Working against the clock: digital surveillance in US Medicaid homecare services

Introduction

In the United States, the role of government in care provision gained growing public attention as the country's care crisis was brought into stark relief during the Covid-19 pandemic. In nursing homes and other institutions, the isolation of quarantine lockdowns, understaffing and mishandled outbreaks, and bleak living conditions combined with longer trends of independent living and “aging in place” have together driven rising preferences among disabled and older adults for home-based care over institutional settings (Reilly, 2020). More Americans are seeking out home- and community-based services (HCBS), which are typically funded through Medicaid, the means-tested public assistance program that provides health coverage for low-income Americans. However, the workforce has majorly failed to keep pace with this growing demand (Kreider & Werner, 2023).

Labor shortages, driven in part by the poverty wages of Medicaid's low reimbursement rates, have sparked a reckoning over support for care workers. Care as a public good became controversial when the Biden administration, as part of its Infrastructure Plan, sought to introduce $400 billion in investment to support both childcare and long-term services and supports (LTSS). The latter workforce, which includes nursing assistants, home health aides, and personal care attendants, would receive pay raises, stronger benefits, and better skills training. However, Republican opponents argued that bridges, highways, and roads were infrastructure; caregivers were not. In an aired ABC News segment, former New Jersey governor Chris Christie sarcastically quipped, “Oh, no, now the care economy is infrastructure … I don’t even know what the hell the care economy is” (ABC News, 2021). Ultimately, lack of bipartisan support led to the funding being excised from the final version of the legislation. But the political moment resurfaced historic tensions in the labor politics of care and its role in the welfare state, between support for publicly funded care provision and views of the private family as the “primary source of economic security and a comprehensive alternative to the welfare state” (Cooper, 2017, p. 9). As will be discussed in this article, Medicaid HCBS has long been a site where judgments over deserving and undeserving state support of care are continuously litigated through the concepts of “fraud, waste, and abuse.”

At the same time as Congress was negotiating the bill's funding, a different type of care infrastructure was quietly being rolled out across the country. In 2016, Congress had passed the 21st Century CURES Act, which included a provision requiring all Medicaid personal care and home health care services that require a home visit to use EVV systems. This mobile app-based method of digital worker surveillance places greater state scrutiny on both workers and public benefits recipients by requiring them to verify care services in real-time. From the start, the mandate elicited vehement, mobilized opposition from disability advocacy organizations and labor unions, who argued that the EVV mandate would erode quality of care, jeopardize access to benefits, and push workers out of the sector (Mateescu, 2021). Meanwhile, state Medicaid agencies struggled to translate the mandate into policy and conduct roll-outs even as a deadly pandemic threw families and their networks of care into turmoil in the years that followed. EVV technology vendors, some of whom came from logistics contexts such as the long-haul trucking industry and had pivoted into homecare with little or no background in the industry, further exacerbated tensions between families and state agencies by designing far more invasive features into EVV systems than were legally required. This included GPS location tracking, digital geofencing that limited where services could be provided, requirements for real-time rather than asynchronous task inputs, as well as frequent biometric verifications through facial or voice recognition technologies.

This article investigates how the digital surveillance and labor management of publicly funded care workers shapes the experiences of care services for people with disabilities and older adults across the US. The findings draw on qualitative interviews with homecare workers, their clients, and disability and labor advocates, as well as analysis of a broad range of federal and state-level documentation and tech vendor materials, news coverage, and other sources.

