On the Frontlines of Scaling-Up: A Qualitative Analysis of Implementation Challenges in a CDD Project in Rural India

This paper analyzes four years of qualitative data observing a large participatory anti-poverty project in India as it scales up from its first phase (covering 400,000 households) to its second (covering 800,000 households). Focusing on the frontlines of change -- at the village level, the analysis finds that the key difference between implementation in the two phases of the project was that facilitators in the first phase deployed a discourse that was carefully "co-produced" with its beneficiaries. Through careful groundwork and creative improvisation, facilitators incorporated the interests of multiple stakeholders on the ground while bringing beneficiaries into the project. However, as the project scaled up, participants were mobilized quickly with a homogenous and fixed script that lacked the kind of improvisation that characterized the first phase, and which failed to include diverse stakeholder interests, objectives, and voices. These differences significantly reduced the intensity of participation and its concomitant social impacts. The study finds that the work of facilitators was embedded in a larger shift in organizational priorities within the project, which in turn was responding to a shift in the political climate.

This paper analyzes four years of qualitative data observing a large participatory anti-poverty project in India as it scales up from its first phase (covering 400,000 households) to its second (covering 800,000 households). Focusing on the frontlines of change-at the village level, the analysis finds that the key difference between implementation in the two phases of the project was that facilitators in the first phase deployed a discourse that was carefully "co-produced" with its beneficiaries. Through careful groundwork and creative improvisation, facilitators incorporated the interests of multiple stakeholders on the ground while bringing beneficiaries into the project. However, as the project scaled up, participants were mobilized quickly with a homogenous and fixed script that lacked the kind of improvisation that characterized the first phase, and which failed to include diverse stakeholder interests, objectives, and voices. These differences significantly reduced the intensity of participation and its concomitant social impacts. The study finds that the work of facilitators was embedded in a larger shift in organizational priorities within the project, which in turn was responding to a shift in the political climate.

INTRODUCTION
In the past two decades, aid organizations, multilateral institutions and national governments have increasingly sought to involve local communities in development. The assumption is that mobilizing poor communities and involving them in all aspects of project design and implementation leads to more effective and equitable service delivery (WDR 2004). Billions of dollars have been spent on such community-driven anti-poverty interventions (Mansuri and Rao 2012). However, the scale and complexity of these interventions make such programs vulnerable to a number of challenges and considerable variations in local capacity (Woolcock 2009). Even if projects are able to build small islands of success in their first phase, they are unable to sustain the intensity and momentum required to induce participation as they scale-up (Mansuri and Rao 2012).
But why does the intensity of community participation decrease as projects expand? One important reason is what we might call 'implementation failure.' Expanding the scope of the project from a few communities to a large number puts pressure on the project's implementation unit, which in turn weakens its intended impact. Several papers and project documents highlight this phenomenon (Mosse 2004, Barron et al. 2011, Swidler and Watkins 2015. Little scholarly attention, however, has been paid to identifying the processes and mechanisms that constitute implementation failure during scale-up, and few studies have compared 'successful' and less successful implementation within the same project. In this study, we draw on qualitative data that cover four years  in the life of one of the largest World Bank-assisted, anti-poverty projects in India. We observed the project at a unique juncture in its trajectory, when it was scaling-up from its first phase (covering 400,000 households) to its second phase (covering 800,000 households). Impact evaluations found stark differences in outcomes in the first phase (Datta 2015) and the second phase (Hoffmann et al. 2017). Our own qualitative work found that the first phase was akin to a mini-social movement that challenged traditional structures of power and patriarchy . In this paper, we examine the changes in the project as it expanded into the second phase in order to explain the difference in outcomes. We use a combination of in-depth interviews, focus groups, and participant observation to trace implementation processes and changes on the ground in both phases.
Focusing on the frontlines (i.e., at the village level), we find that the key difference between implementation in the two phases of the project was that mobilizers in the first phase deployed a discourse that was carefully "co-produced" (Ostrom 1996) with its beneficiaries. Through careful groundwork and creative improvisation, mobilizers incorporated the interests of multiple stakeholders on the ground while bringing beneficiaries into the project. However, as the project scaled up to other districts, participants were mobilized quickly with a homogenous and fixed script that lacked the kind of improvisation that characterized the first phase and which failed to include diverse stakeholder interests, objectives and voices. These differences significantly reduced the intensity of participation and its concomitant social impacts. Tracing these differences back from village level to block, block to district, and then district to capital city, we found that the work of facilitators was embedded in a larger shift in organizational priorities within the project, which in turn was responding to the shift in political climate at that juncture. This paper focuses on the key differences in mobilization strategies adopted at the frontlines by the project that led to the different outcomes between the first and second phases. To this end, it is divided into five sections. Following this introduction, section two summarizes key ideas from the literature in a range of disciplines on inducing participation and introduces the tenets of the main conceptual argument. Section three outlines the methodology and describes the data employed in the analysis. Section four delves into the main argument, and section five provides a systematic comparative analysis of six ways in which mobilizers differed in the first and second phases. The last section concludes with a few recommendations on scaling up in large-scale community driven development (or CDD) projects.

