Speculating with Care: Worker-centered Perspectives on Scale in a Chat-based Health Information Service

Seeking to address barriers to in-person care, governments and non-governmental organizations (NGOs) globally have been pushing for scaling chat- or phone-based information services that rely on care workers to engage with users. Despite theoretical tensions between care and scale and the essential role of care workers, workers' perspective on scale and its impact on care provision is rarely centered early on in decisions to scale. In this paper, we examine care and scale from the perspective of medical support executives (MSEs) who support a chat-based health information service for maternal and child health deployed across multiple states in India. We draw on observations of MSEs' work, interviews with MSEs, NGO staff who implement the service, and families who use the service, and speculative design sessions conducted with MSEs. We find that by centering MSEs' perspectives, we can differentiate between growth of the relationships and heterogeneity that enable social impact, versus scale-thinking that promotes the decontextualization of care. We leverage our findings to discuss implications for scale and automation in chat-based health information services, including the importance of human connection, place, and support for care workers.


INTRODUCTION
In a number of public health systems, information services are increasingly being deployed at national scales [1,4,20].In Global South countries in particular, increasing mobile phone use amongst diverse populations presents new opportunities to reach people with information and medical advice, including precisely those who may face the most significant barriers to in-person care in inequitable and underfunded health systems.Typically operated by governments and non-governmental organizations (NGOs), these services may automate dissemination of health information to registered mobile phones and also answer questions sent in by care-seekers, ultimately aiming to support health management outside of clinical settings.These services often employ trained nurses or allied care workers to answer questions.As these services expand geographically organizational goals for scale.We then held speculative design sessions with MSEs to uncover their perspectives on their work and potential changes to it.
In light of potential changes such as increasing engagement with the WhatsApp service and automation of MSEs' workflows, we find that there were shared values within the NGO such as the desire to increase impact of the WhatsApp service.However, MSEs were also concerned about what their work responsibilities will look like, how to preserve their relationship and engagement with families, and work-life balance.We find that these concerns arise due to entanglements of the scalable and nonscalable that MSEs are uniquely positioned to make visible.Drawing on studies of scale and care, we contribute an understanding of how centering care workers' perspectives can help differentiate between impact and a sole focus on scale by bringing to light how organizational and design decisions affect the relationships and heterogeneity that enable impact.We also address tensions in maintaining a focus on care and human infrastructure amidst limited resources, by demonstrating how even in cases where technology may be more cost effective to invest in, a sense of place offered by human connection is necessary for impact.Through this discussion, we share implications for the role of scale and automation in chat-based health information services that prioritize human connection, place, and support for care workers.

RELATED WORK
We situate our work in multiple bodies of work in CSCW and HCI.We discuss studies of technologymediated care provision, such as in chat-based support groups or helplines.We also describe the work on conversational agents in health domains specifically and recent advances prompting questions of automation.We then summarize work on scale and automation, and its tensions with care labor.In relation to this work, we center worker perspectives in a semi-automated WhatsApp service to understand how their expertise and priorities can inform trajectories of scale and automation.

Technology-mediated Care Provision
A body of work in CSCW and HCI has looked at how care workers, such as doctors, nurses, facilitators, and volunteers, offer informational, emotional, and navigational support via chat or phone-based helplines [24,44,45,62,68,72].These settings may be characterized by long-term interactions, for example in a chat-based support group [24,72], or one-off interactions such as a person calling with a question about a specific health concern [62,63].Depending on the context, prior work has highlighted the different factors that affect how the relationship between care provider and recipient unfolds and how workers might be best supported in care provision.Some studies have looked at facilitation of chat groups, for example for mothers, couples seeking fertility treatment, or youth living with HIV [24,68,72].Yadav et al. 's study of WhatsApp groups for pregnant women and mothers showed that women valued responsiveness, openness to calls for more detailed conversations, and the overall ease of access to medical expertise through the groups [72].Studies looking at more one-off interactions have demonstrated the work of meeting emotional needs amidst the pressures of time and lack of existing rapport.Gallegos et al.'s study of a breastfeeding helpline in Austrialia shows that nurses built mothers' self-efficacy through affirmation and teamwork, whereas ignoring interpersonal and emotional needs in favor of pragmatic advice could undermine self-efficacy [15].Pendse and colleagues describe how the design of mental health helplines could better recognize issues such as wait times, lack of identity-specific services, and unknowns about what to expect on the call [45].They also describe the importance of the identity of helpline volunteers and how they shape decisions around counseling [44].Overall, these studies affirm and characterize the central role that care workers play in ensuring quality remote care (or not), which our study takes as its starting point for centering workers in decisions around scale.
A number of studies note the issues of high workload that come with engaging care recipients through technology, sometimes on top of other in-person care responsibilities.Issues include keeping track of interactions that need responses [24,68], the level of emotional labor required and few procedures in place for considering safety and wellbeing [24,44], and the challenges of maintaining fast response time [24,72].The motivation to keep up with increasing scale of interactions in particular has prompted a number of design recommendations around automation and AI intended to allow care workers to focus more on relational aspects of their work or more vulnerable care recipients.For example, some interventions may share information on a timeline, such as during pregnancy, which could be scheduled and automated [46,73], while AI could be designed to help ensure more in-depth follow up with those more likely to drop out of information dissemination interventions [41].Sentiment analysis could aid facilitators in determining when a particular care recipient or moment in the group conversation needs greater attention [24,68].There have also been moves to automate the triaging and answering of health-related questions, including in multilingual settings [8].This work argues for the feasibility and appropriateness of supporting scale through these means, but workers' perspectives on how their work will actually be impacted by efforts to scale technology-mediated care provision have been understudied.

Chat-based Conversational Agents for Health
In comparison to systems that rely on care workers, there has also been a growing body of work on entirely automated chatbots for health and wellbeing, motivated by addressing barriers such as stigma or lack of access to in-person care.This brings up questions of the precise role of conversational agents in comparison to care workers.Seeking to understand the potential of conversational agents themselves, condition areas for application of chatbots have included mental health [28,31,32], COVID-19 [27], adolescent sexual and reproductive health [49], sexual violence [34], breastfeeding education [73], and maternal and child health more broadly [43].Studies have centered on acceptability and need-finding, looking at what communities desire from chatbots.Goals include sharing information tailored to the local context [27,73], addressing health-related questions [49], testing knowledge, navigational support [34], health coaching [37], and encouraging self-disclosure [31,32].Other studies on chatbots have generally been Wizard of Oz or experimental studies that aim to understand experiences with chatbots; some of these studies begin to look at the role of chatbots in relation to care workers.For example, studies of AdolescentBot and FeedPal, chatbots in South Asia for adolescent sexual and reproductive health and breastfeeding respectively, point to concerns such as ensuring that users know the scope of the chatbot upfront and that it cannot offer consultations with health workers [49,73].A body of work looks at the factors that affect self-disclosure to a chatbot, including perceptions of disclosure from the chatbot itself or use of chatbots as a tool to share information with a mental health professional [31,32].
In terms of feasibility, iterations of large language models, particularly those aligned with the medical domain [57,58,74] or drawing on multiple languages [54], have prompted much speculation as to whether chatbots based on these technologies can support a more flexible range of interactions in medical question answering and be used in more diverse contexts.However, these models, in addition to more general ones, have been shown to exhibit inaccuracies or share irrelevant information [57,58] and have not been rigorously evaluated for situated conversations or key priorities such as health equity or empathy [3,58].Their appropriateness for these cases is also not clear given that these models do not process meaning [5,18].Multilingual language models may also use biased wording, misinterpret words or context, or impose linguistic properties onto low-resource languages, due to continued dominance of English data for training [40].Experts in healthcare and digital health call for consensus on the acceptability of use of these tools and divesting from the idea that they can replace workers [18,61,64].Prior work in HCI has demonstrated that humans play a key role in communication even amidst automated interactions.For example, Perrier et al. 's work on a hybrid SMS system to engage pregnant mothers [46] or both mothers and fathers [47], showed that the automated messages helped establish a relationship while nurses could answer personal or unexpected questions, though there was the possibility that mothers thought the automated messages were also from nurses [46].Mitchell et al. studied the feasibility of text-based health coaching, finding that human coaches could be more empathetic and ask probing questions, while chatbot interactions could be more persistent and consistent in offering options [37].It is evident that care must be taken in understanding potential roles of humans and automation, not just in terms of feasibility, but from workers' perspectives as care providers.

