This ‘data fusion’ can allow strong inferences to normally private information, such as about sexuality, political ideology, health conditions, or home address. In other words, data fusion is a way of breaking down the firewalls between different databases.

The markets for data have developed to the degree where highly precise and personal data can be purchased in large quantities with great ease, with few questions asked about the provenance or purpose of that data. For example, the popular family safety app Life360 sells exact and near real-time geolocation data to the ‘mobility data and behavioural insights’ company Arity, which was founded by the large American insurer Allstate (Keegan & Ng, 2022). That data becomes one more factor that can be included in insurance activities like targeted marketing, product development, actuarial risk analysis, underwriting policies, and/or claims management. These industries and markets have largely escaped social restraints thanks to the scale and speed of data capture, plus the complexity and opacity of algorithmic analysis combined with their secretive and brazen practices (Bednarz & Manwaring, 2022). Legal rules, if any, are easily avoided or arbitraged, while the public is largely left in the dark about the existence of these systems and the extent of their impacts.

To obtain data for fusion, insurers have already begun partnering with manufacturers of networked, datafied, automated technologies, with even more partnerships between these industries coming. They offer rebates to offset the price of getting a smarter bedroom, smarter kitchen, smarter car, smarter whatever. In return, you simply grant them access to the real-time data from those things. The next thing you know, your fridge could be talking with your Fitbit and both of them could be informing your insurer about your eating and exercise habits. Or, as a next step, you could be incentivized to allow insurers (and/or their authorized agents) to remotely control, say, a smart security system connected to your home. Nothing beats the sense of safety (mitigation of risks) provided by the constant vigilance of a professional indemnity agency.

The trade-off of datafication for discounts, control for convenience, may seem innocuous now. But discounts quickly become penalties once expectations about data disclosure shift from novel to normal. As surveillance by insurers ‘becomes more accepted, it will give rise to its own stigma: When disclosure becomes low-cost and routine, those who hold out are suspect’ (Peppet, 2011, p. 1180). Impeding the flow of data, even just to maintain some privacy and dignity, ‘may carry with it the presumption that one is hiding information’ (Peppet, 2011, p. 1180). Insurers are already primed to see every customer as a potential liar and fraudster (Davey, 2020). Many major innovations in the industry are driven by fighting information asymmetries and rooting out fraud through devising better ways to know more about and have more power over consumers (Baker, 2003; Bouk, 2015).

Insurers may test out new products, practices, or programs to volunteer consumers, making them more mandatory as they meet insurers’ goals. That mandate may be official, such as by requiring certain data disclosure or device usage for coverage, or unofficial, such as by penalizing people who don’t disclose data or use devices. Of course, there are no assurances that acquiescing to a mandate will not also lead to penalties as datafication affects risk assessments and behaviour recommendations.

The logics described here can interlock and reinforce each other. Considering just the three logics outlined so far, we can already see how connections are being drawn between ubiquitous intermediation, constant interaction, and total integration—with the remaining logics of hyper-personalization, actuarial discrimination, and dynamic reaction easily plugging into the new regimes of actuarial governance.

Logic 4: Hyper-personalization

Like the convenience of total integration, hyper-personalization is an easy sell to consumers. Everybody wants to feel that their insurer really understands them, providing services and products that match their needs and preferences—all delivered with a personal touch. Personalization has already become a central feature of the user experience and consumer market for digital technologies. We expect personalization to be built into the design of products. We enjoy personalization in applications such as recommender systems for music or shopping (Seaver, 2022). We tolerate personalization in applications like targeted advertising on social media (Crain, 2021). We likely do not even recognize some personalization until it goes away, such as when your internet browser’s cookies are deleted, and your suggested searches and saved logins are cleared. In the name of convenience—which Huberman (2021, p. 337) identifies as a core ideology of surveillance capitalism, playing an important role in ‘sustaining and perpetuating’ this techno-economic system—remarkable degrees of data-driven personalization in all aspects of life have become normalized. These technological and cultural developments in hyper-personalization have, in turn, paved the way for further advances in yet another longstanding aspiration of the insurance industry.