Specifically, I approach EVV systems as welfare technology from two perspectives. Literature on work within the digital welfare state has in large part focused on how frontline administrators’ roles are reshaped by data infrastructures and automated systems, often shifting human discretion and decision-making in new directions (Buffat, 2015). The case of EVV draws attention to how algorithmic labor management converges with welfare state surveillance trends towards “obsessing about fraud, cost savings, sanctions and market-driven definitions of efficiency” (Alston, 2019, p. 20). Business models innovated by gig platform companies in ride-hail and delivery services are becoming the aspirational template for the care sector. Both public and private entities are investing in “care tech” companies that promise to manage care efficiently and at “scale” (Dalmer et al., 2022; Glaser, 2021; Mateescu & Ticona, 2020). Digital platforms mediating domestic labor in particular take advantage of historically embedded racial and gender inequalities as a central driving force of their business models (Gebrial, 2020; van Doorn, 2017). In the context of Medicaid HCBS, EVV systems promise the ability to remotely digitally monitor a disaggregated workforce through an app much in the same way as Uber manages its fleets of drivers. This article explores these dynamics through the lens of homecare workers’ and benefits recipients’ experiences with getting “flagged” by EVV systems. “Flagging” becomes a tool for more finely tuning classifications of different types of paid and unpaid care, mirroring the billing structures of such programs and the limitations of the technology itself. While EVV systems were implemented as a means to reduce “fraud, waste, and abuse,” these categories remain opaque and mysterious to the app's end users. Daily interactions with EVV systems entailed navigating mixed messages from state Medicaid agencies and other actors about what, exactly, EVV systems were measuring, and increased burdens of administrative and interpretive labor to avoid repeatedly getting flagged for “noncompliance.”

Second, I argue that the introduction of EVV systems surfaced long-standing tensions in the employment arrangements between homecare workers, public benefits recipients, and the state. These actors occupy contradictory roles: Medicaid-funded homecare workers are treated as not-quite public sector workers; benefits recipients as not-quite employers, and public long-term care programs as not-quite welfare. The introduction of EVV systems into this triadic employment arrangement destabilized day-to-day relationships, where workers struggled to repair and reconfigure daily routines in order to do their jobs. Medicaid recipients hold authority over their care workers, yet are themselves subjected to state surveillance and often have little control over how they receive care services. At the same time, homecare workers were acutely aware of how minor missteps in their digital timekeeping practices could imperil their clients’ access to public benefits. These mutual risks and precarities in some ways pitted interests against each other, but also created common cause for both labor and disability rights advocacy efforts in pushes to roll back EVV requirements (Scalia, 2019). What the case of EVV highlights is the ways that digital welfare technologies not only shape points of access to state support, but that changing legal, social, and economic relationships are felt at the level of mundane and intimate routines of everyday life.

Methodology

Research design and planning for this project began in August 2019, and fieldwork was conducted over the course of 18 months, from October 2019 to April 2021. Throughout this period, state Medicaid agencies across the US were at varying stages of EVV program implementation; some had only begun stakeholder engagement through town halls, webinars, and formal solicitations for public comment, or were in the process of selecting a vendor and implementation model as well as finalizing policy, while others had fully rolled out EVV systems. Because of the cross-state differences in EVV policies, as well as differing demographics, employment conditions, and urban–rural contexts, this study collected and analyzed publicly available information about EVV implementation in all 50 states, as well as relying on public trackers and industry summary reports created by nonprofit organizations such as the National Association of States United for Aging and Disabilities (NASAUD, 2018).

Due to the onset of the Covid-19 pandemic six months in to the study, fieldwork following March 2020 pivoted to remote methods, such as phone and video conference interviews. To comply with social distancing, in-person events hosted by various institutions were near-universally switched to webinar formats, enabling the author to remotely attend events across multiple states. Three broad areas of empirical research informed this study: First, semi-structured, qualitative interviews were conducted with 20 workers and Medicaid service recipients across the United States, in total including four home health aides, nine personal care attendants, and six client/employers. Participants were recruited through a combination of referrals, posting digital flyers in online forums like local job listings groups for hiring personal care attendants, and events such as information sessions. Second, this project tracked federal and state implementation of EVV through publicly available documents, third-party vendor marketing materials, and industry publications. The author conducted participant-observation of EVV worker training/information sessions and product demos hosted by vendors, state Medicaid agencies, and the Centers for Medicare and Medicaid Services (CMS) which was responsible for providing federal guidance on implementation and assessing compliance with the CURES Act mandate. Lastly, this study draws on local news reporting and online grassroots mobilizations from disability advocacy organizations, civil society organizations, as well as care worker advocacy groups and labor unions such as SEIU1199. The author also conducted informational background interviews with representatives from these groups, which informed research question design as well as interview protocols.