Scale, Automation, and Labor
Scale and care are often seen as antithetical to one another, and especially so when scale extracts, invisibilizes, and standardizes care labor.Tsing's influential work on scale discusses how scalability intends for projects to expand without changing, which excludes diversity and possibilities of relationships that lead to transformation [66].Hanna and Park describe how scale relies on and assumes interchangable units of work, a largely homogeneous set of users, and the datafication of people [17].Systems can also be interscaled, where it is not just one component that is scaled, but multiple distinct elements, which intersect to contribute to a notion of scale [48].Work also shows that scale is not just about material configurations but also about strategic narratives that make systems seem scaled or scalable, when in reality, they rely on hidden labor, such as that of microworkers for AI tools [55].
Prior work in CSCW and HCI suggests that a core aspect of resisting scale thinking is to prioritize strengthening relationships, sharing of ideas and resources that achieve intended impact, and attending to difference and conflict among stakeholders in the system [6,17].Lampinen et al. offer proliferation as an alternative to scale, seeking to recenter impact and diversity over monocultures and linearity [29].Looking at practice, Tseng et al. discuss how, in the context of supporting survivors of intimate partner violence with computer security, care and concepts of standardization and measurement might be synthesized by supporting therapeutic interactions and survivor-centered approaches to evaluation [65].However, Seaver suggests that we also be cautious of the strategic move in some arenas to position care and scale as compatible [55].For example, prior work has pointed out how gig platforms "care" for workers and customers in ways that have little to do with nurturing relationships and more to do with controlling workers in favor of maintaining profits [51,56].In technological projects in many healthcare settings, there is a tendency to view scalability as important for achieving social impact, but also to make up for the high cost of (and subsequent lack of investment into) human resources [11].Mateescu and Eubanks warn that this "risks creating a system where we only value the parts of care that can be turned into data," [35] and conversations in CSCW have prompted questions of how to work towards a politics of care in which design centers care workers over efficiency [23].
A body of work seeks to make the human labor in relation to technologies visible, in efforts to problematize claims of seamlessness and scale achieved through these technologies.Studies uncover how domain experts and data annotators who make data-intensive systems possible are devalued or overlooked in the process of AI development [53,69].For example, in frontline health, the design of AI systems and scalability do not align with workers' workflows and aspirations [19] or their mental models [42].Other work looks at care contexts such as sanitation and elder care, highlighting the invisibility of workers' and care recipients' efforts to integrate technology.Kang and Fox center the perspectives of janitorial staffers in the introduction of AI in their workplace, arguing for the importance of worker perspectives in conceptualizing ethical AI [22].Lazar et al. discuss how a conversational digital pet for home care for older adults elicited workarounds to the way it obscures the human teleoperators behind the avatar [30].Visibility has also been complicated in prior work, as the move to visibilize care labor is not necessarily enough to ensure that voices are heard [50], especially compared to concerns of scale [36].Overall, our work contributes an understanding of how centering care workers' perspectives can align with the provocation to resist scale-thinking in trajectories of organizational change.In particular, we draw attention to contexts of resource constraints, where technological scale is often seen as the only feasible option.

METHODS
Our study was conducted between March and December of 2021, as part of a larger, year-long qualitative investigation of the WhatsApp service and MSEs' work [26].The goal of this study was to understand organizational goals for scaling a WhatsApp service for maternal and child health, and center workers' values and perspectives on the potential effects of scale and related automated workflows on their work.The study was approved by institutional review boards in the United States and India.

Setting
The NGO partners with state governments across eight states in India to support patient education around health conditions (e.g.MCH, general medical and surgery, tuberculosis, cardiology) in two ways: 1) conducting in-person patient education sessions by condition area at district hospitals (tertiary-level care within a three-tier public health system), and 2) operating a remote engagement follow-up service, majority of which is a WhatsApp service that continues to share information and offers families the opportunity to ask questions.MCH is the largest condition area of work, comprising of more than 80% of the patients whom the NGO trains.The in-person sessions are a key point where families learn about and can register for the WhatsApp service.The WhatsApp service was started in 2019 and approximately 52,500 families had registered for the WhatsApp service by 2021.Families can register to receive either antenatal or postnatal care information, and they get messages approximately every other day for two months.Families can ask questions at any time during this period and after the message series ends.The MCH service received approximately 30 to 40 questions a day about MCH and caregiving, at the time of the study.MSEs are trained nurses, employed by the NGO to answer families' questions, among other work.Questions get answered by MSEs in the same language that the question was asked.In the process of answering questions, MSEs collaborate with translators and doctors.Doctors approve or help write responses to families.Over the course of the study, the NGO employed seven MSEs and two doctors.MSEs answer questions between 8am and 8pm, with one shift at 8am -5pm and another at 2pm -8pm, with three hours of overlap between MSEs.

Data Collection
All data collection was conducted remotely via Zoom or phone call by Naveena due to the COVID-19 pandemic and changing levels of community transmission.For all interviews and design sessions, Naveena asked for verbal consent for participation, audio recording, and taking notes.Study phases and participants involved are summarized in Table 1.Zoom, covering both shifts and transitions between shifts.MSEs shared their screens as most of their work is conducted via a computer, and Naveena asked questions as needed about work done off screen such as phone calls or writing notes.Observations were conducted with an orientation to how MSEs conduct their work, any pain points that arose, and the practices and tools that enabled them to accomplish work despite challenges.We also conducted semi-structured interviews with both MSEs, focused on their motivations for joining as MSEs, positives and challenges of their work, experiences collaborating with colleagues, and any future goals they might have for their careers.Interviews were approximately an hour long.Overall, this helped us understand the nature of MSEs' work, challenges, perceptions of their work, and relationships with colleagues.