In their analysis of life/health insurance policies that use self-tracking data from fitness wearables, Jeanningros and McFall (2020, p. 7) explain that these programs are mostly focused on creating data-driven engagement with policyholders based on ‘behavioural assessment, improvement and reward’. This is the idea behind programs like Vitality, ‘the market leading brand in behavioural insurance’ owned by the South African company Discovery Limited and delivered through franchise agreements with insurance providers around the world (Jeanningros & McFall, 2020, p. 1). Through incentives like subsidized Apple Watches, gamification through earning points, and rewards partnerships with companies like Starbucks and Amazon, insurers hope to induce people to track and act on their personal health data. ‘The theoretical claim that people can be gently nudged away from chronic, preventable, lifestyle-related disease toward better, healthier behaviour rooted in behavioural economics is the practical basis of the Vitality program’ (Jeanningros & McFall, 2020, p. 11). In reality, these behavioural insurance programs are used more as branding initiatives, which insurers hope will create more intimate and stickier relationships with the kind of customers who might already be concerned about exercising and eating well.

The life/health insurance industry largely does not use these programs to price risk directly (EIOPA, 2019) as much as to induce behavioural changes in policyholders. This is not because insurers are uninterested in (mandatory) use of such personal data for underwriting and sorting individuals, but because currently the benefits may not outweigh the technical, infrastructural, and reputational costs of using that data. It is a different story with car insurance and the use of telematics or ‘black boxes’ that track where, when, how individuals drive and translate that directly into personalized premium prices (Barry & Charpentier, 2020). In these cases, the branding of such programs have shifted away from slogans such as ‘Know yourself; Improve yourself; Enjoy’ (Jeanningros & McFall, 2020), instead being couched in terms of fairness and accuracy: ‘pay-how-you-drive’ (EIOPA, 2019). (The next two sections focus on examining fairness as actuarial discrimination and accuracy as dynamic reaction.) Rather than promising better living through wellness programs, these tailored policies price your risk using the most accurate data about your behaviours which means you pay only what you owe, not an amount based on the aggregate actions of other people.

In 2019, EIOPA published a ‘thematic review’ of ‘big data analytics in motor and health insurance,’ which was based on an in-depth questionnaire completed by 222 insurance firms from 28 Member States, plus 24 National Competent Authorities (country-level financial regulator) and two national consumer associations. The review is a wealth of information and analysis that reveals a ‘strong trend towards increasingly data-driven business models throughout the insurance value chain in motor and health insurance’ (EIOPA, 2019, p. 6). The review’s findings demonstrate that insurers are self-consciously pushing forward the logics described here, with voices from the sector stating they expect great advancements in abilities and applications over the next few years. It’s worth quoting at length from EIOPA:

Most insurance firms consider that BDA [big data analytics] will enable them to better understand their customer’s needs and characteristics and therefore allow them to develop more personalised products and services. Firms expect BDA will transform product development processes and product customization through the ability to identify underlying patterns in extremely granular data, coupled by the ability to capture and use increasingly available behavioral data from consumers. Insurance firms consider that BDA enables them to develop more granular risk assessments and better segmentation of consumers by means of assessing the risks in areas and segments that was not possible in the past. This results in the definition of new risk factors that enable the development of new products, both in motor and in health insurance, focusing on specific targets, markets and groups of coverage. … In this context [of increased market competition], insurance firms have started to adopt more sophisticated BDA-driven pricing models in order to optimize the profits with the help of the new possibilities offered by technological developments and new data sources. This has enabled a more granular segmentation of risks, increasing the effectiveness of risk selection, and allowing more risk-based pricing. This trend has also influenced the number and type of rating factors used by insurance firms in their pricing and underwriting models, both during the quoting process as well as on the renewal stage. (EIOPA, 2019, pp. 18, 29)

For consumers, hyper-personalization is marketed by insurers as a source of interactive policies, self-knowledge, subsidized rewards, potential discounts, convenient services, accurate assessments, fairer prices, faster claims, speedy payouts, and more. The shadow side of hyper-personalization in the way insurers envision eventually enact it—let alone if taken to its unconstrained logical conclusion—is the demutualization of insurance. In other words, insurance is emptied of its function in society as a way of creating security, sharing risks, organizing mutual aid, and forming bonds of solidarity across communities of different people (Lehtonen & Liukko, 2011). The CEO of Progressive, a major insurer in the United States, has called this logic ‘the statistics of one’. This is based on the idea that insurers will have so much data about each person, they won’t have to rely on analysing aggregate data and models about populations. The ideal scenario for insurers is one where the risk pools have been drained. Every person occupies their own risk pool—or more like a risk puddle.