The rise of technologically scaled care infrastructure

Information and data technologies have long been a part of care systems, particularly healthcare settings such as hospitals. More recently, they are being introduced in settings like nursing homes and home-based care, where the quantification of care and human labor are increasingly enmeshed with public financing and private profits (Berridge & Grigorovich, 2022). The Wall Street Journal noted that the 2019 Consumer Electronics show was dominated by home health tech (Shlagman, 2020). In January 2021, the US Department of Health and Human Services announced its first ever Chief AI Officer, as well as aims to “accelerate AI-centered pursuits across HHS” (Vincent, 2021). These range from information and communication technologies (ICT), task automation, workforce monitoring and management tools, to automated decision-making, as well as clinical tools such as biometric wearables that quantify health and behavioral aspects of care (Corbyn, 2021).

In terms of welfare administration, the role of algorithmic decision-making tools in making public benefits determinations has been a focal point of both scholarly attention and advocacy efforts to fight unjust and dangerous cuts to critical services (Brown et al., 2020; Eubanks, 2018). In the US context, state governments have adopted decision-making tools to assess both program eligibility and to calculate allocations of care hours under Medicaid. Contestations over these algorithms have pointed to how they are often implemented with little public debate or transparency over how decisions are made, and have led to legal challenges over benefits recipients’ due process rights (Brown et al., 2020). In Arkansas, for instance, the state's Department of Human Services introduced an algorithmic system to allocate care hours for disabled service recipients of Medicaid Waiver programs, but this led to scores of individuals experiencing drastic cuts with severe consequences to their health and well-being (Lecher, 2018).

Another domain of government experimentation with care technologies has been with automating aspects of care labor, with investments in robotics, AI assistants, telecare, and workforce management software. These tools come with promises of greater efficiency, optimized workflows, and alleviating burdens of administrative paperwork. However, they have also been critiqued as instruments of austerity by other means, as well as for embedding forms of ageism, ableism, and racism into technology design (Berridge & Grigorovich, 2022; Mort et al., 2013). In 2022, New York State's Office for the Aging, for instance, purchased digital virtual assistants for older state residents, and companies like Amazon are increasingly seeking to contract with senior living facilities to use its Alexa AI assistant (BusinessWire, 2021; Osbourne, 2022). In the UK context, studies of telecare use in adult social care programs have shown how such technologies had the effect of “encroaching on valued aspects of care work, leaving mundane tasks to care workers and creating new responsibilities, relegating staff to ‘machine babysitters’” (Hamblin, 2022). More broadly, welfare reforms in the UK have generated controversy over cost-cutting practices such as replacing overnight homecare visits with remote telecare through a combination of tablets, motion sensors, microphones, and CCTV cameras (Reynolds, 2019). As tech marketing promises greater efficiency and scaling of care, new scholarship has examined how technology innovations often merely redistribute how care is provided, while further justifying worker invisibility, precarity, and low wages (Mauldin, 2020).

The case of EVV systems fuses together these two trajectories within datafied care provision by the state; on the one hand, the administration and allocation of welfare benefits and on the other, the rise of algorithmic management and digital surveillance of low-wage work. As will be discussed, the multiplicity of EVV systems’ function – whether it is a useful tool to support workers or a form of welfare surveillance – provides ambiguous cover for its most punitive applications. In its initial iteration, EVV was designed in the late 1990s by a Registered Nurse, who had sought to solve a common problem across field service occupations where employees are constantly on the move: one of data entry, oversight, and remote communication with management. Today, various design and industry efforts have indeed sought to leverage technology to better support home care workers and counterbalance worker invisibility and isolation by providing them with skills training, opportunities for peer support, and communications channels with a broader care team (Poon et al., 2021; Tseng et al., 2020). Direct care workers are most consistently in day-to-day contact with the people they support, yet they are largely excluded from definitions of frontline care (Hondagneu-Sotelo, 2007, p. 9). This exclusion is reinforced by legacies of racial and gendered occupational segregation and the devaluation of women of color in these occupations (Boris & Klein, 2015; Duffy, 2005; Nakano-Glenn, 2010).