Understanding the
As part of this phase, we later conducted semi-structured interviews with parents who were registered for the WhatsApp service to understand their experiences with it, including their motivations, use, and perceptions of the service, in relation to its intended design.We interviewed four men and five women.As our goal was not necessarily to gain a comprehensive understanding of the user base, but rather to learn about the experiences and circumstances that brought about families' engagement with the service, these interviews provided valuable context.Interviews were approximately an hour long.All participants resided in rural or periurban parts of Karnataka, except for one man who resided in Maharashtra.All participants had newborns, except one woman who was expecting.Participants were 21 to 34 years old and part of low-income households.
3.2.2Speculating on Scale.In September of 2021, more MSEs had been hired to support the increase in engagement with the WhatsApp service, and the NGO had plans to set up more WhatsApp service lines for condition areas beyond MCH and expand geographically.The NGO and research team were interested in understanding how MSEs' perspectives and extensive experience engaging with families could be centered in efforts to handle the increased volume of engagement and availability of the WhatsApp service.Data collection for this part of the study focused on understanding nearand long-term organizational goals for the WhatsApp service and perspectives on how to handle increased engagement (whether through technological means or otherwise).With an understanding of the possible changes that could arise, we then sought to understand the background of the new pool of MSEs and solicit their perspectives on organizational goals and what they seek in their futures of work.
To understand organizational perspectives, we conducted four semi-structured interviews with NGO staff who were leads in the design, technology, and product teams.These teams were chosen because of their direct and most significant involvement in decisions related to design and scale of the WhatsApp service.Questions focused on near-and long-term goals for the organization, what aspects of their current services and workflows they expect to change to meet these goals, and resources and constraints in reaching these goals as an organization.Interviews were conducted over Zoom and ran between 45 minutes to an hour.They were conducted in English, which was the language typically used among staff (as well as MSEs).All leads had been working at the NGO for one to four years, and included two men and two women between the ages of 26 to 36, with backgrounds in social work, health systems, engineering, and design.
We then conducted additional semi-structured interviews with MSEs, drawing on the same questions as the first phase of the study.During the second phase of the study, there were six MSEs total: four newly hired MSEs, one MSE who had recently returned from leave, and one MSE whom we interviewed in the first phase (the other was on leave).This pool of MSEs was composed of five women and one man, all working remotely in Karnataka and between ages 25 and 35.We conducted four more interviews based on MSEs' availability, leading to a total of six interviews across the two phases.All interviews were an hour long.All MSEs we interviewed were women.
We then conducted four speculative design sessions with all current MSEs.Our approach was informed by Wong et al. 's work on speculative design oriented towards thinking through lifeworlds and widespread implications of imagined sociotechnical systems [71].They respond to approaches that typically only account for the self and immediate contexts of use, suggesting that also explicitly considering the broader social systems that people and artifacts exist in is important for understanding the wider implications of artifacts on a wider range of stakeholders [71].We drew on this approach because organizational decisions would affect not just an individual MSE, but also MSEs as a team, their relationship with families, and the impact they would be able to have through their care work-all of which are important for understanding the value they get out of doing care work.We wanted to ensure that the speculative design activities could prompt MSEs to speculate about their own experiences, as well as in relation to colleagues, families, and the NGO as a whole.Inspired by Wong et al.'s suggested tactics, we designed the four sessions to bring attention to people, practices, and institutions, the different ways that people experience infrastructures, and relationships beyond use.
The sessions were conducted weekly, over Zoom, as part of an existing series of workshops on design skills that was being held by a design team member for MSEs (which was an effort to support these skills amongst all NGO staff regardless of which team they were in).Though we communicated that the particular sessions led by Naveena were part of a research project, we repurposed this series so that the speculative design sessions would not require additional time to be scheduled with MSEs.
The first session started with collectively defining the concept of design and sharing MSEs' associations and experiences with it so far.We also introduced the concept of speculative design specifically, unpacking how it encourages thinking through possible future lifeworlds to inform the present, and emphasizing that there are no incorrect answers or approaches to the activities or questions posed throughout the sessions.We also conducted a tree of life activity [52] (refer to Figure 1, left) using an interactive whiteboard, which supported group discussion of MSEs' values, the support and resources they draw on, and their aspirations for work and life.This session was intended to support MSEs in relating to design, building common ground and a sense of group capacities, and becoming comfortable speaking freely and communicating ideas even when not fully formed.
Between the first and second sessions, we prompted MSEs to do a diary activity in which they were asked to note down (whether in written format, with pictures, or just mentally) enjoyable and challenging experiences at work.We prompted MSEs to share these experiences at the start of the second session with the goal of priming them to think about their personal and embodied experiences of work, including how they felt and why, so that scenario walkthroughs could be discussed at a similar level of detail.
The second and third sessions consisted of the scenario walkthroughs, where we posed potential changes in scale and workflows inspired by interviews with NGO staff, and asked MSEs to walk us through what those changes might look like in practice and how they might be affected by them, whether it was positively or negatively.We asked about both open-ended scenarios of greater engagement without posing technological solutions, as well as specific technological solutions, in • "What if there were 100 MSEs answering questions?" • "What if you were answering 200 families per shift?" • "What if families heard about the MCH line through a relative, poster, or WhatsApp forward?" • "What if coordination with medical team/translators and reporting was automated?" • "What if automation could help you answer questions?Where might it help, if at all?" Follow-up prompts included suggestions to walk through a typical day to ground discussion in everyday practices, questions around clarifying or describing an idea in more detail, or reflecting back ideas to synthesize the discussion at various points.As MSEs discussed their perspectives, Naveena shared the interactive whiteboard screen for all participants to view and wrote sticky notes summarizing MSEs' comments (see Figure 1, right).We also prompted MSEs to voice what concerns, questions, or alternatives they had or could see in relation to the scenarios.
The fourth session consisted of prompting MSEs to imagine what future news headlines about the NGO or reviews from families about the WhatsApp service might look like if the scenarios in the previous two sessions played out.These activities were intended to prompt consideration of the broader societal and end-user impacts of organizational changes, inspired by similar activities used in prior work by Wong [70] and aligning with speculative design focused on lifeworlds beyond individuals.We prompted MSEs to consider both positive and negative headlines and reviews, to elicit thoughts on diverse experiences with infrastructures.

Data Analysis
We first analyzed data from observations of MSEs' work and interviews with MSEs, in the form of notes taken during observations and interview transcripts.We used inductive thematic analysis [60] to identify themes.We first generated initial codes, such as "valuing novelty of work", "desire to connect with families", and "stress of managing expectations", leading to themes such as "valuing heterogeneity" and "challenges in handling existing volume of interactions".We also analyzed data from interviews with NGO staff, asking what and how decisions were made regarding the future of the service; codes included "expanding reach of service", "hiring more MSEs", and "preserving the human touch", which led to themes around "achieving meaningful impact", "resources mediate decisions to scale", and "ripple effects of supporting greater engagement".These larger themes make up our findings around MSEs' collective experiences at work and forthcoming organizational changes.
We used our understanding of MSEs' perspectives on their work and the NGO's considerations around scale to inform the speculative design sessions.We used an inductive and iterative process for analysis of data from the speculative design sessions, including transcripts from the sessions, notes taken during the sessions, and artifacts produced in the tree of life and headlines and reviews activities.Naveena and Richard reflected on each session after its completion, discussing interesting reactions or comments from MSEs and any challenges with ideating or encouraging discussion.Naveena created memos based on these reflections and the transcripts and artifacts from each session, drawing insights such as how MSEs perceived the impact of automation on their rapport with families, or how many of the aspects of work they valued had to do with heterogeneity and learning.Naveena then revisited the transcripts of interviews with NGO staff, drawing comparisons between perspectives to ask what it would mean to center MSEs' perspectives in responding to increased engagement with the WhatsApp service.This process led us to think about what participants really sought to "scale", bringing us to engage with studies of scale that helped us differentiate dimensions of scale and their relation to social impact.
We note that by its nature, speculative design requires thinking through lifeworlds that are not necessarily concrete and that participants do not have direct experience with yet.Seminal work in CSCW has demonstrated that situated practice and the context of use are important to understand, and not always aligned with initial plans and design [59].Our findings speak to values and priorities based on workers' experiences so far as MSEs, and we aim for these to be a resource in bringing the design of chat-based services closer to MSEs' experiences.