The logic of hyper-personalization in insurance exists in different simultaneous states of development, with some segments of insurance further advanced in its application than others (McFall & Moor, 2018; Moor & Lury, 2018; Swedloff, 2020). Of course, we should be careful not to buy into the hype of industry claims, mistaking marketing for reality, or assuming that the logic would ever reach an ultimate endpoint of full actualization. The practice of insurance will never escape the necessity of averaging and smoothing out differences among people grouped together in the same risk classification. Even if those classifications get smaller and more refined, there will still be risk pools and statistical probability. But the power of an aspiration is its ability to motivate changes. Actuaries were discussing ‘individualized risk making’ over a hundred years ago (Bouk, 2015, p. 56) and they will likely be working on it a hundred years in the future. Insurers are quite clear about the direction they think (and hope) personalization is heading in the sector. Even if that goal is unreachable, the consequences can be found in the processes and impacts of what insurers do as they push this logic further.

As the EIOPA (2019) review explains, there are different views within the industry about the feasibility of hyper-personalization to the degree of (near) individualized pricing. A large tranche of insurance firms surveyed ‘declared that they will be able to use BDA to move towards individualized policy pricing’ in the near future (EIOPA, 2019, p. 38). Other firms just saw BDA as a continuation of what they already do—‘augmenting and enhancing, rather than replacing, existing pricing techniques’ (EIOPA, 2019, p. 38). The last group stated BDA will increase granularity of pricing and segmentation, but ‘it will not be possible (or desirable) to move to a segmentation of one’ (EIOPA, 2019, p. 38).

The issue of draining risk pools in insurance is not just a purely hypothetical worst-case scenario put forward by external critics. Indeed, industry insiders at the very top have also warned that this scenario is a realistic problem if the logic of hyper-personalization is left unchecked. As Colm Holmes—then CEO of Aviva and now CEO of Allianz Holdings, both massive multinational insurers—said in an interview: ‘The use of data is something I think regulators will have to look at, because if you get down to insuring the individual, you don’t have an insurance industry—you just create people who don’t need insurance and people who aren’t insurable’ (Littlejohns, 2020). An insurance industry supercharged with AI and data isn’t inherently a bad thing. Yet there is little reason to believe that the industry overall won’t use the power of personalization and other innovations to increase its own profits.

More practically, the problem comes when these technologies translate into new methods for squeezing more out of customers and shirking obligations to pay claims. We don’t have to reach the point of artificial intelligence underwriting creating a crisis of demutualization as every person is put in their own risk pool of one. It is already hard enough to combat practices of ‘price optimization’—an industry euphemism for price discrimination—where insurers analyse non-risk-related data, to target people with personalized prices that reflect how much they will pay, not their risky behaviours (Swedloff, 2020). Such optimization can also extend to paying claims, ‘where the compensation paid to the consumer suffering a loss does not only depend on objective facts like the damage, cost for repair, medical expenses etc,’ but also on how much (or how little) each consumer is likely to accept (EIOPA, 2019, p. 47). This kind of optimization has regressive impacts where the most vulnerable and already disadvantaged people—those who are poorer, older, less educated, for example—are also most likely to be put in a position of having little other choice but to accept higher prices and lower payouts. A new report by Citizens Advice, a UK financial assistance organization, notes that discriminatory pricing introduces an ‘ethnicity penalty in the insurance market’ (Cook et al., 2022). Now such practices—and the sensitive data sets they rely on—can be laundered through opaque machine learning, thus giving human actuaries plausible deniability when discrimination and deception is uncovered (Burrell, 2016; Prince & Schwarcz, 2020; Swedloff, 2020).