LITERATURE REVIEW
Explaining why projects are unable to sustain high levels of participation requires a detailed examination of the factors shaping dynamics at the field level; however, the literature on this is fairly thin. Studies that have examined the implementation and ground-level functioning of development projects point to two broad challenges: elite capture and misalignment of project goals and staff incentives. We consider these two potential challenges in depth.
First, scholars argue that during the initial mobilization stages of any communitydriven project, facilitators tend to reproduce existing structural inequalities. Owing to their position in the social hierarchy, community facilitators are said to play an influential role in several aspects of a project. For instance, their preferences can heavily influence the deliberation process around project selection, and these preferences, as much as the preferences of prime movers within the community, are reflected in project proposals (Platteau and Gaspart 2003). This 'elite capture' is considered one of the greatest risks of poor facilitation, and participatory development projects routinely suffer from it.
Arguably, there are multiple junctures in the everyday life of a participatory development project that are open and susceptible to elite capture. However, facilitators or local elites are rarely a homogeneous category whose interests are in direct conflict with 'marginal groups.' Typically, there are multiple alliances across both categories that help reproduce inequalities. And yet, little careful analysis goes into showing how this is done or can be undone. This paper details how in Phase I villages facilitators engaged head-on with the messy business of preventing dominant groups (and themselves) from co-opting the process, and how in Phase II villages the pressure to scale up resulted in less effective facilitation.
Secondly, scholars have also suggested that facilitators are often constrained by the fact that they are working within the structures and hierarchies of donor and government interventions. For instance, although their job requires intensive and constant engagement with experimentation, facilitators are given targets (mobilize X communities in Y days) which might force them to adopt shortcuts to persuade or force people to participate, using messages for recruitment that are quite different from stated project goals. Mosse (2005), for example, finds evidence of this in Western India. As the project he studied scaled up, the fieldworkers, or jankars, working closely with but junior to the community organization project staff, began to "regard themselves primarily as project employees (if not private contractors), with the power to assess work and sanction payment" (p. 114). In essence, their incentives rarely aligned in a manner that allowed them to innovate on the frontlines.
The consequences of misaligned incentives can be many: To begin with, they can seriously hamper the quality of the participatory process. In other words, pressures to scale up quickly can be at odds with the emphasis on intensive participation, and this may influence the frontline workers' ability to deliver. Most critically, when constrained by the structures within which they operate, facilitators can at best induce 'passive' participation, which rarely leads to radical, empowered change. In other words, facilitators working within the parameters of a government or donor-funded project can only bring change in so far as it aids paradigm maintenance, rather than truly 'rocking the boat. ' Mansuri and Rao (2012) refer to the former as "induced" participation. Inducing participation requires a fundamentally different approach, they argue; one that is long term, context sensitive and committed to developing a culture of learning-by-doing among its staff through honest monitoring and evaluation systems. 1 It works best within a "sandwich" of support-an effective state from above and civic action from below (Fox 1993). Mosse (2004) offers an example of what may constitute high-quality facilitation.
Writing about a DFID-funded participatory project in western India, he notes the positive role that "skillful brokers" (i.e., facilitators) can have on project success when they 'interpret' the project goals in locally relevant and meaningful ways so that stakeholders (both project participants and a broader network of supporters) realize their significance and understand how their own interests are best served by them. This paper is closest to Mosse's work in that it also focuses primarily on the work of brokers. Our main contribution is to unpack how the policies and principles of participation find meaning in the daily work of their agents of delivery on the ground (i.e., the facilitators and mobilizers). We take the analysis a step forward by 1) illustrating 1 The work that project facilitators do must be separated from what the participants can do. The facilitators may well be constrained to operate within existing paradigms, but they need not circumscribe the spaces where project participants can or cannot operate. On the contrary, as we illustrate in later sections, the Phase I facilitators often encourage women to leverage their skills outside JEEViKA in a way that best furthers their interests, whether that means paradigm maintenance or 'rocking the boat.' On the other hand, Phase II facilitators pay very little attention to what women do beyond the realm of the project. the differences in practices when projects scale up, 2) demonstrating how incentives 'misalign,' and 3) explaining the variation in implementation between two phases of a project. The Phase I villages were selected randomly from a larger quantitative sample in Muzaffarpur and Madhubani districts. These two villages were matched with a set of Phase II villages using propensity score matching methods (Imbens and Rubin 2015) on the basis of village-level data on literacy, caste composition, landlessness, levels of outmigration, and availability of infrastructure from the 2001 government census. 2 In order to improve the quality of the match, field investigators visited all the possible Phase II villages for a week, studying their geography and their economic, social and political structures to select the two best matches for each Phase I village (see Table 1). Within each village, first, preliminary studies were conducted using participatory rural appraisal (PRA) methods to gain an understanding of the local political economy.