Positionality
The authors include Naveena and Richard, researchers based in the United States, and Shirley, Nupoor, and Victoria, research and design staff at the NGO in India.Given conversations in the CSCW and HCI communities on the roles and responsibilities of researchers in relation to worker groups and advocacy [13], we note that the research collaboration, despite seeking to center worker perspectives in organizational change, is still external to the worker community (as opposed to explicitly led by them).We proceeded with the speculative design sessions, however, due to the fact that the NGO staff felt it was important to understand MSEs' perspectives through such sessions.Also, MSEs themselves voiced in interviews that they appreciated the flat organization of the NGO and ability to have input into design decisions.These values were reflected in the fact that the NGO sought to support MSEs' skills as designers through the existing design workshop series we integrated into.However, it is still important to acknowledge that hierarchies can be implicit or shift over time.MSEs provided feedback that it was very valuable to engage in the questions involved in the session and also to share those views with other MSEs, which is not something they did regularly.Our hope is that the speculative design sessions were helpful in and of themselves towards introducing forms of interaction and perspective-sharing that MSEs may continue to find useful in future.

FINDINGS
Our findings start by describing the work of MSEs, highlighting aspects of their work that contribute to satisfaction.We then describe the dimensions of scale that the NGO was considering at the time of the study.Finally, we highlight MSEs' perspectives on these changes, and how these dimensions would interscale to change their work.

Care: Receiving and Giving
Critiques of scale have brought attention to how scale depends on being able to expand without rethinking basic elements, obscuring the heterogeneity of these elements and the potential transformations they enable [17,66].We start by describing the experience of being an MSE and caring for families, highlighting where differences and change are currently possible and contribute to the satisfaction that MSEs get out of their work.
4.1.1Growing and Skilling as an MSE.Most MSEs were trained nurses or nursing educators who, after periods of time doing clinical nursing work, sought to find more flexible and less physically taxing work.Multiple MSEs noted how working on the WhatsApp service was an opportunity to still do care work, while also working remotely and having more manageable shifts.As one MSE described, "Working hospital side is a different thing...We have to do the night shift... and handle the life of the families.So I thought let it be nursing, but a kind of office setup... Timing will be fixed, and then after that, we don't need to worry about work and all."In tree of life and diary activities, many MSEs emphasized how their nursing background was a shared strength, noting "medical field", "nursing", or "public health" as part of their trees' trunks.This allowed them to engage with families, but many MSEs also noted that the nature of the work was an opportunity to branch out into other skills and domains.
MSEs' work consisted of documentation, coordination, and engaging with families, all of which used various digital tools and language skills.MSEs received questions from families via a software that allowed them to navigate all WhatsApp chats in one dashboard.In terms of documentation, MSEs manually recorded questions received from families as well as metadata such as date and time received and category of question, in Excel sheets.MSEs also periodically collated this and other data to share with other teams within the NGO, reporting information such as the number of questions received per day, frequency of language, and what could be answered by the FAQ bank or not.Upon documenting families' questions, MSEs sent them to translators via chat if needed, and then to the medical team, to either get approval for a response constructed with the help of the FAQ bank, or get input on creating a new response.
Much of this work could be described as manual and repetitive, and while prior work describes this as a source of disappointment for workers [69], MSEs said it helped them (at least initially) learn new things beyond their clinically-focused nursing training.For example, one MSE noted during a speculative design session that they were using this opportunity to gain technical, design, and language skills: "...in our nursing field nobody teaching us the technical part, so here we're learning about the technical part as well... we don't know some languages but we are learning them now, so it's about the learning and everything is coming."Their work on the chat platform also occasionally involved more varied work, for example changing or debugging the automation logic for the chat software, which offered an interesting challenge, as one MSE described: "I love to do the tricky things, like in the platforms and all, some like flow mistakes will be there, then I love to find those mistakes, then correcting those mistakes by myself." Multiple MSEs noted that picking up technical skills was the most exciting part of their job, and some felt that these skills would be useful in the future given the increasing digitization of healthcare they had observed in India.
In addition to their core work via the chat platform, MSEs noted that one aspect of their current work environments that was satisfying was how it was possible to work with other teams and projects.This was enabled by a relatively flat organizational structure and the NGO's recognition of MSEs' healthcare background and experience engaging with families through the WhatsApp service.For example, MSEs were involved in weekly design team meetings.One MSE described during the diary activity how she was asked to present on healthcare delivery during one such meeting, which allowed her to revisit and hone her presentation and communication skills: "It has been long time that I have not done.So by doing that I felt, yeah, it's good.It's something I have learned." MSEs could also target their involvement according to their interests.For example, one MSE explained how she discussed doing more in-depth technical work with a design team member: "I'm also interested in computers, so I would like to spend more time there, like generating the reports [about WhatsApp service usage] and presenting in a different way...And I'm learning Google Data Studio also for all these things.So I am growing." In another example, another MSE described helping start up a WhatsApp service for a different condition area.This heterogeneity was core to MSEs' satisfaction with their work, as one MSE described in an interview: "It's [work is] like continuously changing, even now today, it's a new project for us.And today we finish the project and tomorrow it will be again a new project.So till now, like I'm really excited and also this is what keeping me more, you know, like moving on in this organization." Diversity of work was also an opportunity for self-discovery, as one MSE described in the diary activity: "So as I'm working on the other new projects... it helps me to understand my strengths more." In doing this range of work and learning the skills needed, MSEs valued strong communication amongst junior and senior MSEs, made easier by their shared professional background.At the time of the study, four MSEs had newly joined and relied on more experienced MSEs to show them the ropes.MSEs felt they could easily reach out to senior MSEs, a relationship that enabled problem-solving, as one MSE described: "...the commonality we have is nursing... and along with that, we have support like teaching, training.Because we are new, we didn't know anything, we don't even know the language actually, so the coordination [with other MSEs], the communication and understanding... like if I have any problem, I talk to my [senior MSE], she will make out a way." 4.1.2Caring for Families.MSEs also do a significant amount of care work, and their interactions with families brought them satisfaction, as well as concerns around work-life balance.Typically, MSEs respond to families' questions via chat, sometimes asking follow up questions, such as gestational age or age of the newborn.This aspect of their work allowed MSEs to use their nursing background and support families but "from like a far distance", as one MSE described.However, MSEs also conduct "user calls", or random calls to families to personalize the WhatsApp service for them and to better understand their experience with the service so far.More rarely, they also sometimes contact families experiencing emergencies or challenges navigating in-person care, offering emotional and sensemaking support.For some MSEs, calls were where they were able to feel most connected to families and the impact the service was having.For example, families sometimes shared their worries, positive experiences or increased confidence with caregiving, or sensitive concerns that they felt uncomfortable talking about in in-person clinical settings.One MSE described how the calls get closer to the satisfaction she received as a staff nurse: "Texting was okay, but phone calls were really good.With texting, we'll have the questions and appropriate response and families will send feedback, good service.With phone calls, some families, there are minute things.Surprisingly, once I talked to a father, usually mothers will talk more, but he was taking so much care.'I take care of my wife, I will share [the messages] with my wife... ' It felt like they can share anything." At the same time, MSEs expressed how both messaging and calling families could be stressful at times.There could often be insistence from families on faster responses or a barrage of requests around things the WhatsApp service was not designed or allowed to do, such as writing prescriptions or diagnosing conditions.One MSE described handling the pressure to respond to questions on the WhatsApp service as one of the most challenging aspects of her job as there was nothing she could do to address the bottlenecks of coordinating with the medical team or translators: "That's a difficult part for me and they [families] are also not wrong.I mean they have some expectations also, that's why they are asking.But yeah, we can't do anything for that." Another MSE explained how once they called a family, they might call back several times, including in the early morning or late night.She explained her concern that even if she turned off the phone, "...they will be like irritated, like these people, we are calling them and they are not picking up, all these things they might feel." Despite the range of experiences that MSEs had with individual families, all MSEs shared a deep connection to collectively working towards greater impact in the organization's mission of supporting caregivers.In the tree of life activity, multiple MSEs noted aspirations such as "help families in every possible way", "need to see the [NGO] at the high level", and "to grow with the organization", demonstrating their desire for the NGO's work to support both families and MSEs.There was strong agreement in the design session that MSEs could have more impact by being part of a collective effort, in response to an MSE's reflection on the connection between her personal and work goals: "It is only if [I am] in this organization, I can help more people.If I'm single, and I'm only one who is having a dream to help families, I can't do that." There was also an association of impact with providing the NGO's services in more places.During the headlines activity, multiple MSEs envisioned the NGO offering maternal and child health services globally, with one headline reading "[The NGO] has signed an MOU to provide MCH services all across the world." There was clearly a desire among MSEs to see the organization serve more people-questions and contestations as to how this might happen make up the focus of the following sections.