Logic 5: Actuarial discrimination

The ability to collect, analyse, and connect more data about more risk factors opens the way for scores and judgments based on seemingly arbitrary correlations. It doesn’t matter if insurance companies know why people who drink coffee after 5 p.m. or have low credit scores or are implicated by whatever other random factor may correlate with higher risk. What matters is that the pattern has been identified in the data and can be turned into ‘actionable insights’ that justify price discrimination. While this describes the opaque operations of machine learning (Burrell, 2016), such practices fit well with how actuarial calculation already largely works (Glenn, 2003). Both are built on an epistemology less concerned with knowing why a relationship might exist and more with showing a probabilistic connection. If you ever want to frustrate an actuary, start by asking them to explain the causal validity of factors used to assess, predict, rate, and price risk. Actuarial decisions by insurers do not need to be—and indeed very often are not—justified by clear causal relations or based only on objective facts (Glenn, 2000). Even if such requirements happen to be mandated and enforced by regulators, the subjective judgements, assumptions, and stories that constitute actuarial practice are still unavoidable (Glenn, 2000; Porter, 1995). The political decisions are hidden under technical decisions—starting from what datasets are included, what questions are unasked, what interpretations are used—and the techno-political confabulation only intensifies from there (Barry & Charpentier, 2022; Horan, 2021; Simon, 1988).

With both machine learning and actuarial science, more direct forms of knowledge, such as your answers on an insurance questionnaire, might be treated as less truthful than proxy inferences about you based on other sources. As Neil Sprackling (2022), President of US Life and Health at Swiss Re, explained in a video interview for credit rating agency AM Best, ‘We’ve got a number of [predictive models] going. One specific example would be, we’ve got something called a smoker propensity model: It can predict whether you smoke or do not smoke without actually asking you the question or testing you for it.’ People lie all the time. Insurers would rather trust indirect data about consumers than direct information from consumers. More generally, ‘[e]vents in data science are constituted not from experiences but from those traces of experience which can be datafied’ (McQuillan, 2018, p. 261). In this political epistemic system, ‘direct apprehension’ of information is devalued because it centres an inferior subjective viewpoint, rather than a superior objective god’s eye view (Haraway, 1988). Or in lieu of god, the machine will do. The epigraph for this article—about how an insurance executive claims that artificial intelligence, thanks in part to his firm’s $500 million investment into data analytics company Palantir, allows the insurer to learn things ‘only god knew about’ (Sugiura & Lewis, 2022)—shows the extent to which these metaphors of machinic power being next to godliness are taken seriously by industry leaders.

There is now a whole cottage industry of companies who claim that their technologies can discover truth and detect fraud by processing everything about consumers except what they say. These include start-ups like ForMotiv, which calls its machine learning model a ‘patent-pending Digital Polygraph’ that analyses ‘thousands of behavioral cues, or someone’s “digital body language”, collected while they engage with a form or application’ (ForMotiv, n.d.). This data includes how long it takes to complete an online form, information typed into fields then deleted, and the movement of your cursor on the page. That these companies make explicit reference to contested science like lie detection to describe their service’s spurious capabilities seemingly matters little to the major financial and insurance institutions that use these services and treat them as accurate. Here truth is judged not on scientific soundness, but statistical significance for profit-making.

With the introduction of new techniques like ‘accelerated underwriting’ via machine learning, we once again see an amplification of already existing practices and longstanding problems (Barry & Charpentier, 2022). The system—the scale of data, the power of computation, the speed of decisions—derives legitimacy and authority from its own (claimed) capacity. I’m reminded of a quote by Luckey (2018), billionaire founder of Oculus and Anduril, who stated while speaking at the Web Summit conference: ‘technological superiority is a prerequisite for moral superiority’. Fundamentally, insurance is a regime premised on employing superior information technology to establish and enforce a superior moral economy of risk, worth, merit, and the fair administration thereof (Fourcade & Healy, 2013, 2017; Zelizer, 1979).

There is a clever rhetorical framing in insurance discourses where industry executives and experts are loudly and publicly against ‘unfair discrimination’ (Meyers & Van Hoyweghen, 2018). The sentiment is expressed in white papers, textbooks, conferences, webinars, and everywhere else. In an interview for trade magazine The Actuary, Sprackling said the industry is retiring the term ‘proxy discrimination’ because ‘there is no widely accepted definition’ and is returning to ‘unfair discrimination’ (Abrokwah, 2022). A term coined as far back as 1909 by a Kansas insurance regulator, ‘unfair discrimination’ continues to be a popular and productive concept for the industry (Miller, 2009). Everybody can agree: ‘unfair discrimination’ is unintended and unwanted. However, if we pay close attention to the discourse, we’ll find that those two words are always conjoined for a very specific reason. Doesn’t the term seem redundant? The problem for insurers is not discrimination, but the unfair modifier. Why? Because insurers have a very specific definition of ‘fairness,’ which is based on actuarial judgements of inclusion and pricing in risk pools.