Marketing materials and trainings directed at workers have generally framed EVV systems as a worker-support tool. However, the dividing lines between management and state surveillance may be tenuous and easily overstepped. In India, the rollout of the Shield 360 app, intended to monitor and update daily work targets of accredited social health activists (ASHA), who serve as community health workers supported by the Indian government, prompted mass protests and sit-ins demanding removal of the app. The mostly female workforce found that the app gave managers remote access at any time to information about their every move, and the ASHA workers’ union argued that the app undermined trust among the poor, rural communities they served (Bansal, 2021). In many domains, algorithmic management has been posited as the solution to the challenge of decentralized worker control. EVV technology companies have been explicit in their comparisons to the gig platform economy, calling for the “Uberization of home care” as the path forward to modernizing the industry (Alaya Care, n.d.). At the same time, a growing body of labor literature has documented how these business models have enabled employers to exert significant control over workers while divesting themselves of employer responsibilities through independent contractor misclassification, and have been linked to greater economic precarity, lack of benefits, and exclusion from the social safety net (Dubal, 2017; Rosenblat, 2018). One potentially emergent dynamic may be parallel ways that governments may have growing digital control at a distance over care workforces while continuing to push risks and responsibilities onto paid and unpaid care workers and families (Glaser, 2021). What this may portend is a future of increasingly extractive, technologically mediated labor management of the bottom of welfare state workforces’ hierarchies.

Surveillance between welfare and labor

A significant dimension of EVV systems’ impacts lies in the historically ambiguous position of care workers in the US welfare system and, within that, long-term care's reliance on Medicaid as the primary source of funding. Homecare work is shaped by the racism, sexism, and deep stigmatization of poverty and disability that have long shaped care infrastructures in the US. But this has underpinned not only societal devaluation, but the perception and treatment of care workers themselves as welfare dependents, performing labor that should rightly be the domain of private, unpaid family care. Historians Boris and Klein (2015) have traced the emergence of homecare programs out of earlier US legacies of slavery and domestic service performed primarily by Black women. Homecare aide jobs were viewed by early builders of the modern American welfare state as a means of pushing poor women out of welfare dependence and into waged labor, thus creating a “jobs program on the cheap” (Boris & Klein, 2015, p. 69) These legacies continue today as low wages keep a majority of homecare workers living below the federal poverty line, with more than half relying on some form of public assistance, such as Medicaid, food stamps, or cash assistance (PHI National, 2019).

At the same time, legislative contestations over wage and labor protections were often undermined by the argument that raising worker wages would make care provision inaccessible to too many Americans, thus pitting workers and clients against each other (Dowling, 2022; Iezzoni et al., 2019). On a functional level, then, Medicaid homecare workers make up part of the public sector workforce, yet are not treated as such. In federal legal rulings they have been termed “quasi-public employees,” thus trapped “at the nexus of the public welfare state and the ‘private’ labor market,” and excluded from being able to make demands of the state through collective bargaining (Bigley, 2022, p. 253).

Another important factor is long-term care's dependence on Medicaid, which historian Gabriel Winant has called “the poor stepchild of health insurance,” but whose budget has nevertheless been steadily growing since the 1980s (Winant, 2021, p. 18). Medicare, the national health insurance program for people over age 65, paradoxically only covers primary, acute, and post-acute care in the short-term, meaning that public long-term care in the US has, since its creation, been associated with poverty and the stigmatization of welfare. This has led to distrust of both workers and care recipients, where policy discourses around overseeing care provision have centered on the conundrum of homecare as “a virtual black box – an unknown to the consumer and to policy-makers” (Boris & Klein, 2015, p. 160). Assumptions regarding welfare recipients as fraudulent and untrustworthy also combine with popular perceptions of the “disability con,” or the idea that many disabled people are faking or exaggerating their disabilities (Dorfman, 2019).

Thus, algorithmic tools to detect, deter, or predict fraud are becoming more common in welfare and worker management contexts alike. One prominent example has been the Dutch government's use of SyRI, a risk calculation model to calculate whether an individual is at higher risk to commit benefits fraud, which was legally challenged for selectively targeting low-income neighborhoods, exacerbating discrimination on the basis of race, socioeconomic, or migrant status (Henley & Booth, 2020). In the care sector, the rise of formal auditing techniques and technologies have proliferated in online care marketplaces, including tools like AI assessment products to allow consumers to risk score care workers’ “trustworthiness” before hiring them (Harwell, 2018; Ticona, 2020). In both contexts, algorithmic tools may be critiqued not only over how they may be biased and lack transparency but also for the targeted ways they are deployed, and the legal power relations that encompass not only citizen–state relationships, but also the subjugated relations experienced by low-wage workers.