Scale: The Why and How
Philips notes in her work on scale that the "...construction of scale is multidimensional... dimensions of scale can be rendered interdependent-or interscaled-in ways that contribute to their cultural impact, durability, and power, " [48].She describes how a "bigger wedding" means more people, but it also means "a grander space, more food, fancier food, more and higher-alcoholic-content liquor for free, a fancier wedding dress... " and so on [48].Further, these are often assumed to go together, so that more attendees "naturally" implies other aspects of scale.In this section, we describe the different pathways through which the NGO was considering increasing engagement with the WhatsApp service.We find that scaling engagement in these ways has ripple effects, but does not automatically lead to greater scale in other aspects of the intervention-these decisions are instead mediated by cost of resources, limitations to technological innovation, and a concern for maintaining the "human touch".

Having an Impact.
At the time of the study, one relatively short-term goal was to use existing infrastructures to increase engagement with the WhatsApp service.The NGO aimed to expand the number of hospitals at which the NGO conducted patient education sessions and therefore introduced the WhatsApp service to families.The NGO also wanted to increase the "conversion rate" of families who took the training and then proceeded to sign up for the WhatsApp service.Another goal was to increase active engagement more generally from those who did register-not to simply inflate use but to ensure the service is as useful as possible.The design lead explained why increased engagement is important for expanding the impact they knew was possible based on the current state of the WhatsApp service: "We want to support families more.Based on early findings from evaluations and our experience, engagement isn't low because people don't have questions, it's low because people don't know they can ask questions or what kinds are okay to ask.We don't want to be a spam service, we want reading, responding, questions, so we can make an impact." In seeking to expand impact beyond maternal and child health in the long term, the NGO sought to create lines for more condition areas.This effort was already underway with services being built out for cardiac health.Finally, they also wanted to offer the service in more countries.
More engagement across state and national boundaries meant considering another dimension of scale-language.The NGO viewed tailoring language and script to each family's preferences to be very important for building rapport and ensuring comprehension.In this sense, scaling language was not about attempting to create a lingua franca, but rather to support a greater diversity of languages in recognition of the relationships they helped build.
Another change that the NGO was considering that would increase engagement was publicizing the WhatsApp service beyond hospitals, making it more likely that families could register without having been to a hospital or attended a training.This was already happening to some extent, as families would sometimes share during calls with MSEs that they heard about the service from the posters in the hospital, or from relatives or a neighbor.According to the tech lead, this would avoid engagement being "fundamentally limited by the footfall of the hospital, the willingness of the nurse to actually conduct the session, and a lot of other external factors... " He thought this could also lead to a greater diversity of people registering for the service, as communities who are comfortable with going to hospitals may also be more privileged in terms of class, caste, religion, or other identities.This could also lead to a greater diversity of mindsets with which families register for the service, as shared by the tech lead: "...the attitude that someone who's coming in cold will be you know, they are... going to be curious about this, or maybe like desperate, where they really need help.And some of the mental, like, state of mind of mothers who are signing up for this service will be different, a lot more diversity... " This potential trajectory was informed by the desire to offer support to more families, but according to the product lead, it was also related to the fact that it was resource-intensive for the NGO to work with hospitals and monitor the patient education sessions.It was also too difficult for nurses to consistently set aside time in their already overloaded schedules to conduct the sessions.The hope was that the NGO could engage with hospital workers in other ways, and an in-person touch point for families could be more about inspiring them to think about and build confidence in their role as caregivers.Meanwhile, the tech lead postulated that even if more MSEs were hired to support a scaled WhatsApp service, it would be easier to grow a centralized team of MSEs than a decentralized implementation team and hospital partners.Thus, we see how the NGO was reconsidering how impact is best achieved when working with particular assets and limitations-expanding the WhatsApp service allowed the NGO to reach more people, even outside clinical settings, with fewer and more manageable resources.

