The coordinates of scaling: Facilitating inclusive innovation

The desire to ensure that the benefits of successful small-scale social innovation are more widely available has led to a plethora of frameworks that seek to scale such innovations. We review 20 extant frameworks for scaling and distinguished four directions: up (producing changes in laws, policies, institutions or norms), down (resource allocation to support implementation), in (ensuring organizations have the capacity to deliver the type and number of good practices required) and out (geographically replicating or broadening the range or scope of good practices). In addition to these directions of scaling a generic pathway, or process, to achieve scaling is also discernible across many of the frameworks reviewed. This involves five phases: identifying, planning, implementing, learning and adapting. We stress the need for a more dynamic and systemic approach to scaling, as well as one which anticipates, addresses and assesses the extent to which scaling is inclusive of marginalized groups.


| INTRODUCTION
There is a growing interest in scaling social innovations 1 in development to increase their social impact. In recent years, there has been a flourishing of 'scaling science' to improve social impact and benefit society (Gargani & Mclean, 2017). Moreover, while scaling up 2 is frequently mention as desirable, there are only a few frameworks that address how to influence scaling so that socially excluded populations also benefit. The existing scaling frameworks in the literature rarely incorporate specific strategies that require planning and implementing scaling that considers the needs of the most marginalized groups.
The purpose of this paper is to conceptualize a scaling framework that parsimoniously incorporates the essential components of the plethora of extant approaches to scaling and to strengthen a concern to 'leave no one behind'. We emphasize that scaling socially inclusive innovations usually occurs within the context of complex adaptive systems (CASs) and that within this context there additionally needs to be a stronger and clearer focus on the inclusion of marginalized groups; and we suggest how this can be achieved. Furthermore, although existing frameworks prescribe a series of steps to achieve scaling, they rarely consider the complexity, or the 1 We consider socially inclusive innovations to be 'the development and implementation of new ideas which aspire to create opportunities that enhance social and economic wellbeing for disenfranchised members of society' (George et al., 2012, p. 663), promoted through organizations whose primary purposes are social (Mulgan, 2006, p. 146). 2 The use of use scaling up and scaling is used interchangeably in the literature. unpredictability, of the broader system they aim to change (Paina & Peters, 2012).
A CAS approach addresses the interaction amongst different agents that are in and out of the system to be scaled (Hall & Clark, 2010). A CAS approach is potentially useful to understanding scaling in complex environments and to understand scaling failure (McVeigh et al., 2016;Paina & Peters, 2012). Furthermore, intervening organizations often overlook the interaction between the systems that they want to change, and the root causes of exclusion, or discrimination, in the scaling process. With the emphasis on 'leaving no one behind' in the Sustainable Development Goals (SDGs, 2015), this should be addressed and recognized as a priority. Given that the goal of scaling effective interventions is to promote social impact, it is important to have a scaling framework that analyses exclusion and discrimination, from a systems perspective.
The paper is divided into two sections. The first section describes existing scaling frameworks: how scaling is conceived and recommendations for its achievement. Using a constant comparison methodology, we identify five common phases of scaling and four distinct directions of scaling, representing a parsimonious synthesis of the range of terminology currently in use. The second section suggests an inclusive scaling framework that addresses inclusion and suggests a range of participatory mechanisms to centre inclusion as a scaling objective. By stressing inclusion as an implicit aim of scaling, we seek to refocus thinking about scaling into what social impact would mean for marginalized communities-for reaching towards the marginalized and promoting social benefit for all.

| METHODOLOGY
In order to understand how scaling happens and is promoted for socially inclusive innovations in development, we conducted a systematic search and review (Grant & Booth, 2009, p. 102). The scope of this review was to explore scaling frameworks that addressed socially inclusive innovations targeting vulnerable populations. We narrowed the review to the following inclusion criteria: (a) scaling up of innovations in development contexts (primarily low-and middle-income countries) and (b) for identifiable discrete innovations as interventions, good practices, or pilot projects of development actors, such as community-based organizations, non-governmental organizations (NGO), international NGOs (INGOs) and the public sector, with the intent to improve the living conditions, which included health, education, employment, housing, land tenure and agriculture, and technology.