METHODOLOGY
Following this, qualitative data were collected in 11 cycles over four years. Every over the four-year period.) The first set of participants was selected to be representative of different socioeconomic strata in the village, and subsequent participants were selected via a mixture of purposive and snowball sampling. We interviewed women who were members of JEEViKA, their husbands, and key informants in the village such as religious heads, village council members, moneylenders, subsidized food shop dealers, landlords, and public officials.
Qualitative data were collected via a variety of methods: a) unstructured interviews, b) focus group discussions (FGDs), c) passive observation of group meetings, and trainings, workshops, mobilization drives at several levels (village, block, district).
Interviews were typically conducted in participants' homes, and FGDs in public spaces such as the village temple, school or panchayat office. These interviews in the field helped us systematically compare the impact of JEEViKA in its two phases, as well as the differences in quality of facilitation in the frontlines.
In order to help understand this shift in organizational dynamics and management priorities, we supplemented these data with in-depth interviews with JEEViKA staff. We interviewed a total of 24 state, district and block project managers, training officers, area and cluster coordinators, consultants and young professionals all the way from the village level to the capital city. In all, we conducted interviews and 64 FGDs at multiple levels.
Counting each FGD as one transcript we have 776 transcripts of data (see Table 2). The interviews were combined with direct observation of project activities and focused on understanding how the project and its frontline workers were responsible for the changes experienced. Interviews took one to two hours. They were conducted in the local language (Hindi or Maithili), simultaneously recorded, and then transcribed verbatim into English. All 776 transcripts were then coded in NVivo, after which the data were analyzed inductively in three steps. JEEViKA, therefore, decided to mobilize the women at the community level slowly by first gaining their trust and gathering input from them. In order to build a strong base of SHGs, JEEViKA brought in trained SHG members from SERP-Andhra women who had benefitted from the project-to mobilize and motivate the women in Bihar.
Alongside that, they carefully recruited frontline workers (Area Coordinators [ACs] and Community Coordinators [CCs]), looking for several years of experience in mobilization and participatory rural appraisal methods. Most of those ultimately recruited were former community workers from the National Bank of Agriculture and Rural Development (NABARD) and the Women Development Corporation (WDC). These frontline workers were sent on 'immersion trips' to Andhra Pradesh to learn from the SERP project. As we will see in the next section, along with the women from SERP, the Bihari women were largely successful in gaining trust at the village level, identifying the most vulnerable women and households and gradually mobilizing them into the project.
Careful groundwork and implementation resulted in a project that was perceived to be transparent, legitimate and successful in its first phase. An evaluation conducted in 2011, which used matching methods with retrospective questions on changes observed by respondents since 2006, found that the project's impact on women's empowerment (measured as physical mobility, participation in decision-making, political participation and confidence in undertaking collective action) was large (Datta 2015). In fact, compared to JEEViKA's success in achieving its targeted economic impacts, its social impact was substantially deeper. By giving economically and socially disadvantaged women access to a well-defined network of people and new systems of 'knowledge,' JEEViKA broke down long-standing normative restrictions and significantly shifted gender boundaries in these villages .
However, when the project expanded to its second phase, it had positive but less impressive results (Datta et all 2015, Hoffmann et al 2017. While the goal of debt reduction was achieved to some extent in Phase II, the social empowerment effects (women's mobility, social capital, participation in the public sphere) were not significantly different from zero. 4 In this paper, our aim is to illuminate the why behind this phase shift, something that remains invisible in the quantitative study. Why did this distinction between Phase I and II in JEEViKA's social empowerment effects emerge? Through our in-depth interviews, we found that the pressure to scale up and scale up quickly led to some of these differences. Owing to its success in the first phase, the project was able to get a buy-in both at the village and at the state government level.
Having established that the 'model works' in its first phase, the emphasis in the second be brought under the fold of JEEViKA. In addition, as the project expanded and scaled up to cover three times as many districts as quickly as possible, the 'phase shift' (Mosse 2005) brought with it entirely new organizational dynamics that sped up the mobilization phase and prioritized quantifiable targets over slow experimentation and adaptation. As one of the State Project Managers told us, "manpower, money and monitoring-all three were our key ingredients in the first phase, and we ignored them all in the next phase." These three ingredients succinctly capture all the factors that management identified as key. In terms of manpower, the facilitators recruited in JEEViKA's first phase were an intrinsically motivated group that had a much more dynamic and flexible approach than the second phase facilitators. In contrast, in the second phase, facilitators were less carefully handpicked and lacked the experience and drive of the first phase facilitators. Four thousand new staff were rapidly recruited en masse and degrees were emphasized over work experience in the field. "…there's a certain kind of zeal missing in the next generation. Something was lost in mass recruitment. It became hard to explain to them that JEEViKA is a path for women, not a destination. " 7 Moreover, the proportion of field-level staff to the number of SHGs declined in Phase II. In other words, the size of the workforce could not keep up with the number of SHGs, or for that matter with the increasing number of interventions that were gradually rolled out on the SHGs. "… [in the first phase] our only agenda was to mobilize women properly. Now far too many things are part of the mandate." 