Discrimination is fundamental to the operations of insurance companies. In simple terms, discrimination in insurance starts with two parts: Actuaries use statistical analysis of data to create risk classifications and underwriters then decide how to assign individuals to each risk classification. These classifications form the basis for how different people are treated by insurers in terms of their policies, prices, coverage, claims, and so on— if they are even offered insurance at all. Methods of classification and discrimination have changed over time based on what is financially profitable, technologically possible, and socio-legally permissible (Krippner & Hirschman, 2022). But what remains essential is the ability to discriminate based on the risk classification that insurers create and assign to individuals.

Take away this ability to discriminate and you simply do not have a private insurance industry. But discrimination is also bad public relations; especially now, with so much critical attention on social justice and algorithmic bias. As noted in a paper by the Chief Risk Officers Forum, a European industry group, about the ‘right to underwrite’—or the ability to classify, price, and choose how insurers will or won’t cover risks—the term discrimination is ‘usually associated with a negative bias and synonymous with prejudice,’ which is why they advocate for using the term ‘differentiation’ instead (CRO Forum, 2012, p. 3). The paper was motivated, as they note, by the industry’s worry that EU anti-discrimination law ‘poses a major challenge to the insurance industry as it restricts fundamental insurance pricing’ (CRO Forum, 2012, p. 2). When insurers loudly decry ‘unfairness’, they are hoping that people will assume insurers are also against discrimination, while also assuming that we all share the same conception of fairness. In reality, insurers view discrimination as a purely ‘neutral’ action that is justified by their highly idiosyncratic idea of ‘actuarial fairness’ (Meyers & Van Hoyweghen, 2018), which can, and often does, run counter to more commonly held moral conceptions of fairness.

At its base, actuarial fairness is a theory of justice based on paying only what you owe, and receiving only what you are owed, according to your own risk (Heras et al., 2020; Landes, 2015). An insurance policy is fair only if the price an individual is charged accurately reflects their own risk exposure—based on an ever-evolving set of factors and methods used to calculate that risk—and if people bearing the same risk are charged the same price. Similarly, an insurance claim is fair only if it is based only on actual losses within conditions of appropriate risk mitigation. Within this definition, any form of cost shifting or cross-subsidizing within risk pools is unfair, even if it is done to achieve other forms of social fairness. With more sophisticated methods for measuring risk based on more detailed rating criteria, and thus the creation of more precise segmentation of people, fairness ‘became a matter of computation capacities, rather than a philosophy of shared losses’ within a community (Barry, 2020, p. 174). This shift was already occurring by the 1950s and only ramped up with rapid advances in digital computing and bigger data from the 1970s onwards.

In their own way, the other logics and technologies described in this article all contribute to, and are created for, the pursuit of actuarial fairness. A report by the Australian Actuaries Institute (2016, p. 4) on the impact of big data on the future of insurance plainly states that ‘increased individual risk pricing will make premiums fairer in that they will be more reflective of that risk’. Underwriting actuarially fair prices requires information and discrimination—the former to assess risk, the latter to sort it. With detailed lifestyle and behavioural profiles, insurance companies argue that their prices will be more accurate, more complete, and thus fairer. Premiums will truly reflect the choices each person makes and the risks they assume. Those who lead safe, careful, or uneventful lives and demonstrate their willingness to make decisions along the lines the insurance company demands won’t have to bear the load of those who are risky and rash, or more recalcitrant to such recommendations.