The literature covered from 1995 to 2019 and considered the first studies on scaling frameworks in development (Uvin, 1995). The types of documents reviewed were as follows: (1) scientific academic research on scaling frameworks for innovations and (2) grey literature of reports and manuals on scaling frameworks from local, national and international organizations working on development, for example, US Agency for International Development (USAID), World Health Organization (WHO) and World Bank (WB). We excluded scaling up in nondevelopment contexts, businesses and scaling business-like models.
The data were collected, first, by conducting a literature search in Academic Search Complete, Web of Science and Google Scholar that included the key words 'social innovation' 'development' 'scaling up approaches' OR 'scaling up frameworks' OR 'scaling up methodologies', which resulted in 13 entries in Google Scholar, two in Academia Search Complete and four in Web of Science; when changing the term to 'innovation', the results were 189, 36 and 30, respectively. We extended the search to 'innovation' because using the term on 'socially inclusive innovations', the search yielded in no results. We eliminated the duplicates, reviewed the abstracts and selected those articles that mentioned a scaling framework with a number of steps to scale.
We also added ExpandNet, 3 a known source for scaling that has a scaling-up bibliography organized in 10 themes, with 369 titles focusing on scaling and 92 focusing on scaling up frameworks, with another 36 titles on development. We eliminated the duplicates with the articles selected on the first step, resulting in 52 new documents.
The frameworks were selected by reviewing the articles applying the following criteria: (a) covers a description of a number of steps to achieve scaling and/or (b) describes scaling directions and/or (c) defines scaling. Finally, we filtered those results that had used the same scaling framework. The frameworks were classified by policy area of focus and, if applicable, by organization sponsoring the framework. We used a constant 3 ExpandNet is a global network of practitioners, scholars and academics to advance the science of scaling up. Available at: https://expandnet.net. Last accessed May 2020. comparison analysis (CCA) to explore the differences and similarities of the 20 frameworks, and as a result of that analysis, we came up with five stages for scaling and four directions addressed in the next section.
The purpose of the CCA is to create new theory and/or to analyse documents (Leech & Onwuegbuzie, 2008;Onwuegbuzie et al., 2012). The CCA method has three stages of analysis: (a) open coding (identify thematic codes relating to particular ideas/activities/ actions); (b) axial coding (identify codes into similar categories) and (c) selective coding to, in our case, create a new framework, but with the continuous process of monitoring new data (document analysis) to compare with the collected data (already analysed documents) (Onwuegbuzie et al., 2012, p. 13). The process of data collection reaches a point of theoretical saturation when no new categories are found in the subsequent analysis; in this case, no other different distinct stages were found in the scaling frameworks analysed (Onwuegbuzie et al., 2012).

| Findings-20 scaling frameworks
In order to understand how and when scaling might happen, we felt it was important to explore and analyse the different extant frameworks. In general, scaling frameworks define a direction of scaling and include a set of steps to follow; some also show the types of strategies employed to scale practices, programmes and policies.
The CCA open coding indicated different levels of completeness of the 20 frameworks reviewed; for example, if they have an explicit definition of scaling, refer to specific units of scaling or include steps to scale. A second stage of coding considered if a framework identified process issues to scaling. The third stage developed coding for five common phases, using a common language and is described in Sections 2.1.1 and 2.1.2. Table 1 summarizes the 20 approaches to scaling identified through the literature search. The scaling frameworks in Table 1 do not address differences between types of organizations and their paths to scale. For example, differences between scaling for a national programme that is implemented by the government and for a local project implemented by a community-based organization are not addressed. Indeed, the literature of scaling describes the approaches of scaling for non-profits, international NGOs, governments, social enterprises, communitybased organizations and private sector, as essentially being the same. The Nesta UK framework defines the scaling target as the social innovations, defining these broadly as 'new products, services and models that both meet social needs and create new social relationships or collaborationsthey're "social" both in ends and means' and 'can be generated from within any sectorpublic, private or socialor from citizens and social movements. They may generate financial value, but don't have to' (Gabriel, 2014, p. 7).