8 The sudden increase in the work force also meant that the new facilitators could not be trained directly by the state, but instead had to rely on a second tier of trainers at the district level. Also, unlike the first phase, where their training involved immersion in similar projects in other parts of India, particularly in Andhra Pradesh, in the second phase there were no such immersions. On the contrary, they had targets and were expected to begin mobilizing women from their very first trips to the field, thereby skipping an initial learning phase. These differences in recruitment and training led to "…a frontline that was a dedicated group of social workers in the first phase, and a dedicated group of salesmen in the second phase." 9 7 State Project Manager, Patna. 8 Nalanda AC. 9 District Training Officer (DTO), Cycle 2, Ramganj.
Phase II project managers regretted having to move forward quickly without first setting in place robust systems of monitoring and feedback. "…something worked in the first phase, but we didn't quite know what. There was not enough time to capture institutional learning from the first phase to feed into the second phase… [F]or instance, we had no idea how much time it takes to build a community." 10 Changes were made to the project design without feedback or a full understanding of what aspects of project design had helped achieve success in the first place. As we will see, one of the main changes-introducing the livelihoods component during institution building, rather than focusing on only building community institutions first as they did before-was a decision made without sufficient feedback. This led to less participation.
Besides facilitation, one of the key changes that had an impact on the second phase was limited resources-money-both for the community and for the project. With a sudden scale-up, limited resources meant there was little room for the kind of slow learning that was possible initially. The second phase focused, therefore, on emulating the first phase and its processes rather than innovating on the frontlines. ACs and CCs who worked in both phases often complained that during the second phase they felt Drawing on the qualitative data that focus on daily practices and interactions in the field, we show in this paper that JEEViKA in Phase I villages mobilized women into the project and sustained their participation by deploying a discourse that was "coproduced" by the facilitators and the women themselves, a discourse that incorporated multiple interests. We draw on Eleanor Ostrom (1996) for the original formulation of "coproduced" as a process by which "…citizens can play an active role in producing public goods and services of consequence to them" (1073). However, some form of 'active role by citizens' or partnership is now so common in service delivery that it renders such a definition trivial. Joshi and Moore (2004) further redefine "coproduction" or "institutionalized coproduction," as the "provision of public services (broadly defined, to include regulation) through regular, long-term relationships between state agencies and organized groups of citizens, where both make substantial resource contributions" (1).
Coproducing discourse is in many ways similar to coproducing public services.
Coproduction of service delivery involves blurring the conventional distinctions between outsider authority and insider knowledge, between public and private. Similarly, coproducing discourse involves purposefully obfuscating boundaries between the formal project script coming from the project HQ and the everyday circumstances of the women in the villages.
In Phase I, facilitators brought a script including flowcharts, images, training modules etc., but on the frontlines, they constantly changed that script by gathering input from women. They reminded mobilizers as well as beneficiaries to rely on their own experience of poverty in order to help other women. The facilitators may or may not have had an in-depth understanding of the context of every village, but they did understand that no village or community or 'elite group' or 'marginalized group' was homogenous and that there were multiple interests cutting across each category. Creating a common subjective understanding of the complexity of the problem and being sensitive to the diversity of interests was key to effective implementation. As one facilitator pointed out, "…we are not here to preach (hum pravachan dene nahi aate). This is an opportunity for us to ask every woman her own individual story of why she is poverty stricken. Some say, 'because my husband is sick or disabled,' some say 'because we routinely suffer abuse in the hands of our employers,' some say 'it is because of an outstanding loan,' some say 'it's just how things have been for generations.' We must hear each story. The women themselves must hear each story." 12 In other words, the SHG and VO meetings or the training meetings were treated as a space to contest and forge, rather than to impose, meanings of and solutions to poverty. The ultimate aim was to weave divergent narratives into the project's discourse and together agree on a set of principles and negotiated language. This then provided a framework into which the future actions of both the facilitators and the women would fit.
This kind of coproduction of discourse was key to the success of the project in its first phase, but, in Ostrom's (1996) terms, the "citizen contribution" of women to this discursive coproduction was hard to routinize. As a result, by the time the project got to its second phase and was under pressure to expand rapidly, facilitators trained with a homogenous and scripted discourse of poverty. On their part, the women, rather than arriving at the solution themselves, were told that the project itself was the solution. Far from being coproduced, in Phase II discourse was imposed from above, was closer to the script and lacked the kind of improvisation characteristic of Phase I. In effect, mobilization was rapid, but women found themselves alienated from the purpose of the project, and the project became an entity outside of them rather than synonymous with them. In the following sections, we discuss six different aspects of this argument, and how both implementation success and failure manifested in the field.