Achieving Impact.
To respond to greater engagement with the WhatsApp service, NGO staff indicated the need to scale its supporting workforce, namely the number of MSEs they hired.They felt this would be feasible for a number of reasons, one being that they were confident of the hiring pool of retired nurses or nurses looking for work outside of clinical settings.They also found that working remotely is an attractive option for this workforce.A design team member noted that many MSEs work remotely, including some from before the COVID-19 pandemic.In her experience, "...when we start telling this is going to be work from home, there are a lot more people who are open to do this." This was especially important given the nursing context MSEs were coming from-most MSEs noted they appreciated the flexibility of location and time because it let them attend to caregiving roles for children or other family members who were dependent on them.However, members of the design team were also concerned about whether having more MSEs, for example up to tens or hundreds, could reduce their sense of community or connection to the mission of the NGO, affecting collaboration and motivation.
Another response to increased engagement was a narrowing in on the organization's mission.One design team member shared that MSEs might also have to do fewer calls with families in emergencies, and ignore more questions that do not fall under the behavior change mission: The level of customization means, we call some of the families if they're in an emergency situation and tell [advise] them, so I think most of them will come down.Not just that but we answer to each and every hi, hello, some of the hospital related complaints, bank-related queries...We keep answering all of those questions, telling them to go back to the hospital and ask the people there, so I think we might stop answering all of those things...Because that's not the mission of our services, the mission is behavior change.The NGO also wanted to make it easier for MSEs to respond to families, so that even with greater engagement, the response time would not suffer, and possibly even improve.At the time of the study, responses could take a few hours as MSEs went through the pipeline of answering questions.The tech lead noted that this process could not withstand a higher volume of engagement, however: "...we get X number of questions in a day, and we are able to manage it right now, even though it's manual, because we have, like a few MSEs... if the number of questions even doubles tomorrow, our team is going to be completely overwhelmed, because they're pretty much at capacity right now." To support greater capacity, the NGO was considering automating parts of MSEs' workflows, namely the flow of information between teams and the documentation work.There were longer term aspirations towards implementing a triage system as well that would help alert the medical team as to which questions could not be answered by MSEs with the help of the FAQ bank and would need their additional expertise.However, the tech lead felt that the feasibility of this was limited in the short-term given that language technology was not advanced enough to support processing of local languages and dialects and diverse grammars and spellings.He described how this was not ideal for the NGO's purposes given the stakes of the health domain: "...the margin of error that we can deal with is very low.So, you know, we really can't afford to have like, false negatives." At best, he foresaw the very far-off possibility of piloting a triage system for English (which the WhatsApp service received very few questions in) and Hindi questions first.
Beyond backend workflows, a major question was whether direct interactions with familiesasking follow up questions or directly answering their questions-could or should be automated as part of speeding up response time.Again, the limitations of language technology were salient.Additionally, NGO staff were concerned about losing "the human touch" that they felt made the service unique.One design team member noted that chatbots are common in India now, and families would recognize when a chatbot is fully automated.For example, minimal or repetitive conversations were a known sign of automated systems, according to a design team member: "WhatsApp is popular in India now, and most of them know if it's a company phone number, then it's a bot.But ours is not like that... we're sending them answers in their particular language, in whatever style they use...And there's no convo that'll happen with a bot... here what happens is, they [MSEs] build a conversation which builds trust, which has human touch... [With commercial chatbots] they're the same set of questions [from the chatbot] that I have to answer whenever I have some problems, but here it's not the case, it's a human asking different questions according to your problem." Design team members were also concerned about maintaining customization of language as part of the human touch.The NGO was considering hiring more translators or MSEs with diverse language skills to support faster responses, but they were still not sure if the level of detail required for transliteration and translation could be maintained with significantly increased engagement.
Though it was not an immediate consideration, the tech lead did feel that at some level of scale, humans simply cannot answer each and every question.This is where the possibility of automating some responses to families came in, and he felt that then the goal would be to make the WhatsApp service as human-like as possible, indicating a desire to address some of the issues with standardized chatbot interactions: "So there's a sense that you are talking to a human being which might get lost and typically does get lost after you reach a certain scale, because literally human beings cannot respond to so many things.So I wish we can like, design the service to be as human-like as possible... because it helps allay their fears as well... versus like a bot with a standard template message that is just like, you know, telling you follow ABCD, that's a little robotic." We also found that it would then be an organizational decision to consider what sorts of risks automated responses would entail.It would mean accepting some margin of error, rather than striving towards error-free interactions.Suggestions around handling a margin of error included having audits, sending questions with certain keywords that indicate more serious cases to humans, or having a human monitor the automated decisions so it can become more accurate over time.

Futures: Scalable and Nonscalable Worlds
Delving into the always present connection between scalable and nonscalable components, Tsing argues that scalable worlds depend upon nonscalable worlds, even if this is not acknowledged [66].In this section, we show how MSEs viewed dimensions of scale and nonscale, such as faster response time and the maintenance of the human touch, coming together to constitute their work, and the specific futures they saw for themselves amidst such changes.