The frameworks aim to scale pilots, projects, programmes or policies, often advocating for a variety of practices-not well defined-that leave their implementation to their use in the field. An exception to this is that the Self-Evaluation for Effective decision-making Systems for communities to adapt learning and expand (SEED) addresses only community-based projects as the 'size' of the unit to be scaled to larger programme adoption (Taylor & Taylor, 2003). More consideration is therefore needed to understand the practicalities of scaling for one type of organizational entity versus another type, as well as a set of interventions that are contextually relevant. The focus in these frameworks is generally on the reasoning to scale and not on the scalable unit for the frameworks, whether that is a pilot, a programme or a policy to scale. However, the Scaling up Management (SUM) framework is one approach that does differentiate scaling process by the scalable unit: pilot, demonstration, capacity building and campaigns; for example, in a pilot project, scaling up comes after the innovation phase (Kohl and Cooley,n.d.,p. 4). The Nesta UK approach explains scaling of social innovations in a spiralling process to achieve systemic change, in six stages: (1) prompts, inspirations and diagnoses; (2) proposals and ideas; (3) prototyping and pilots; (4) sustaining; (5) scaling and diffusion; and (6) systemic change (Murray et al., 2010, p. 12-13).
In the 20 frameworks, scaling is a process that usually goes from small to big; some frameworks like 'Making it Big' are more explicit in this regard than others, such as the learning process approach (LPA). ExpandNet is one of the approaches that describes in more detail in nine steps and considers scaling as 'expanding, replicating, adapting and sustaining successful policies, programs or projects in geographic space and over time to reach a greater number of rural poor' (ExpandNet, 2010, p. 17). Other definitions of scaling include the improvement collaborative approach definition that focuses on the growth of the intervention from improvements that serve a small group to 'a significantly larger population, such as an entire region or country' (USAID, 2008, p. 20). The International Fund for Agricultural Development (IFAD) approach addresses scaling as quality of impact and sustainability (Hartmann & Linn, 2008, p. 8). This framework includes drivers and spaces; the drivers are the enablers to scale up (e.g. strong leadership), and the spaces are opportunities or potential obstacles to scale up (e.g. policy space). In the frameworks reviewed, scaling T A B L E 1 Scaling frameworks (inclusion criteria and constant comparison analysis applied)  In Milat et al. (2015).

Name of framework
in itself, is either a goal or a step to achieve sustainability. The 'five configurations for scaling' approach considers scaling 'to make a durable and profound change' (Westley et al., 2014, p. 3).
The 20 frameworks address scaling as a phase but, differing in the number of phases as well as the components they drop in for each phase. Barker et al. (2016) identified two models of global health scaling frameworks: (1) nonsequential and (2) sequential approaches. The latter follow a particular ordering of phases to achieve scale. The 'framework for going to full scale' explains linearity from developing the scalable unit to testing it and then to fully scaling it up (Barker et al., 2016, p. 7). The 'scaling up population health intervention guide' underlines the steps in an order with their objectives, strategies and challenges, as well as different tools required for each phase. The 'State Implementation and Scaling up of Evidence-based Practices' (SISEP) describes an implementation process to scale from the exploration and adoption to innovation and sustainability. The SUM framework approach emphasis is on the scaling plan to identify the need to scale and to establish the preconditions to implement the scale as well as the Implementing Best Practices Consortium (IBP) approach that begins defining the need for change. A few frameworks incorporate a final stage to evaluate the scaling process such as the IBP and the LPA. Moreover, the LPA considers participation and knowledge transfer as a precondition to scale up. The LPA stages aim to progressively achieve the organization's maturity by achieving expansion as the ultimate goal in a way that the organization is able to address new problems and create new solutions to replicate (see Figure 1). The SISEP, like the LPA, highlights practice improvement and evaluation as key, and the innovation comes at the end of the process. The SISEP emphasizes the need to develop capacity in terms of professional development and practice improvement. The SISEP includes identifying different stages in the process to implement the project and the importance of evaluation. The transition to scale is depicted in most of the approaches in a sequential order; however, for some like the GHLI-AIDED (Assess, Innovate, Develop, Engage, Devolve), the stages are not followed one after the other, but these can be reiterative over the process.