As Kiviat explains in a study of the political controversy in the US over using credit scores in insurance pricing:

Actuarial fairness does not avoid the moral minefield. It simply, if implicitly, holds that people are always accountable, regardless of whether the data look the way they do because of personal fault, structural disadvantage, simple chance, or some other factor. Algorithmic prediction is imbued with normative viewpoints—they are viewpoints that suit the goals of corporations. (Kiviat, 2019, p. 1151)

In a history of the moral justifications used by the insurance industry over the last 150 years to justify risk classification and actuarial discrimination, Baker writes:

In practice, the fairness argument has been mobilized in public policy debates, not to protect the rights of low risk individuals, but rather to promote the freedom of insurance organizations to classify insureds through any means they wish. While some 'low risk' individuals may believe that they are benefited by risk classification, any particular individual is only one technological innovation away from losing his or her privileged status—the reality that lies behind the widespread concern with genetic testing by insurance companies. … This is not to say that risk classification is always and everywhere a bad, but rather that the fairness justification for classification does not carry all the moral force that its proponents assert. (Baker, 2003, pp. 275–276)

In 2011, the European Union Court ruled that insurers can no longer use gender as a risk factor for justifying different premium prices for individuals. However, the court decision was not based on some social ideal of gender equality, but rather on the theory of actuarial unfairness (Frezal & Barry, 2020). It was unjust for insurers to discriminate against people based on broad aggerate categories like gender, rather than based on data from each individual’s own actions and habits. In other words, the court sent a signal to insurers that they needed to speed up the progression of the logics outlined here. In response, the industry started investing further in telematics and other systems that can collect personalized behavioural data and contribute to justifications for discrimination and intervention based on factors that each individual can control and thus be held responsible for (Meyers & Van Hoyweghen, 2018). Whereas demographics and genetics are accidents of chance, behaviours are actions by choice. Of course, reality is far more socially complex, but those features must be smoothed away for the sake of moral clarity. If risk pools cannot become fully demutualized via individual assessment, then the next fairer option is to make them homogenized via behavioural analysis.

Logic 6: Dynamic reaction

Ours is an age dominated by dynamism: the annihilation of time by speed. The instantaneous communication of information for anticipatory decision. Temporal compression as social transformation.

Insurance—an industry that has made a public virtue out of its slow, ponderous stability, with names like Prudential and logos like the Rock of Gibraltar—has given in to the siren call of acceleration. Every phase of the insurance process is being innovated with speed in mind. However, it’s not just speed for the sake of doing things faster; it’s speed in the sense of dynamic reactions to changing conditions. Time is money, but action is power. Whereas previous logics have focused on getting data through interaction and integration, or justifying its use for personalization and discrimination, this last logic is about turning data into action. We can see how this logic is playing out by looking at some new capacities for dynamic reaction that are significant in their own right, while also stacking together to turbocharge insurers. The sequence in this section moves from ‘accelerated underwriting’, to ‘insurance-as-a-service’, to smart compliance.

Accelerated underwriting is based on using non-traditional data, machine learning, and predictive models to automate risk pricing for insurance policies. This term, currently, largely refers to developments in life insurance. The technique is designed to expedite the process of assessment, decrease the cost of underwriting, increase the acceptance rate for policies, and make application a less invasive experience for customers by waiving tests like ‘paramedical exams and fluid collection’ (NAIC, 2022, p. 2). Instead of these direct forms of information, accelerated underwriting relies on much broader variety of factors. According to a report by the US National Association of Insurance Commissioners, a non-exhaustive list of ‘variables used by some accelerated underwriting models include customer disclosures, prescription history, digital health records, credit attributes, medical information bureau data, public records, motor vehicle reports, smartphone apps, consumer activity wearables, claim acceleration tools, individual consumer risk development systems, purchasing history, behavior learned through cell phone usage, and social media’ (NAIC, 2022, p. 5).

There are no industry standards for the data and models companies use for accelerated underwriting. Each insurer is collecting its own data, creating its own models, and/or contracting with a data/technology provider. A report by Munich Re, titled, ‘Accelerated underwriting: The new paradigm for risk selection’—observes that, ‘[t]he traditional method of applying for life insurance is quickly becoming the exception rather than the rule’ (Munich Re, 2020, p. 1). Adoption and experimentation with these accelerated underwriting techniques is spreading beyond life insurance. As previous sections have explained, almost all other coverage domains are investing, investigating, and/or implementing the forms of alternative data and machine learning needed to bring acceleration and automation into underwriting. While these techniques can be transformative on their own, they can also support or catalyse other forms of dynamic reaction in insurance.