"
Sometimes I feel we need more members".One immediate reaction to the idea of greater engagement was satisfaction-MSEs shared a sense of responsibility to ensure the WhatsApp service can serve the families who need it.At the prospect of getting 200 questions per shift, approximately quadruple the current volume, one MSE shared: "We feel now responsible for that.If we are answering 200, we are helping and reaching out to the families, so we feel very happy with that."In terms of actually managing the increased workload, MSEs imagined that hiring more would help with time management, assuming the hiring kept up with the increased engagement.Newer MSEs shared how already, they struggled to finish their shift on time, and that handling so many questions made it difficult to consistently choose the most appropriate responses from the FAQ bank.One MSE shared how she felt that she would appreciate another MSE helping answer questions during her shift to alleviate some of the pressure: "Sometimes I feel that... in [shift] hours, we need more members.One person is answering now, so I wish [hiring] other persons means it will be in correct time and we can manage also." One specific issue with managing time for some MSEs was that they did not receive many questions from families in their own language, which meant that they needed to rely on translators more heavily.This in turn added to response time significantly and required MSEs to keep track of and return to families who were waiting on a response once they received a translation.Hiring more MSEs and allowing for more specialization according to language would avoid needing to rely on translators and reduce wait time, while also allowing MSEs to draw on their strengths, as one MSE shared: "...If we have at least one language, one MSE, at least we can manage well, because [right now] I have to wait for the translation.Sometimes it may be [translated] earlier or sometimes it may be later.So waiting for that and searching [for the family's question] again, it is causing delay also sometimes.So I feel... Telugu questions I will answer, Hindi questions other person can answer." MSEs also suggested that if there were many more of them, it would be extremely important to maintain strong relationships and communication amongst one another.They felt that currently, it was the ability to ask for help and have close communication during shift changes that allowed them to learn from one another and ensure a quality service for families.As one MSE said, "...when it comes to the relationship and communication, it has to happen.Because when we're working, it is [through] the communication only which we are able to work."With tens or hundreds of MSEs, participants were concerned that there would be challenges in communicating without some sort of structure that let MSEs know who exactly to ask for help.MSEs speculated that it might not be possible to have as much one-on-one support between senior and junior MSEs, but some suggested group trainings and testing of knowledge before starting to answer families as alternative methods.They also foresaw a role for more peer learning, where teams of MSEs, even if new to the work, could troubleshoot issues together.
MSEs also emphasized how a larger workforce could lead to a more deliberate process for allowing MSEs to specialize in areas they are most interested in upskilling in.While currently, MSEs could participate in design decisions or technical areas if desired, a larger workforce of MSEs could allow for specialized teams for particular types of work beyond managing the chats.As one MSE said, in her experience working with new MSEs, "Some of them are more interested in technical part, some of them are interested more in the medical part...So we can like differentiate and if we give them a training, it would be good for them and us also." 4.3.2"Automate... except answering the families".MSEs were largely in support of automating their backend workflows-automated documentation work, information flows between other teams, and, ideally, translation, would help with the time management issues described above.However, MSEs speculated that automating the actual conversation with families and responses to them would be infeasible and actively harmful because it would not be possible to build rapport or trust in the service without human conversation and follow-up questions.For example, one MSE said, "They [families] will think only the computer is sending us some messages and that's it, there is no conversation...We are asking them what is the baby's age, when does the baby get delivered, all these things we're asking... " and another MSE followed up, saying, "We need interpersonal rapport, then only we can give a better response and a better service." Observations of MSEs and interviews with families affirm the assessment of circumstances and risk that MSEs accounted for in answering families' questions.MSEs needed to decide on the importance of unpacking certain details as mentioned above, which would determine whether particular symptoms should be seen as worrying or not.MSEs also accounted for the family's emotional state, emphasizing when to not worry or when to see a doctor and how quickly.They also kept in mind local practices when answering-for example, while providing instructions on breastfeeding, they might also mention what not to feed babies to proactively address practices that can lead to complications, such as feeding honey.
MSEs also pointed out how some families already perceived the service as entirely automated without any humans available to respond, prompting them to question the impact of further automation, even if more human-like, on perceptions of and engagement with the service.MSEs shared multiple experiences hearing from families that when they perceived a lack of human presence behind the service, they did not engage or ask questions-this happened even with the current design, which was intended to convey supportiveness in the tone of the messages and explicitly informs families that nurses are available to answer their questions.In one example, an MSE described how a family she had called felt that they would not get human engagement based on the automated messages until she informed them otherwise: "I have called one family...I asked them why you are not asking any questions to us, they said I thought it's a computer, like computer is sending a response to us like daily messages, so I thought there is no manpower behind this, so I was thinking if I ask question also, nobody will be there to answer back...So I told her no you can ask the question, and there will be a nurse, doctors are there, who is giving a response to your questions... whenever you want you can ask the question, we are there for you, and then she said okay." MSEs were concerned about families' lack of confidence in an automated service to the extent that for the review activity, some MSEs imagined that families might have negative feedback if they perceived any sort of automated or repetitive response from the system: "It is only robotic service, there are no humans observing" or "We can't trust these responses.This is not related to what I asked.I'm going to another platform where humans help." In interviews, most parents did indicate that they thought nurses or doctors, or at least the NGO as an entity, answered their questions.They explained that this was because they were told so at the hospital where they registered, or because the response messages to questions indicate that a team is working on answering their query.Crucially, the perception of health workers responding to questions was what convinced families that this was what they called "proper info" that they knew they could rely on.At the same time, one mother said that she did not know where the messages were coming from.Overall, we found that it was not a given that families all perceived the service in the same way, affirming MSEs' experiences with needing to clarify this with some families.
Even with automation of only backend workflows, MSEs were concerned about their future role.If much of their work of coordinating across teams or retrieving responses was automated, MSEs wondered what they would do with their time instead.Asking "Why are we here then?", one MSE laughingly said, "Only we need to sit and see if the computer has given correct answer or not," indicating dissatisfaction with such use of their expertise.In response, another MSE suggested that they can work on other skills: "We will engage ourselves in projects.Creating the templates, flows, whatever the things are there, we will go there and we will become a technical persons." Another factor that MSEs saw affecting their relationship with families was if families started hearing about the service through means other than the in-person education sessions.MSEs were in agreement that families who heard about the service through health workers, relatives, or neighbors generally had a sense of what the service could offer and had confidence in it.However, they had numerous experiences with families who heard about the service through, for example, posters at hospitals, assuming it could prescribe medication or help with navigating the hospital system, only to be frustrated that this was not within the scope of the WhatsApp service.As one MSE explained, a personal introduction was preferable to leaving the service up to interpretation: "...relatives, friends, and the doctors, and the nurses, if they will refer our number, then it will be more effective... From the one-side communication from the posters or other pamphlets, it is effective but not that much compared to the healthcare members." 4.3.3"We also have families".MSEs also foresaw their work-life balance being affected with greater interaction with families.For example, MSEs were already inundated with questions and call-backs whenever they called families.They foresaw such issues increasing with more families to call.The prospect of needing to answer more questions also affected how MSEs viewed working from home.As one MSE pointed out, though working from home was desirable, it also meant balancing family expectations with work, which would become more difficult with intensified remote work: "They are expecting us to do more work at home, whether we will finish before our duty timing also... we're working alone and [family members feel] my daughter-in-law or my daughter, we also need to work with them for a few minutes...So if we are sitting in front of laptop whole day, then it will be like not looking good." Then there was also the mental burden of care work and the pressure of managing questions in general.New MSEs felt they could be better mentally prepared for the pressure of answering questions and the many different topics that could come up that need to be addressed in unique ways according to the needs of the family, as one MSE shared: "It definitely differs upon individual right... while training or giving any suggestions, it also should [be communicated] like there will be certain times where you have to face these situations, so you have to be ready, you have to be more confident.This is the environment and you have to get adapted."

DISCUSSION
Our findings describe the relationships and opportunities for growth currently present in MSEs' work, and their views and concerns on how to maintain these aspects of their work amidst growth of the WhatsApp service.Below, we describe how centering care worker perspectives offers a pathway towards thinking about what prior work calls the growth of experiences and learning [29], in contrast to growth of technology itself.We also discuss how the context of resource constraints mediates decisions around technological scale versus investment in human resources, and how this might be interrupted through a focus on place in relation to technology.Throughout, we share design implications for chat-based health information services that account for the relationships and specificity brought out by care workers' perspectives.