Some of the scaling frameworks have a higher level of complexity than others by delimiting the scaling directions, whereas others promote an understanding that scaling takes place intertwined in complex systems. The GHLI-AIDED, ExpandNet, the Five Configurations for Scaling up Social Innovation and the SEED emphasize unpredictability and complexity; that scaling does not happen in a vacuum. Complexity is understood as a system where the intervention adds or contributes to part of a change, the GHLI-AIDED specifically addresses CAS; unpredictability is part of the scaling process. The IBP and the GHDLI address scaling complexity as part of a larger systemic view of change. The Five Configurations for Scaling Up Social Innovation is a CAS model to scale; the scaling departure point is the analysis of complex systems (Moore & Westley, 2011;Westley & Antadze, 2010) or cross-scale interactions or 'panarchy'. 4 The principal characteristic of the model is the unpredictability of the factors associated in scaling social innovations. This approach is useful in understanding the fluidity of change in an organization that is aiming to scale up. Another framework that addresses context as a key factor is 'Guide for Fostering Change to Scale Up Effective Health Services', which addresses scaling strategies 'that best suits the environment' (2007, p. 26). The environment is also a factor in the ExpandNet framework defining it as the 'conditions and institutions that are external to the user organization but fundamentally affect the prospects for scaling up' (ExpandNet, 2010, p. 16).
Most of the frameworks have focused on scaling health systems such as the GHLI-AIDED that seeks to disseminate innovations and understand how scale up works in low-income countries. The same is true of the 'improvement collaborative approach', which can also be used in other policy areas (Milat et al., 2014, p. 4). Frameworks like the SUM promote a generic set of questions to develop any scaling strategy: the what (model), the how (methods), the who (organizational roles) and where to scale up (dimensions) (Kohl and Cooley, n.d., 2003, p. 2). It is also clear that scaling is interdisciplinary and cuts across different fields such as health, education and agriculture (ExpandNet, 2010;Fixsen, 2009). Furthermore, scaling frameworks involve multidimensional processes (Hartmann & Linn, 2008) that require some common elements, such as favourable policies, collaboration and developing organizational capacities, amongst others.

| Five common phases across scaling frameworks
Across the frameworks we reviewed, there is no consistent understanding of the elements of a comprehensive framework to support organizations to scale. Some frameworks are better developed than others, to include different scaling dimensions and a detailed phase process of how to scale. However, there are stages to scale that overlap in these frameworks, and we use the constant comparison method described above; we identified five emergent common phases that are now described below: identification, planning, implementation, learning and adapting (IPILA) (see Figure 2). In Melanesian mythology, Nuga is the father of the Kiwaians of New Guinea. He was carved from wood by Ipila. To avoid being lonely, Nuga asked Ipila-to 'scale up'-to carve three more like himself.
1. Identifying: Identifying the scalable unit is the starting point for scaling. A definition of this varies according to the different frameworks. 'Identifying' include subcodes that emphasize context assessment and suitability of the innovation. As an example of the first, the GHLI-AIDED emphasizes the environment and the conditions rather than the innovation itself (Bradley et al., 2011, p. 18). The SISEP framework (Fixsen et al., 2013, p. 2) calls this phase an exploration and defines it as 'identifying the need for change, learning about possible interventions that may be solutions, creating readiness for change, learning about what it takes to implement the innovation effectively, developing stakeholders and champions, deciding to proceed (or not)'. In Table 1, a number of frameworks include this phase to varying extents (see Frameworks). A practice classification hierarchy (Hancock, 2003;Jonasova & Cooke, 2012) may be used to classify practices by the level of the evidence provided and to estimate their general applicability. The classification starts from a basic level of an identifiable discrete community practice, which is considered the 'innovation', with the highest level of unit to scale being a 'policy principle'. In general, the scaling frameworks refer to scaling of 'good practices' or 'promising practices' that are small-scale projects with some evidence that they can be replicated. 5 2. Planning: Most scaling frameworks (see Table 1: e.g. 2, 3,4,5,7,9,11,13,17,18,19 and 20) include a scaling planning process. This is the stage where the organization analyses what is feasible and has collected enough evidence to replicate the practice. The tools to plan the process differ, but most of them detail the steps to follow and the strategies to employ to scale the practice. The scaling up population health intervention guide in Milat et al. (2014, p. 13) addresses a basic question to plan scaling: 'Has a plan that creates a vision of what scaling up will look like and a compelling case for action been developed?' The SUM scaling planning is a four-task process that includes the creation of a vision in the first place (Cooley & Linn, 2014, p. 7, figure 5). 3. Implementing: A stage that covers a variety of strategies that are context driven. This phase will cover resource mobilization, stakeholder engagement and training to improve capacity to perform well and deliver (see Table 1: e.g. 2, 3, 5, 9, 10, 11, 13, 14, 17, 18 and 20). For example, the Guide for Fostering Change to Scale Up Effective Health Services includes a preimplementation phase that is supporting demonstration to then proceed to scale and selecting the 5 Jonasova and Cooke (2012, p. 6).