Our traditional relationship with insurance is both ongoing and occasional. That is, the insurance contract is always active, as long as we don’t break any conditions of coverage, but we only engage with it on a quarterly or annual basis at renewal time. The technical, social, and legal limitations of this relationship have restricted how often insurers could adjust premium prices and roll out new products. Insurers are now beginning to break through these limits as new digital systems and business models shift to time horizons of weekly or even daily possibilities for change.

The growing trend of ‘usage-based insurance’ or ‘behaviour-based insurance,’ which is most common in automotive insurance, is the next step away from traditional insurance towards more dynamic insurtech. These are the kinds of services, for example, discussed in earlier logics where insurers use vehicular telematics or ‘black boxes’ to track driving data, reflect it back to drivers as scores, provide feedback for how to improve, and incorporate usage or behaviour directly into premiums in the form of ‘discounts’ off a base rate. The emphasis on ‘discounts’ is not incidental. It is good marketing: Consumers react much more positively to price discounts than penalties, which some insiders euphemistically call ‘negative discounts’. And it is good legal compliance: Regulators in many jurisdictions restrict the ability of insurers to increase prices of policies during contracts, which is why some insiders argue for higher base rates to give more room for manoeuvring down. While usage/behaviour-based insurance incorporates alternative features for assessment and pricing, these products are often part of otherwise traditional insurance contracts where coverage persist continuously even when not actively in use (e.g., driving).

However, if we follow the logic of dynamic reaction even further, we reach (still early) experiments for service models called ‘on-demand insurance’ or ‘insurance-as-a-service’. Rather than a persistent contract, every activation of insurance is treated as a new transaction with a new contract, thus allowing for maximum flexibility in dynamic policies. For a clear picture of how on-demand insurance might operate, it is worth quoting at length from a recent academic review of the market landscape and business model:

In on-demand insurance, the contract ends when turning insurance off. A customer can choose to turn on insurance from different providers at different moments. The mode of activation can range from manual to automatic: coverage can be activated manually, recommended by a smart device and activated manually by the user, or activated automatically based on criteria like location, activity or context. (Röschmann et al., 2022, p. 608)

This multiplication of contracts is not merely a technological quirk. It’s also a way of innovating around regulations that restrict how often policy features like premiums and other conditions can be changed.

Although calculating adequate premiums in the underwriting process is more difficult, short contracts have the advantage that they open the possibility of continuous underwriting. Premiums can be dynamically adapted each time the customer seeks coverage. Hence, the contract wording and the price (experience pricing) can be adjusted at each activation taking into account newly available information or adverse development. (Röschmann et al., 2022, p. 620)

The on-demand model is where the rubber really hits the road for applications of systems like accelerated underwriting. This is still a nascent area of insurtech and a niche product in the market—and will likely stay that way for quite a while. Though, as my conversations with insurers and entrepreneurs in the sector indicate, there is growing interest in how the model can mature and how its features can be adapted for other insurance products. For many people working within the insurtech sector, the difference between optimists and pessimists is just a difference in the time/cost horizons for when the technologies will be good enough to support these capabilities. Throughout this article I have described how most of the techno-political logics are developing on a long continuum of intensification. However, eventually all of the logics hit a point where matters of degree become differences in type. Here we can see how doing more faster is not always just doing more of the same.

If we move to the next phase of the insurance process, compliance and claims, we can see how this logic is operating now and where it is heading. In addition to price adjustments, dynamic reaction also looks like the ability to better enforce compliance with policy conditions. Consider a particularly egregious example of how American insurers cover CPAP machines: doctor-prescribed masks worn while sleeping by millions of people with breathing problems. The machine is loud and bulky. The mask is constricting and awkward. As with any treatment for a chronic ailment, sometimes people cannot wear the uncomfortable mask every night. Sometimes they, or their partners, just want a bit of peace and quiet. Little did patients know, however, the machine was also a spy for insurance companies. The CPAP was recording and sending usage data to insurers, allowing them to track when and for how long the mask is used (Allen, 2018). Some patients found out the hard way that their insurer had been watching them sleep when they stopped covering the cost of the (extremely expensive) machine and supplies because the patients failed to strictly comply with the prescribed use. The CPAP can be lifesaving but go a few nights without wearing the mask and your insurer will decide you don’t deserve to have it anymore.