Flagged for noncompliance

As part of required stakeholder engagement around EVV implementation, the Virginia Department of Medical Assistance Services (Virginia DMAS) solicited public comment, and received over a thousand negative complaints from both care workers and Medicaid recipients. Among them was a mock job description satirizing the unworkable demands of personal care attendant jobs and the pressures of digital surveillance. In addition to describing such “benefits” as no health insurance, lost paychecks due to glitchy software, and working with busy, overloaded case workers, other bullet points included, “Your employer will be a stressed out family who [is] so overwhelmed and under threats by the government for anything from abuse to fraud.” The last bullet point warned, “You must have a talent for following guidelines that are contained in multiple regulations without any clear instruction of how to interpret regulations. All support agencies will have different interpretations, so you must be able to interpret the interpretations” (Virginia DMAS, 2020).

This section focuses on the experience of having one's digital timesheet “flagged,” both as a continuous threat and a reality that altered benefits recipients’ daily life and fueled worker precarity. The perplexing work of “interpreting the interpretations” aptly describes workers’ and clients’ efforts to make sense of what it meant to “comply” with EVV systems, and what it meant to get flagged for failing to do so. On its face, the CURES Act requirements appear straightforward. The legislation mandates that EVV systems collect six data points: (1) type of service performed, (2) who is receiving the service, the (2) date and (3) location of the service, (4) who is providing the service, and (6) the time the service begins and ends. But in practice, workers in multiple states began constantly experiencing their time submissions being flagged and/or rejected (Mullaney, 2018). Some of these instances were attributable to technical issues, including inaccurate or failed GPS location readings (especially in rural areas), glitchy timekeeping that led to minor discrepancies in clock time, failed biometric verifications, and other errors due the fact that EVV's low-income users often did not have access to new, well-functioning smartphones or could not afford consistently active phone data plans (Mateescu, 2021). However, another major issue was that it was often challenging to avoid getting flagged, and it was not clear to EVV users what getting flagged entailed. A major driver of the EVV mandate was a 2016 projection from the Congressional Budget Office that claimed that EVV implementation would lead to cost savings of $290 million over a 10-year period, by cutting down on “fraud, waste, and abuse” (CMS, 2020). Technology vendors had claimed that EVV systems make it much more difficult to falsely bill for services, and serve to safeguard clients against neglect or abuse by identifying gaps in care. If the data does not match – such as a worker clocking out at an unauthorized service location – they may be denied payment or face other penalties.

What constituted fraud, however, and how administrators would decide whether a flag indicated fraud or mere error, was generally opaque to both workers and benefits recipients. This question was particularly vexing for live-in care workers and family members who were paid through consumer-directed programs, where paid working time was often only a fraction of the amount of unpaid time spent providing care. As one interviewee who worked as a personal care attendant for her developmentally disabled son asked, “How does a mother commit fraud, when you're with your child 24/7?” While most interviewees affirmed the real possibility of care worker neglect or abuse, they expressed skepticism over whether EVV system data could reveal anything meaningful about quality of care from time and location data. In states like Arkansas, where financial management services company Palco administered the state's EVV app, AuthentiCare, the company had assured users that flags for triggering an “unauthorized location” alert were merely “informational geofencing messages” that didn’t necessarily mean workers were being accused of fraud. However, the days or weeks it took to review the data delayed or fully halted workers’ paychecks and could spell financial ruin (Eubanks & Mateescu, 2021). Error, fraud, or otherwise, the result was often functionally punitive no matter what the final assessment was.