F I G U R E 2
Common stages in the scaling frameworks appropriate scaling strategy (quantitative, functional or political scale-up; IBP, 2007, p. 26). The improvement collaborative approach includes an implementation package that needs to be based on what already works, and its components will be defined by what already exists (USAID, 2008, p. 7). 4. Learning: The learning process provides the validation to scale and requires knowledge transfer inside the organization and amongst others. For instance, the improvement collaborative approach highlights shared learning as one of the steps, and Korten's learning approach is about organizational learning and using it to acquire knowledge. The learning approach includes learning to be (1) effective, (2) efficient and (3) expand (Korten, 1980). The SEED approach includes three learning dimensions at the community level from the paternalistic approach to one that empowers communities (Taylor & Taylor, 2003; see Table 1, e.g. 1, 5, 6, 7, 8, 11, 13, 15 and 17). 5. Adapting: The process of scaling may be unpredictable and complex; some but not all the scaling frameworks take account of this. The adaptive models emphasize that scaling is not a linear process. The GHLI-AIDED is a nonsequential model that emphasizes the interconnection and non-linearity of their components. This approach addresses scaling up as multifactorial, hard to predict and in which sometimes it is difficult to directly identify cause-effect relationships. This CAS ethos is also a feature of the IFAD approach, as it defines the scaling process as adaptive; likewise, the SEED approach uses an adaptive learning model as a continuous exchange amongst the community and other stakeholders involve (see Table 1, e.g. 2, 8, 11, 15, 16 and 17).
We recommend the attempts at scaling address, in a deliberative manner, each of these five phases, but we do not prescribe the exact methodology for doing this. Specific methods can be gleaned from existing approaches but should in the first place be contextually relevant and emergent, ideally through participatory approaches. As well as identifying five thematic phases in the scaling processes we reviewed, we were also able to discern four emergent directions of scaling from the frameworks in Table 1, column 5.

| Finding a common language for the Scaling Directions
The frameworks we reviewed offer a plethora of terminologies that were sometimes overlapping and sometimes contradictory. We now briefly consider some of these. Uvin (1995) defined 'scaling down' as 'processes whereby international organisations (IOs) change their structures and modes of functioning to allow for meaningful interaction and cooperation with grassroots organisations and NGOs ' (p. 495). Scaling down refers to processes that aim to increase impact without becoming larger and focusing on fewer strategies (Uvin et al., 2000(Uvin et al., , p. 1416. Hancock (2003, pp. 5-6) distinguished 'scaling down' as shifting responsibilities to a lower level by deconcentrating and devolving, 'scaling out' as replicating between countries and 'scaling up' as advocating to influence policy (Korten, 1980). Moore et al. (2015) distinguished between 'scaling up' and 'scaling out' and introduced 'scaling deep' as a direction that implies a change in the culture, values and beliefs (Moore et al., 2015, p. 75). Hartmann and Linn (2008, p. 14) explained three directions to scale: the first is expansion and entails scaling a pilot project that, in some cases, the current organization might not be able to carry forwards and needs to pass it on to another organization with different capabilities. The second direction is replication and occurs through a franchise model between different types of organizations, for example, from NGOs to government. The third direction is spontaneous diffusion and is spreading the practices through replication where the spillover has formal or informal channels. Hartmann and Linn's formulation thus interweave process and direction.