Refusing to allow insurers to audit your daily life and domestic habits raises a red flag. Maybe the insurer will hike your rates—or levy ‘negative discounts’—to reflect your lack of sharing. A claim might be denied because you weren’t using the required devices producing the right data, so the insurer assumes you are defrauding them. Or your coverage may just be cancelled altogether if you don’t agree to wear, install, and use the smart tech mandated—perhaps even provided free of charge—by the insurer. If, at some point, you decide not to lead a smart life, inhabit a smart home, and drive a smart car, then the insurer will be alerted so that it can adjust your policies and premiums accordingly (Sadowski, 2020b).

This is the flipside of usage-based insurance where insurance companies mandate people behave in certain ways. Mandates are not new features of insurance policies, but the ability to constantly monitor and instantly react to mandates has greatly increased. With these new powers also comes an expansion in the types of conditions that insurers can practically require, enforce, and account for when paying (or denying) claims. Pushing this logic to a more radical endpoint does not need to catapult us into a science fictional realm. It would just take more intensive application of already existing things: real-time data triggering smart contracts that automatically change premiums, payout claims, or cancel policies if certain parameters are met. Each of these things have independently been heralded by the insurtech sector as trends paving the way for the future of insurance, which means they have also received vast amounts of attention and investment. Bringing them together in a more synergistic approach, along with some modest technological advancements, is all it would take to push the logic of dynamic reaction to the next level and make the scenario into reality.

Conclusion

As legal scholars have observed, ‘within a regime of liberal governance, insurance is one of the greatest sources of regulatory authority over private life’ (Baker & Simon, 2002, p. 13). The industry’s ability to record, analyse, assess, sort, exclude, discipline, and punish people, in ways that have immediate, profound, material impacts on their lives, often surpasses the power of government agencies. We tolerate the existence of such a regime because insurance is meant to play an important role in society as an effective way to pool risks and provide mutual aid to those in need, whether because of personal choice, structural constraint, or random catastrophe. This is a model of insurance as a mechanism for organizing social welfare, a method for redistributing community losses, and a means for creating group solidarity. But that ‘collective and cooperative view’ of insurance is in direct competition with—and has long been directly opposed by (Walters, 1981)—the ‘individualistic and profit-oriented’ view of the insurance industry (Frezal & Barry, 2020, p. 128). It should come as no surprise which view of insurance is dominant. The former is only present symbolically in the marketing campaigns of an industry driven by the interests of the latter.

Insurance is not unique in succumbing to the imperatives of capital, but it is uniquely adept at exploiting people when they are most vulnerable by controlling access to essential services for pooling risk and managing uncertainty. Ultimately, as it is currently being implemented—surveilling populations, individualizing risk, enforcing compliance, disciplining policyholders—insurtech is helping push the industry even further away from public utility toward strictly private benefit. The industry seems dead set on developing in ways that undermine the foundational forms of collective security and cooperative solidarity that constitute the function of insurance in society. This is not a new trajectory by any means; but it is one that is intensifying in different ways. At the beginning of this article, I discussed the idea of total life insurance as a concept for thinking about how, in the not-so-distant future, insurtech could become far more interconnected, integrated, and intensified. As I’ve shown through analysis of six techno-political logics, elements of this future are already here, and we sit at the cusp of further developments in the powers of this governance regime.

Beyond being an analysis of insurance and technology, I hope this article also serves as laying out a critical agenda that attracts and guides even more scholars to study how these logics are materializing in the world—with their moral economies and technical systems, long histories and near futures. In doing so, we must pay particular attention to how each individual piece snaps into place, coming together to form a regime of insurtech as governance. What’s at stake is more than just changes to how we each individually interact with our insurers. Instead, this is about understanding how a powerful institution that is fundamental to the operations of modern society continues to change, and what dynamics and imperatives, whose desires and interests, are steering that change. The stuff of insurance is far too important to be left to the insurance industry.

Author biography

Jathan Sadowski is a Senior Research Fellow (ARC DECRA) in the Emerging Technologies Research Lab at Monash University. He works on the political economy of technology, with a focus on the FIRE sector. He is the author of two books, Too Smart (The MIT Press, 2020) and The Mechanic and the Luddite (UC Press, 2024), both on technology and capitalism.