ANALYSIS: COPRODUCTION IN PRACTICE
This section outlines the three stages of community facilitation in JEEViKAvillage entry, mobilization and everyday life-and provides a systematic overview of how facilitators operated in these stages in Phase I versus Phase II.

I. Village Entry
The first step of a project at the community level is called 'village entry.' In this step, it is project protocol to first get approval from key stakeholders in the village and then conduct a social mapping to find the poorest households. Following this, facilitators proceed with gathering women from those identified households into the project's fold until they reach 'saturation' (i.e., when the project has mobilized 80 percent of its target audience).

a) Doing a thorough power analysis versus getting 'buy-in'
In Phase I, village entry began with a series of informal conversations with the key stakeholders in the village-local politicians, traditional leaders, moneylenders, school teachers, landowners and other influential men and women-as a means of information gathering. In these initial stages, ACs and CCs collected a large amount of information on caste dynamics, land-use patterns, informal moneylending practices and patriarchy, and tried to locate women's interests and positions in each village. "…it was our main task to figure out how to work within these setups so as to be as least disruptive as possible. We had to avoid being seen as threatening, but we also had to get them to reveal reliable information about the village." 13 In these initial stages, facilitators repeatedly emphasized that there was no substitute for time. For instance, one of the district level community trainers, formerly an AC in the first phase, explained to us how it took him between one to three months to get the information he needed to form SHGs.
"I remember going to a village once where for several months no one would speak to us.
The environment was extremely tense, but no one would tell us what was going on. It took six months of probing in and around the village to figure it out: We had happened to enter the village at a moment when a twenty-five-year-long feud between two dominant caste groups over a piece of land had reach its culmination and both groups were waging war against each other. The entire village was split up on the issue. We immediately took note of this and knew that village-wide meetings were impossible, as were inter-caste SHGs, at least in the initial stages. We had to find an alternative strategy." 14 There were insights gained in this stage that were specific to each village, but some that were universal as well. All ACs and CCs, for instance, mentioned three things that they learned early on. First, they understood that they were about to operate in a context with an extremely high density of failed projects that had microcredit components similar to JEEViKA. As a result, there was a strong mistrust towards such projects and villagers were skeptical of putting their savings into them. Second, the facilitators learned that a long history of identity politics had created a culture of 'caste-based handouts': People expected to be treated as beneficiaries, and expected more if they belonged to particular castes. Finally, they learned that the villagers' closest point of reference for outsiders coming to collect demographic data was the census officers who were derided for gathering all their information by talking only to key stakeholders and only stopping by the first few houses in the village. As we will see later, the project's messaging during the initial mobilization stages rested heavily on disassociating itself from these three things.
Once information was gathered, an ongoing conversation between facilitators and key stakeholders began in Phase I villages, in particular between local leaders from traditional caste panchayats and elected Panchayati Raj Institution (PRI) members (Mukhiya, Sarpanch, Panch, and ward members.) 15 One of the DTOs gives an example, "…In a few villages in Purnea district, for instance, we were not offered a room to stay in by the PRI members, and spent several nights sleeping on the streets or in the village temple. When they didn't want to listen to us, we started a sanitation drive. We began by going around the village and picking up trash, cleaning the public spaces, etc. We did whatever it took in those initial days to gain their trust and hung around till they finally listened to us." 16 Once the village heads were prepared to cooperate, a series of negotiations and network-building exercises began. "… their first question is always, 'What do we get out of this?' The hardest part is to get the Mukhiya and his men to think beyond the 'what are we going to gain' (humko kya fayeda) mentality, and focus their attention squarely on the fact that no one has anything to gain; we are here to help, not give subsidies. (hum sahayta dete hain, chhoot nahi) We give nothing, and ask for nothing in return…" 17 Phase I facilitators claim to have treated these negotiations with utmost care. They could not be seen by the rest of the village as colluding with or privileging certain social groups over others, but they could not be seen as antagonizing them either. "…[G]etting their blessings, and then respectfully bypassing them to reach the community was our hardest challenge. And it helps to be seen as social workers in this regard, and not as political brokers." 18 19 In Phase I, getting support was seen as crucial not only in the initial stages but was also constantly nurtured throughout the life cycle of the project. To give an example, two years into the project, once women started stepping out of their homes and going to banks, police stations, and other public spaces in order to get their JEEViKA-related Once key stakeholders in the village agreed on the merits of the project, the next step was to identify households to include in the project. This method of identifying the 'poorest of the poor' (or POP in JEEViKA's parlance) involved using census data and hamlet-level information, but also brought in the community to identify the most vulnerable households in a way that established greater ownership and legitimacy over the process.
Facilitators in both Phase I and II demonstrated a comparable understanding of this process. What set them apart is that in Phase I villages, ACs and CCs shared a healthy skepticism towards the process. They knew that it was susceptible to being captured by a few groups, that some voices were louder than others when identifying target households, and most importantly that this list would have to be revised with time as they gained a greater understanding of the community, and the community understood more about the project. In a nutshell, decisions about the community were made and changed in the field, with facilitators corrected by participants and vice versa, thereby establishing a model of engagement that was neither entirely bottom up nor top down but which evolved collaboratively on the frontlines.
In Phase II villages, facilitators collected data entirely as a means of arriving at a number of target households. The more qualitative data collected in this stage were not utilized during the process of mobilizing women. Moreover, while the ACs and CCs did the mapping exercise and gathered the necessary data, mobilization was ultimately left to CRPs from another village to whom the numeric target was transferred. However, the qualitative data and the informal knowledge on village dynamics collected in the initial stages were not transferred to these CRPs. As a result, often the households identified by the ACs and CCs in the initial round as the poorest were not the same the households finally mobilized into the project by the CRPs. 21 Former AC, Nauganj.