Another example, ExpandNet, focuses on the effectiveness of interventions, growth, expansion and replication, mostly of health interventions. The first two directions to scale occur 'when authorities at high levels of government were persuaded that an approach adopted at a lower level of government was worthy of replication (horizontally) at the same level or (vertically, upward) at higher levels, when donors drew the same conclusion, or both' (Manor, 2007, p. 18, emphasis added). A third 'direction' is similar to the 'spontaneous diffusion' dimension explained by Hartmann and Linn (2008) and is also called 'spontaneous'. The functional direction is similar to the dimension defined by Uvin et al. (2000) and is when the organization increases the number of activities. There are other directions comprised in the table that emphasize the types of strategies used, for instance, a direct approach that is to increase impact by scaling operational expansion, scaling through advocacy or multiplicative activities (Edwards & Hulme, 1992a, 1992b.
As we have illustrated, the scaling literature is, well, scaling! However, it is developing with overlapping and sometimes inconsistent concepts such as directions, processes and dimensions. Given the importance of the idea of scaling to overall social gain, it is necessary to develop a parsimonious nomenclature, capable of incorporating existing ideas but also one that provides a framework for further research and practice, especially, one that promotes social inclusion. Table 2 suggests a reclassification of scaling directions that folds multiple scaling types into easily understandable terminology with face validity; these have been derived from those found in the frameworks reviewed in Table 1. Our conceptualization does not necessarily seek to replace existing particular models or approaches but rather to allow a more effective synthesis of research and practice across commonly understood directions of scaling. We also differentiate 'scaling' from 'scaling up', which is often used interchangeably and confusingly. Scaling up in our classification seeks changes at the structural level in policies and laws. Scaling out requires the organization to replicate their model geographically. These two are the directions that are currently named in the literature. Scaling in and scaling down are two directions that are often confusingly subsumed in scaling literature but actually require different strategies and have different goals. In the first, the focus of change is within the organization, while in the second-scaling down-the focus is more on the community and context of change. Table 2 summarizes the coordinates-up, down, in and out-of scaling directions. In reality, scaling will often involve moving in several directions at once; recognizing that CASs move with change, with components moving at different rates and possibly in different directions; and that linear attempts to scale in a single direction may be unrealistic and ineffective. However, each systematic approach to any of these directions of scaling should require passing through the particular phases already described above. However, once again, these should be seen as organic and dynamic, not as categorical, fixed or restrictive. In Table 2, we indicate the four directions and associated descriptions, strategies, goals and examples.

| Socially inclusive scaling
The scaling frameworks reviewed in Table 1 do not explicitly address exclusion and discrimination, nor is the CAS perspective often explicit in this regard either. ExpandNet includes the principle of respect for human rights, equity and a gender perspective (ExpandNet, 2010, p. 8); but as to how to include specific strategies to make that happen is not clear in the framework. In this section, we suggest to how to make explicit a scaling approach that incorporates vulnerable populations into the process of scaling, thinking of scaling not as a way to increase numbers of people but to include people who have been excluded. Scaling failure often arises through not reaching vulnerable populations, such as persons with disabilities (Amin et al., 2011) because the models used failed to recognize the complexity of the broader system in which they operate and how this may marginalize some groups.
Scaling frameworks should introduce nondiscrimination as a key feature and address inclusion as a process and goal (Huss & MacLachlan, 2016). Making scaling inclusive is challenging and demands different types of strategies to achieve inclusion in order to tackle different needs amongst vulnerable groups . Carter et al. (2018, p. 3) highlighted a number of challenges that are specific for inclusive scaling practices: (1) understanding the wider contextual ideologies and vested power of individuals and groups, (2) reaching the most marginalized, (3) dealing with longer times frames, (4) coping with reversals and backlash when working on political and culturally sensitive issues, (5) turning theoretical models and emergency evidence into clear operational guidance and effective practice and (6) learning how to measure the impact, cost-effectiveness and sustainability of interventions (Carter et al., 2018, p. 27, our italics in the above).
Scaling frameworks focus on the product and evidence of previous intervention successes, rather than on the processes and outcomes for vulnerable groups, whose 'outcomes' are often poor compared with the mainstream. This, of course, is the very reason for the emphasis in the SDGs on 'leaving no-one behind'. Sometimes, the task practice of scaling can become the centre of the scaling strategy, diverting attention from who gets toand those who do not get to-participate in it. Carter et al. (2018, p. 8) also argued for the scaling processes to be explicitly inclusive, using interventions that target the most marginalized and incorporating inclusive goals that work to change social norms. For scaling inclusive interventions, the speed of the scaling and the cost are important. Equity in scaling may well mean that interventions targeting the most marginalized communities will be more expensive and take longer (Carter et al., 2018).