Learning from Care Worker Perspectives
Literature on scale in CSCW and beyond offers provocations against scale-thinking, suggesting a divestment from the idea that systems can change without rethinking their basic elements [17].Rather, a focus on relationships and heterogeneity is what can enable transformation [66], and ideas of what to grow or expand need to be based in meaningful impact, rather than what is easily quantifiable or reproducible [29,65].Prior work also proposes the idea of proliferation, as opposed to scale, in which artifact ecologies can support the promulgation of ideas or offer templates for bringing initiatives into different value-based communities [29].Our findings show how a shared value among workers and the NGO as a whole was wanting to increase impact-bringing what had proven to be valuable information and opportunity to ask questions to more families.However, their desire to maintain the human touch of the WhatsApp service indicates how the goal is not just to simply clone the WhatsApp service or increase engagement for engagement's sake.Rather, MSEs described some of the most satisfying work they did as learning and engaging with families to support their caregiving work.NGO staff also were motivated by the understanding they had gained over time of how the service was useful in supporting caregiving work and could improve in meeting real needs for more information.
While there was a shared focus on impact, prior work has discussed how in moments of organizational change, there is a tension between organizational values and the challenges of maintaining them while growing [6,7].We acknowledge that in many settings care teams can often be overlooked when their needs are in conflict with scale or cost [19,36,50].However, by focusing on an organization aiming for social impact and linking the perspectives of care workers with the impact of interventions, we seek to interrupt the idea that scale and cost effectiveness through technology are the most important or sole paths towards impact, even when they seem necessary to prioritize.Prior work has discussed how prioritizing care labor can be a source of more ethical design [19,22].Our findings show that to this end, care workers on the frontlines not only can provide insight into their priorities, but also insight into the specific relations that allow for impact in health information services, which could help prioritize flourishing of learning and relationships, rather than expansion of technology itself.
We find that care workers provide unique insight into how different dimensions of scale come to affect care provision and to what extent.In our findings, their perspectives highlighted where scale might no longer be in service of impact, teasing apart the differences between sheer scale and meaningful growth.In the case of automating interactions with families, MSEs were able to point out where the efficiency of automation did not contribute to positive engagement with the service.Their experience showed that families preferred to engage with humans, while perceptions of automation hurt engagement, even with a supportive tone or explanations that nurses were answering.Prior work has also demonstrated people's desire to work around automation to engage with humans behind the scenes in care provision [30].Our findings point out where efforts towards growing or handling engagement might not be served by using automation to expand reach or simply differentiating between human and automated messages.Rather, growth of engagement might be about how families can receive a meaningful introduction to the service, and how a larger team of MSEs might be supported in continuing to call families to personalize the service.This leads to the dimension of scale involving expanding where and how the WhatsApp service is publicized.MSEs had an understanding of how this expansion would, and already does, affect quality of interactions with families.Their support for introductions by health workers, neighbors, or relatives could inform the design of standardized tools like posters, but more importantly, could also steer wider advertising of the service more towards leveraging local care networks such as frontline health workers or primary health centers.
Organizational changes might also require consideration of how MSEs could remain involved in work that honors the care provider-recipient relationship and MSEs' own interests and desire for growth.MSEs valued the mix of technical and care work and the learning opportunities it provided.We found that automation could support backend information retrieval and management to make workflows less stressful and support response time.This could mean having ways to track the status of constructed responses as they move between MSE, medical, and translation teams, or support finding the most tailored responses in the FAQ bank even if one is not familiar with its contents.As for interactions with families, MSEs pointed out the ways that their conversations and line of questioning helped build rapport.We found that their communications required deep understanding of context, culture, risk, and emotions.MSEs were also clear that serving only as a check on automated systems would be unsatisfying.Thus, despite the hope that increasingly human-like interactions can satisfy both scale and the human touch, fundamental issues with recent advances-such as large language models' inaccuracies and lack of understanding [5,57,58], and underrepresentation of marginalized groups in health-related data [2,21]-indicate otherwise and would diminish the value MSEs could get from their work.On the other hand, other possible benefits of freeing up MSEs' time spent on backend workflows could be the ability to retain the diversity of their interactions with families and change organizational trajectories.NGO staff felt that as they scale, they would need to narrow in on the behavior change mission by, for example, no longer responding to emergency calls from families.This is despite these calls being an indicator of the diverse forms of support families may need in caregiving work and struggle to find elsewhere.However, automation of backend workflows could allow MSEs to continue offering emotional and navigational support in such cases, or reach out to more families for user calls, which they expressed helped them feel especially close to and supportive of families.That being said, some backend workflows, despite being repetitive tasks, were an opportunity for MSEs to become familiar with various digital tools, which was useful given their observation of increasing digitization of healthcare.This points to the importance of contextualizing automation in workers' identities and backgrounds, which shape what skills and learning opportunities workers are seeking.Even if work changes, centering workers' definition of growth and learning could help determine how that value could be brought into future work responsibilities-in MSEs' case, this could mean contributing their expertise to more technology-or design-oriented projects within the NGO, as they suggested.
Our findings also point to how in the process of growing, even when investing in more human resources rather than technology, the potential change among worker communities needs to be thought through carefully.Though MSEs were already remote, their relatively small numbers enabled close ties, which encouraged MSEs to ask for help and learn from experience.Growing in numbers combined with remote work could make it more challenging to maintain this collaboration, especially during key workflows like shift changes, where thorough and timely coordination is required.These potential challenges to how while remote work might seem scalable in terms of space and associated costs, growing in terms of strengthening relationships requires intentionality.MSEs pointed out how they might need changes in team structure to still enable strong communication with more experienced MSEs, opportunities for peer learning, or alternative assessment strategies to evaluate learning.In these ways, care workers can speak to the rhythms and modes of collaboration that enable quality work and impact.A focus on growth of relationships is also not automatically a good thing, without accounting for boundaries and individual workers' wellbeing.MSEs described how care work can be taxing, for example, when receiving calls at all times of the day, managing families' expectations of the service, or generally handling constantly incoming questions.Being able to support greater connection may require thought in terms of organizational resources for setting boundaries and supporting self-care.For example, this might look like establishing norms around turning off work devices, or better preparing workers during training for the intensity of managing workflows during an actual shift, as one MSE suggested.

Taking Care Amidst Resource Constraints
The context of our study demonstrated the very real effects of resource constraints on decisions around growth and organizational change.NGO staff discussed how it required significant resources to support the patient education sessions in hospitals where the WhatsApp service was introduced.In contrast, managing the more centralized technology and human infrastructure behind the WhatsApp service was easier to manage and more cost effective.Of course, the potential decision to make the WhatsApp service standalone (that is, no longer introduced through in-person sessions) was also motivated by a question of whether the sessions were effective, given the challenges of running them amidst nurses' busy schedules; NGOs still operate within the more fundamental issue of underinvestment in public health infrastructures.At the same time, this prioritization of technological aspects of the intervention based on cost effectiveness, which is often normalized in resource-constrained contexts, exemplifies how scale can come to mean expansion without rethinking system elements-prior work has discussed how a loss of a sense of place, or connection to activity, history, or geography, is associated with unfettered scale [33].Yet, we found that expanding the advertisement of the WhatsApp service would in fact have ripple effects, shaping how families approach the service with less or varying context and how MSEs would then need to adjust their approach to engaging with families.This presents the question of how a sense of place can and must be recentered in decisions to expand technological interventions, so that we might resist-at least in some ways-the logic that cost effectiveness means overlooking humans.For example, the NGO was considering changing the aim and duration of in-person sessions to focus more on inspiration around caregiving, while making the educational content available remotely.Simultaneously, they sought to find other ways to continue engaging with health workers that they would find beneficial.In the same vein, strategies for sharing the WhatsApp service through community health infrastructures, as mentioned above, could align with families' sense of place.
Overall, these potential trajectories indicate the possibility of persistence in understanding how relationships and a sense of place can be maintained or created anew amidst growth.
We note that resource constraints also took the form of technological limits, such as not having technical resources that could support local languages or high-stakes contexts like healthcare.To some extent, this is precisely why hiring more MSEs with diverse language skills made sense, and MSEs themselves noted that this could allow them to use their language abilities.However, it is notable that NGO staff saw the possibility of language technology improving for Hindi, a highresource language, first, which would then shape which languages that, for example, an automated triage system could support early on.This demonstrates how technical resources could shape the trajectory of organizational change.Prior work has noted that it is not necessarily demand that results in language support for a technology, but rather economic prowess [38].If an organization is reliant upon such technologies, it is possible that organizational decisions end up following the same economic incentives, since those incentives shape what is available.Relatedly, given the high-stakes context and potential margin of error with multilingual language models [40] (and large language models for the medical domain [57]), the use of human labor to reduce error, as suggested by one NGO staff member, could also be disproportionately intensified in this context, similar to how needing to create datasets from scratch for systems in the Global South places additional burden on frontline workers in multiple domains [53].It then becomes even more important to consider how to center the strengths and interests of care workers-not just to make up for a lack of technical resources but to more actively maintain the aspects of work that support learning and strong relationships.It also may be valuable to prioritize support for parts of interventions that may not have as many dedicated technical resources, for example, ensuring support for low-resource languages by using rule-based classifiers.

Table 1 .
WhatsApp Service.Data collection for this study was informed by a larger study of MSEs' work supporting the WhatsApp service and families' experiences with the WhatsApp service.In early 2021, there were two MSEs supporting the WhatsApp service, working remotely within the state of Karnataka.We conducted a total of eight hours of observations of MSEs' work via Proc.ACM Hum.-Comput.Interact., Vol. 7, No. CSCW2, Article 361.Publication date: October 2023.Participants across study phases