II. Mobilization
Succumbing to the pressure of numeric targets set by the ACs and CCs during the initial stages of Phase II forced the mobilizers to lead an enrollment drive fraught with shortcuts. This stage in JEEViKA, called the 'mobilization drive,' is a crucial moment in the project's life because the messages communicated in these initial stages had a significant impact not just on women's decisions to enroll but also to continue to participate.

a) Focusing on individual material outcomes versus collective capabilities
In both Phase I and Phase II, messaging was communicated, and mobilization was executed, through flipcharts. In Phase I villages, women were encouraged to engage with the information in the flipcharts, rather than memorizing them verbatim. Most flipcharts had images from folk tales that demonstrated virtues of solidarity, self-help, creativity and resourcefulness, especially in the face of adversity.
"…we thought of these as tools that we hoped women would use as and when necessary, whether in the livelihoods space or within their households or in any other space. As we shared them, we also emboldened women from different backgrounds to share their own stories of deprivation. A poor Dalit woman has a story, but so does a poor Brahmin woman. This kind of dialogue between the tools and different experiences of poverty is necessary." 22 In short, the Phase I facilitators allowed for JEEViKA to be a platform where understandings of poverty and its solutions could be contested and forged in a participatory manner. This was in contrast with Phase II, where a dialogue was missing.
Instead, facilitators went straight to discussing the merits of JEEViKA and the ways in which it could help women save money, start livelihoods and decrease their debt burden.
The women existed as an audience for messages that only the facilitators fully understood. This difference in mobilization resulted in marked differences in women's expectations: Women in Phase II repeatedly stated that they had not seen results yet.
They were frustrated that their short-term goals were not met and threatened to drop out of the project. Moreover, the moment of mobilization into the project barely constituted a memory for them: When asked to recollect it, they would struggle to remember and would provide lackluster responses. In Phase I villages, on the other hand, women recollected this as a rite of passage, narrating names of ACs, dates of joining, events, etc., with great enthusiasm. "I felt so much kinship with this project when Amit bhaiyya first came and explained it, that I suggested the name JEEViKA. We were all enthused about it. The idea that as women we could come together to do anything was transformational!" 23 In Phase II villages, women were told that a primary reason to join was "lower interest rates" (i.e., loans/credit at less expense than other informal sources) or "high paying jobs" or "toilets" (i.e., grants for building household toilets), whereas in Phase I villages, women were told that the reason to join would be "coming together to help each other" in order to solve their individual problems, such as reduced debt burden and eradicating poverty. In other words, while the mobilization process in Phase II focused directly on outcomes, mobilization in Phase I focused on developing collective capabilities and self-reliance.

b) Turning first movers into eyes and ears of the community versus using them as agents of the facilitators
When JEEViKA first enters a village, it attracts a set of first movers, (i.e., intrinsically motivated women who self-select into the project early and devote time to setting it up). In Phase I, facilitators got much of their work done through these first movers, making them their primary conduits into the community. Many of these women were taken to Andhra Pradesh for an immersion trip which left an indelible mark on them; They came back with a mission to re-create in their own villages what they had seen. An oft-repeated sentiment was "… our didis in Andhra Pradesh are so empowered, that they not only have access to banks now, but can also run all functions of a bank by themselves!" 24 Second, the ACs and CCs encouraged these first movers to share their experiences in their own neighborhoods in order to mobilize more women into joining 23 Cycle 4, Ramganj. 24 Cycle 6 and 7, Several interviews and FGDs, Ramganj and Saifpur. the project. In fact, for most women in JEEViKA, their first point of contact was not the mobilization event but instead their relationship with first movers in the village whom they trusted. When asked why they had chosen to join in the first place, it was not uncommon to hear them say "…because Shakuntala didi (a first mover) asked me to." The project relied on the first movers' knowledge of the community and their networks to mobilize women. They were truly seen as brokers whose work was to represent the community's interests and in the process also construct and cement these interests.
Because these brokers came from different backgrounds, the ACs and CCs learned to move away from a unitary representation of the 'poor.' They were able to understand and convey to potential participants that there were often multiple, even contradictory, interests within the category of 'rural poor women.' Finally, after the facilitators left the village, the first movers were the women who shouldered the responsibilities of the facilitators in their different neighborhoods. They tackled women's questions and concerns and helped sustain participation in the project. Many of these first movers also went on to become mobilizers and leaders in their SHGs.
The role of the first movers becomes even clearer when seen in comparison to Phase II villages where their centrality is missing to some degree. In Nauganj for instance, the facilitators' and CRPs' first stop was the home of Poonam Devi, a former mobilizer with SGSY, an SHG-based program similar to JEEViKA that had a long history of fraud in the village. Poonam Devi herself was seen to be involved in this fraud, but the ACs had little knowledge of this. As our researcher notes "…Poonam Devi has been a part of the loss-making SGSY group as noted by those who were reluctant to join.
Her presence made the community suspicious about the credibility of [JEEViKA] CRPs." 25 Far from being seen as trustworthy, Poonam Devi was eventually seen by the beneficiaries as project staff, representing the project's rather than the community's interests. There was also a lot of misinformation right after the facilitators and CRPs completed the mobilization drive and left the village. For example, men and women inquired about whether this was a government initiative, and whether they would have to travel beyond the village as part of JEEViKA membership, but there was no one to give them the information.