For each of the five-scaling process and four scaling directions we have identified, Carter et al.'s above challenges for inclusiveness should be considered. We do not recommend specific questions to probe for inclusiveness in a particular way but rather an ethos of assessing the extent of inclusiveness, appropriate to the local setting and conditions that constitute the specific context. It may be in some contexts that certain aspects of inclusion are more necessary to address than others. Examples of how the five scaling phases can be made more inclusive are provided in Table 3. Table 3  promote more inclusive policy development and evaluation. The dimensions were derived empirically from an extensive review of the relevant research and also informed by United Nations declarations and conventions. What the key issues are for inclusion in any scaling project should of course be determined in a participatory and inclusive manner, as part of the first phase of scaling-the identification process, so that the identification phase becomes not just about what to scale but also about how to do it in an inclusive way. It is important to try and assess and have some type of measure of the degree to which inclusion is achieved. This allows us to identify which aspects of inclusion-or groups of people-have been easier to address than others and may allow projects to benchmark against future performance. It also allows projects to identify how well things work in the different contexts. Table 3 indicates how such an assessment method can be used to measure inclusion. As illustrated in the table, this approach can be used to assess the extent of socially inclusive scaling for each of the five phases and therefore for each of the four directions relevant to a particular attempt to scale. Figure 3 presents our composite, three-pronged approach to scaling, incorporating (a) the five phases of scaling, (b) the four directions of scaling and (c) the multiple points across the phases and all directions, where inclusive actions should be taken to ensure that socially just scaling occurs and that marginalized groups are not left behind.
The application of socially inclusive scaling that we have adopted supports a right's-based approach. Although we have focussed on the approach of Huss and MacLachlan (2016), other approaches may also guide the identification of crucial themes for inclusion. For

Sources of Evidence
Adapting to more inclusive practices (adapted form Themes 1 and 2, from Huss & MacLachlan, 2016).
Following the first cycle, this phase may overlap with several of the previous phases, especially implementation, and should be with the targeted population's participation. An organization works with their targeted populations on their plan with potential partnerships to facilitate quick changes.
How is the organization adapting to unanticipated challenges concerning inclusion?
Risk assessment and possible solutions emerge from previous consultation with the targeted populations.
A reviewed working plan whose modalities can change according to what is needed and is engaged with by the targeted populations How is the organization addressing unpredictability without compromising the inclusion of vulnerable populations? A specific group is charged with the responsibility to explore how things are now being done differently form before-how the work has evolved to be more inclusive.
F I G U R E 3 A three-pronged approach for a socially inclusive framework example, Forgasc (2011) highlighted the importance of practices that are participatory, accountable, accessible, equal and nondiscriminatory results-based, and appropriately resourced, which recognize the interaction between gender and disability, involve partnerships and are replicable and appropriate to recognize the context-specific requirements, if the practice is to be transferable to other contexts (Forgacs, 2011, p. 8). Han and Shah (2020) recently developed 'Ecosystem of Scaling Social Impact' framework, which considers financing, organizations, technology and data, strategies, institutional infrastructure and government policy and, although complimentary to our own, should also take account of just how greater social inclusion can be built in to the process of scaling. Indeed, regardless of the overall approach adopted, it is essential to include targeted groups, from planning to implementation and evaluation. In our view, a common mistake for the scaling frameworks reviewed previously is to see the intervention as an independent unit, disconnected from, rather than embedded in, the broader system that it must influence.

| CONCLUSIONS
There is great interest and value in understanding how successful scaling can be achieved. Scaling frameworks provide different perspectives on how scaling should happen; however, they generally do not promote inclusiveness as a core element in their approach. We argue that unless inclusion is integral in the process of scaling, then the harder-to-reach will be excluded, further contributing to their marginalization. No attempt at scaling social interventions can be considered legitimate, if it fails to reach those who may benefit most. Our review of 20 different approaches to scaling found commonalities, and we have integrated these into five phases and four directions of scaling: in, out, up and down. We welcome comment and empirical data exploring both the theoretical and practical value of the three-pronged approach (phases-directions-inclusion) outlined here.