The ways that EVV systems decontextualize time and location information away from their social contexts in particular provoked worry that the data would tell the wrong story. In response, workers devised various strategies, including keeping meticulous diary records and taking screenshots as counter-data, placing Post-It notes in various locations so as to remember to log numerous tasks in real-time, and walking their clients into backyards to sign off on timesheets in a location with better device connectivity. Workers in rural states and individuals without home internet spoke of traveling to locations with public WiFi, such as public libraries and parking lots of fast food chains, in order to regularly double check and submit time entries. Angela, a disabled Black woman, described how her worn-down smartphone's weakened battery life meant that she and her attendant had to rush any activities conducted outside of her home out of fear that the battery may die during a shift, meaning that her attendant would not be able to clock out in time. While some policies allowed for manual corrections to timesheet data, workers nevertheless expressed fears of coming under scrutiny for too many corrections. One benefits recipient described her struggles to get more clarity from her case manager: “they have told us, we’re only allowed so many corrections before we get penalized. No one's told us exactly how many corrections or what the penalty is. So, you know, and we’ve asked [the case manager] and they just ignore it. They just, like, brush right over.”

While the concept of “administrative burden” has typically been used to describe the web of complex bureaucratic hurdles imposed on citizens to access public benefits, the case of EVV shows how such burdens reshape the labor process in the form of data entry (Herd & Moynihan, 2019). Tara, a white woman who oversees two care attendants to provide services for her son through Texas’ Medicaid Community First Choice (CFC) program described how a simple trip to a doctor's office would require as many as seven different checkpoints through EVV:

If attendants take him there, they will have to clock out of one service, clock into “transportation.” Once they get there, they clock out of “transportation,” and back into CFC hours. Once they get done, it's clocking out of CFC, then back into “transportation.” Once they get back home, it's clocking out of “transportation.” It's too much for them to do.

Part of this meticulous breakdown of care into individual timed tasks is a legacy of the standardization of paid care work as a means to minimize what the state pays out, that have long represented the tensions between the “universalism of bureaucracies and the particularism of caregiving” (Abel & Nelson, 1990, p. 12). Because care work is impossible to fully routinize, management practices since the 20th century have turned to work discipline through scheduling, squeezing out the more intangible, relational aspects of care from compensation (Winant, 2021, p. 239). I argue that this continuous, invisible labor of attending daily to EVV systems’ opaque, unreliable, and idiosyncratic demands can be thought of as part of the material effects of welfare digitalization. In Ducey's study of nurses experiencing a hospital's economic restructuring, she describes how budget cuts were felt not only in their caring relationships to patients, but as care for objects – through the work of “continually adjusting, and adjusting themselves, to the objects and equipment around them,” creatively making do when the objects necessary to providing care broke, became scarce, or didn’t work as they should (Ducey, 2010, p. 21). Rather than enabling more efficient care provision, the introduction of EVV systems creates a new layer of responsibilities around data entry and the care and maintenance of work equipment.

Moreover, the slippages between timekeeping or technical errors and the categories of “fraud, waste, and abuse,” were further complicated by technology vendors’ ambitions to deploy data mining and predictive analytics. An industry White Paper from healthcare services company Optum, for instance, promotes the potential uses of EVV to identify patterns in historical data that are “dependable predictors of improper billing” in Medicaid services. Some tech vendors’ marketing rhetoric has also promised new ways that EVV can reduce “soft fraud” – a term used by the insurance industry to describe the practice of exaggerating or omitting information from an otherwise legitimate claim, in contrast to more overt “hard” fraud. What “soft fraud” entails in the context of care provision is not clear. The extent and purpose of GPS location verifications and geofencing was a subject of fear and speculation among workers and benefits recipients for precisely the reason that further inferences might be made from their physical movements, ranging from probing phone calls from case managers to broader circulations of their data within Medicaid bureaucracies. In a study on how probation officers perceive and produce narratives from GPS ankle monitor data points of their parolees, Shklovski et al. (2009) make the observation that “the social nature of the GPS trace's production – the activities, relationships and contributors that give it meaning – are absent from the trace as it circulates within the technological system.” While there was no direct link between EVV data and benefits determinations or care hours allocation decisions, a common concern was that these processes would be or were already linked. Indeed, some healthcare start-ups have already indicated the future utility of broader EVV data integrations towards this end, going well beyond the CURES Act legislation's original intent.