III. Everyday life of the project
Once the process of mobilization is over, the community requires a period of focused support to form SHGs and to ensure their regular functioning. Through direct observation of meetings, trainings and daily interactions between the project and the beneficiaries, this section details the role played by the facilitators and mobilizers during the everyday life of the project.

a) Ritualization versus rote recitation
One of the main contributions of the facilitators was to ritualize the process of a regular schedule of SHG meetings. As we outline in another paper (see , SHG meetings in Phase I villages were very disciplined, with the time and place of meetings treated with great sanctity. When groups were being formed, JEEViKA facilitators and eventually the CMs taught SHG members the significance of sitting together in a circle, and the implications of the formation for inclusion and voice. They also taught them how to introduce themselves in every meeting with their own name (rather than their husband's name, or their caste affiliation), and finally how to sing the JEEViKA song and prayer together. JEEViKA shorthand for these rituals is called the Panchasutra, 26 and one of the roles of the CM is to ensure that women follow these rituals and the principles embodied by them, because it eventually becomes a key criterion for establishing loan eligibility for an SHG.
The Panchasutra is a seemingly prosaic practice, but the orchestration of this social performance of introductions, songs, prayers and slogans, and their deliberate repetition every week for seven years has had a profound impact on the construction of a collective identity. It has given women a new group-based identity that is separate from their caste, marriage and kinship identities. This identity is transformative in its own right and can be mobilized at appropriate moments to make performative claims (Appadurai 2014: 18). For instance, the JEEViKA song, 'Badhte Kadam,' 27 gives women confidence, and they report enjoying the act of coming together and singing, but it is also leveraged appropriately in other contexts. Manju didi, a veteran CRP in Ramganj talks about how 26 Panchasutra literally means five principles, and JEEViKA's panchasutra includes regular meetings, regular savings, regular inter-loaning, timely repayment and keeping the book of accounts up-to-date. 27  28 Butler (2015) refers to the relationship between forms of linguistic and bodily performativity.
In Phase II villages, on the other hand, a failure to repeat these themes meant that Panchasutra, the songs and practices involved with coming together, were never ritualized in the same way. On the contrary, women struggled to remember the introductions, often thought of the song as a monotonous necessity repeated by rote learning and felt the rituals were burdensome or distracting to their core business of saving and borrowing. Because they considered the meeting space to be concerned exclusively with financial transactions alone-devoid of rituals-absenteeism in meetings was high, and many women sent money through other family members and neighbors who attended the meeting, rather than taking the effort to come themselves. As a result, JEEViKA, in its second phase, was less successful in creating a sense of community among women and thus relatively ineffective in building the foundations for what was to come.

b) Engaging head-on with elite capture versus keeping local politics at bay
Livelihoods projects are subject to the same risks of local elite capture and clientelistic politics as other community-based projects. Ensuring that decision-making around loan disbursements are inclusive and democratic requires a significant amount of handholding. In Phase II villages, a perverse mechanism was triggered because facilitators asked SHG members to 'elect' their leaders (i.e., the president, secretary and treasurer) without any supervision. 'Elections' took the form of all the SHG members coming together and nominating a few people for the three leadership positions by acclamation. Although such an election might have had the appearance of being democratic, in the absence of proper facilitation well-connected and upper caste women took control over the election process and pushed their own candidacies. In effect, such a method established a power relationship that was open to abuse throughout the lifecycle of the project. The most deleterious consequence of this was that SHG leaders could easily manipulate the rest of the group members to give preferential treatment to some over others. Along with this they became adept at representing their own interests as community concerns. In one of the groups, for instance, a few upper caste (Chaudhury) women were found to borrow repeatedly from the group at 2 percent interest rate and lend to non-joiners in the village at 5 percent. For a while, the remaining women, who were from the Halwai community, were unaware this was going on. On finally uncovering the malpractice they complained to the CM, who refused to take action because she was a Chaudhury herself. The consequences of this are far-reaching, particularly because it lowered trust and increased apathy towards the group and its processes over time.
In Phase I villages, however, ACs and CCs carefully engaged in community politics, not just in the initial stages of elections, but throughout the project cycle. First, the election of the SHG leaders was handled carefully, democratic selection through deliberation rather than a 'show of hands' election was emphasized, and members were encouraged to elect individuals with motivation and competence rather than influence or wealth to leadership positions. Moreover, the work of those selected was constantly supervised and every two years the leadership positions were rotated among the members to avoid entrenchment of power. As the project moved along and SHGs were federated at the village level, facilitators incentivized leaders to ratify whether or not the most marginalized women had been included in the project's fold and had access to loans.
Finally, and most importantly, the separation of powers between the CM, the group and its leaders was made clear by the facilitators: During meetings, they painstakingly repeated that leaders are the SHG's representatives and the group has the power to assess the work of and sanction the payment of their CMs, who on the other hand, are there to take notes and do the bookkeeping; it is up to neither of the two-group leaders or CMs-to arbitrate who gets a loan. The role of the members is to monitor the actions of their leaders, compel them to defend their promises, and spawn new leaders in the process.

CONCLUSION
In summary, in Phase I villages, facilitators took immense pride in being seen as 'fieldworkers' and, occasionally, in putting in extra hours for mobilization, while Phase II facilitators lacked intrinsic motivation and constantly felt stretched for time. In Phase I, facilitators positioned themselves as catalysts for a change process already on its way, rather than as initiators of change or determinants of the outcome. The refrain of "…neither have we come to take anything, nor to give anything" (na hum kuch lene aaye hain, na dene hain) drove home this message most clearly.
This resulted in stark differences in the impacts on women. First, women saw themselves as participants who were not just contributing ideas and relying on implementers to deliver but were also personally responsible for the success or failure of their SHG. In effect, they too went beyond their stipulated responsibilities to make the Second, in Phase I villages, facilitators succeeded in creating a safe and enabling environment for women themselves to take action in the public and private space.
Women reported that "…JEEViKA built our leadership skills and instilled confidence in us to tackle our community and our household. No school, no formal education can teach you that." 29 In meetings women were able to openly discuss gender norms that constrained participation for different communities and encouraged each other to come up with solutions themselves. For instance, in livelihoods activities, while most JEEViKA women invested in traditional handicrafts, others chose to 'rock the boat' by investing in what is considered as men's work (PDS procurement, running a 'kirana' or small dry goods store) or taking up work traditionally done by women of castes much higher than their own (e.g., Madhubani painting, which is traditionally done by Brahmins and Kayasthas, was adopted by Ram and Paswan women). Similarly, in the public sphere, while some JEEViKA women went on to run for Mukhiya's office and fight the incumbent, some chose to ally with the incumbent, while some chose the path of social work completely outside of politics. In short, Phase I facilitators were able to expand opportunity structures for women, building a woman-centered mini-social movement that 29 Ramganj, Cycle 7, Veena Paswan, CLF Treasurer, Paswan tola.
was leveraged beyond the project boundaries. This building of capabilities and creating a wider institutional environment that propelled women forward to break the shackles of normative injunctions outside the project was lacking in Phase II.
However, as the project moved into its third phase, the shortcomings of the second phase began to be understood. Many of lessons from Phase II were incorporated and changes were made to procedures. First, the number of CCs to the size of Panchayats in the field was balanced in order to ensure that the coverage was even across the state.
(Prior to this the number of CCs assigned to each Panchayat was fixed regardless of its size.) Second, emphasis was placed on updating the Management Information System (MIS) so that monitoring and any subsequent action necessary could occur in a more timely manner. Third, the project began a pilot program whereby the data generated from the MIS was taken to the field, so that in the case of delays, SHGs could take action themselves. Fourth, at the state level, the organogram changed in order to increase the number of State Project Managers (SPMs), Project Managers (PMs) and regional managers needed for some key thematics, especially those important for institution-and community-building. And finally, frontline workers began to be graded not only on the numerical targets achieved in the field in terms of number of SHGs mobilized, but also on their theoretical and practical knowledge base, thereby attempting to incentivize quality as well quantity of mobilization.
To conclude, this paper summarizes the dynamics of community-level facilitation in a participatory development project in rural Bihar. The paper draws on qualitative analysis spanning four years. By contrasting the project in two different phases-in its introductory phase and then in its expansionary phase-we find that the difference in outcomes between the phases can be explained in part by the fact that scaling-up brought new organizational dynamics that prioritized targets over processes and quick expansion over improvisation. 30 By identifying six different stages and processes of facilitation and showing how each of these come under strain as the project scales up, we explain implementation successes and failures within the same project.
The arguments presented in this paper have implications for both policy and research. First, if induced participatory development projects desire to mimic organic social movements, a shift is required in project culture and incentive systems. Regardless of the phase of the project, facilitators must be incentivized to learn, adapt and improvise on the frontlines by gathering input from those who are beneficiaries of the project.
Second, our analysis suggests that scholars must spend as much time understanding the processes of implementation of a project as we do in understanding its impact. The escalating pressure to rapidly scale up such projects without first adequately understanding the mechanisms underlying their successes and failures can dilute project effectiveness and undermine the potential for lasting change.