The “flagging” of workers’ timesheets for minor missteps has led to both added administrative labor and impacts to workers’ pay, and suggests broader changes in how accountability is configured. Zhang (2023) argues that state reliance on digital surveillance methods to monitor care remotely brings informational constraints that shape the definition of accountability rather than vice versa. As a result, in cases of noncompliance “the lack of documentation is not merely a smoking gun for fraud but constitutes the violation itself” (Zhang, 2023, p. 47). Location data tells supervisors very little about care relationships, but the existing infrastructure of consumer smartphones’ location tracking capabilities makes it an easy proxy upon which to pin policy goals. This slippage means that auditing processes are detached from questions of care quality, but do create a mechanism by which digital worker surveillance can shave off program costs through more granular divisions between “verified” and “unverified” care. As will be described in the next section, workers and service recipients were both incentivized to comply with and manipulate EVV systems’ visibility into their care relationships.

Complicating employment relationships

In addition to shifting relationships to the state, the rollout of EVV systems surfaced embedded tensions, ambiguities, and contradictions in everyday working relationships. In the US, homecare workers labor under a variety of employment structures. Some are directly employed by private home health agencies, which may provide services for a mix of private-pay and state-funded services (Medicare and Medicaid) to which the agency bills for reimbursement. In states like California, which has one of the largest homecare workforces in the country, the state's In-Home Supportive Services (IHSS) program pays workers directly, and a strong union presence has granted these workers greater protections as public sector employees. Since the 1990s, consumer-directed programs emerged as an alternative to agency-based models, promoted by disability rights and independent living movements as a way for service recipients to regain autonomy over how they receive services. These programs enable service recipients to directly hire, train, and manage workers, who may also be family members or other kin, while a third-party intermediary processes billing and Medicaid reimbursements. Poster (2011) has used the term “multi-surveillance” to complicate notions of surveillance as monolithic, pointing to configurations where workers experience many layers of surveillance from different actors with differing or conflicting interests. The triadic employment relationship of homecare workers, between employer–employee–client (and the state as a powerful but in many ways invisible fourth actor) meant that EVV systems brought out different, conflicting incentives to both comply with and evade surveillance.

Initially, EVV systems were designed for an agency model. At a 2019 information session for agencies, EVV sales representatives painted a picture of a command center-like station where an administrator could have a bird’s-eye view of the workforce, represented as multitudes of tiny dots dispersed across a digital map. Intimate activities like bathing, dressing, meal prep, and managing medications could trickle from dozens or hundreds of clients into the system in real-time, transforming everyday life into data and billing codes. Just as the use of beacon technologies and other IoT (Internet of Things) devices have transformed management in the hospitality industry, so too, a sales representative touted, the homecare industry could modernize the way it manages its workforce.

This model conflicts in many ways with how consumer-directed programs work, where service recipients are typically the “employer of record.” In interviews, the shift from a paper-based system to digitally recording care services in real-time brought out feelings of guilt and responsibility from service recipients, who feared the new rules would strain working relationships. Ruth, a white woman who helps her disabled daughter by supervising her personal attendants and managing paperwork, felt frustrated at having to enforce strict time discipline over the workers: “The last thing we would like to do is to add this element, you know, to their day schedule and make them feel that this is some kind of a situation where they’re monitored when they’re five minutes late.” One of the touted aims of EVV system roll-outs was to add another layer of oversight and accountability over this deinstitutionalized employment relationship, ostensibly to protect workers as well (Jain, 2019). But, as one worker pointed out, EVV systems do not track if workers are underpaid for their labor, only if they are overpaid for providing unauthorized services. In some ways, service recipients benefit from receiving additional unpaid support, given that care hours allocated by the state are often not sufficient to cover the support needed to live safely and comfortably. But at the same time, they were acutely aware of how difficult it was to retain employees, given the low wages and lack of benefits, and went to great lengths to maximize workers’ pay within the limits of their care plan. One service recipient spoke of spending hours emailing with her financial management company, trying to figure out why her workers’ checks were coming up short. In some instances, service recipients have resorted to dipping into personal savings to pay their workers when state-issued paychecks were halted over “flagged” EVV data (Eubanks & Mateescu, 2021).

In some instances, workers and care recipients were alike incentivized to make care work legible to EVV systems. But in others, both parties felt compelled to protect the other from surveillance, or to pressure them into surveillance. Some workers spoke of concern over the fact that their role of inputting all “service locations” into EVV systems effectively created a geographic itinerary of their clients’ entire lives. Similarly, clients sometimes agreed to download the EVV app on their personal cell phones because workers feared being tracked on their off-duty hours. In contrast, the requirement by some EVV systems to use facial recognition or voice verification also pitted workers and clients against each other, where workers felt they had to pressure their clients to repeatedly pose for the phone camera in order to go through verification to access the timesheet submission portal.

These tensions highlight how workers and care recipients experience “privacy extremes” in relation to the state (Gilman & Green, 2018). On the one hand, they are hypervisible as digital surveillance very directly enters the intimacy of home life. Yet EVV systems do not measure the factors that erode care quality, such as poor support, precarity, and burnout. Such issues are deprioritized in favor of forms of quantification that render invisible the often exhausting labor of producing “proof of care.” Thus, EVV systems perpetuate the illusion that data collection is a passive process. In doing so, state bureaucracies are able to exert greater labor control without added responsibility. In contrast, care recipients are positioned as employers, but have little in the way of resources or authority to create the working conditions that make quality care possible. This shift is part of a broader trend. In the UK, Hayes (2015) has observed how the deregulation of care sector employment has led to “conditions in which surveillance has emerged as a new regulatory dynamic” (Hayes, 2015, p. 171). In other sectors, gig platforms exert control through opaque algorithmic decision-making, and by selectively quantifying only what matters for maximizing profits and probing for signals of risk in user behavior. In the context of Amazon Mechanical Turk, Irani (2022) argues that these “algorithms of suspicion” are particularly insidious because they don’t operate by any rules that define what makes a “good” versus “bad” worker, but merely operate by pattern-matching: “[t]o be an outlier, rather than breaking a rule, can be enough for the company to flag a worker as a problem.” Surveillance tools like EVV govern by defining what “compliant” care looks like through the constricted lens of time and location data, thus shaping the nature of welfare provision through its instruments of measurement.

Conclusion

Contemporary welfare systems are being transformed through datafication, not only through reconfigurations of how decisions are made about citizens but also how the workforces that make up the welfare state are managed, evaluated, and (often) exploited. In this article, I have argued that digital surveillance of homecare workers enables Medicaid bureaucracies to engage in continuous and small-scale decision-making about what counts as “compliant” care through flagging workers’ time submissions. What the case of EVV highlights is how data-driven technologies shape public services beyond the more easily identifiable and discrete moments of high-stakes decision-making, such as benefits determinations, which may make scrutinizing them and constructing public accountability more challenging. EVV systems, on the face of it, are merely timekeeping tools, a common “low-tech” solution as old as waged labor and far less sophisticated than complex algorithms. But as Benjamin (2019) shows, even a “low-tech” device such as a zip code can be used as a seemingly race-neutral technology to perpetuate oppressive, racist structures.

The use of digital surveillance to monitor care workers in particular magnifies a contradiction at the heart of many state care infrastructures: that while profound inequalities and exploitative labor conditions are lamented as a threat to the stability of our care system, they are nevertheless treated in policy-making and technology design alike as essential to its functioning. When care workers are “flagged” for non-compliant data inputs, those small discrepancies are often merely a reflection of the complex relationships and needs that these workers must balance to do their jobs. In the absence of substantive public investment and supportive human infrastructures, the turn to fine-tuned metrics and digitally enforced work discipline offers an appealing solution to the distribution of resources. However, more so than other forms of algorithmic decision-making in other domains – such as healthcare or the criminal legal system – the case of EVV puts ordinary people in the position of becoming full-time data workers. As this article has shown, both care workers and their clients struggled to produce data that documented their care relationships. These challenges were not only due to poor technology design or algorithmic bias, but point to deeper questions about the aims and objects of accountability in public programs. As consumer technologies like smartphones can be deployed as data collection tools for government bureaucracies across many contexts, more research is needed to understand how these small, yet constant interactions are changing people's relationships to the state and to each other.