Product Development Partnerships

A strategy for modernizing the social sector tech stack?

By Edmund Korley and Temina Madon

As some of you may recall, The Agency Fund launched our first program in partnership with the technology community South Park Commons. Ever since, we have remained optimistic about technology’s role in social service delivery. We envision a global web of digitally powered social services that uphold people’s dignity and agency, delivered by organizations using data to learn and evolve like tech startups. We imagine social institutions that can adapt rapidly to new contexts – continuously deepening their cost-effectiveness and impact as they scale. 

In this sense, we’re hopeful about the promise of AI for the social sector. Yet as organizations rush toward these new technologies, we see some worrying trends. For example, many social sector organizations assume they need to build custom AI infrastructure. They are hiring consultants and contractors to reinvent the wheel – without realizing that free and open-source options often exist. Also, relatively few social sector organizations have the tools to carefully evaluate the performance of AI models in production, which is risky when serving vulnerable populations. 

Of course today’s AI products and platforms aren’t built for the social sector; they’re built for private sector software developers. The staggering pace of innovation also makes it hard for stretched nonprofit engineers to keep up. And nonprofits do have unique needs – from lack of network connectivity in the field to high data costs and lower literacy among users. With few market incentives to build and maintain software for the social sector (see the challenges facing Open Data Kit), these problems won’t solve themselves.

To help, some donors and large corporations have begun connecting private sector engineers with social sector organizations, usually as volunteers. But accommodating engineers – and maintaining what they build – can be costly. Especially with new technologies like LLMs, donors don’t always know how to deploy help, who to work with, or where to begin. 

This blog shares our recent experience tackling these problems, highlighting our emerging product development strategy and sharing a vision for how donors might join forces to accelerate the building of tech solutions that work for the social sector.

Learning from neglected tropical diseases 

There’s a useful analogy for this challenge: the successful development of drugs, vaccines, and diagnostics for malaria, dengue, tuberculosis, and other tropical diseases.

While designing software for social service providers is a far cry from developing life-saving drugs for infectious disease, the two face similar market failures. At the start of this century, tropical diseases were almost completely neglected by the pharmaceutical industry, largely due to the lack of potential profits. Most people suffering from these diseases were living in low-income countries, with limited purchasing power. 

In the early 2000s, a group of activists, academics, scientists, and donors established bridge organizations, called product development partnerships (PDPs), which have effectively served as “pseudo” pharmaceutical companies. They’ve pioneered new products, while also creating new financing models to stimulate private sector investment in neglected disease R&D. PDPs were the first to implement an “advance market commitment,” for pneumococcal vaccines; this financing mechanism was later used to accelerate the development of COVID-19 vaccines.

By coordinating across partners, today’s PDPs – like PATH, the Drugs for Neglected Diseases initiative (DNDi), the Foundation for Innovative New Diagnostics (FIND), and the Global Alliance for Vaccines and Immunization (GAVI) – are undertaking the full scope of biomedical R&D. Through grants to university partners, they discover novel drug and vaccine targets and new diagnostic technologies. Where promising leads are found, they contract with a range of private firms for the downstream development, clinical testing, manufacturing, and distribution of life-saving products.

PDPs have established an entirely new operating model for innovation. Designing tech solutions for the social sector may require a similarly paradigm-bending approach.

How we’re tackling the challenge

In our efforts to develop software for the social sector, The Agency Fund has taken some lessons from the tropical disease playbook: in many ways, we are acting like a pseudo tech company. 

In fact, we are an impact accelerator for nonprofits and social ventures, investing in organizations that activate human agency. Yet many of our grantees leverage scalable technologies (like AI chatbots and digital media) to support the delivery of social services. We work closely with these organizations to adopt emerging technologies. While we do not build or commercialize products end-to-end, we support product prototyping, integrations, reference architectures, and user testing.

Embedding with our grantees gets us close to the “customer base” we care about. It has taught us to prioritize usability and design, and to avoid one-off solutions that are customized to any one organization. We have come to embrace both open-source and commercial products, where they exist, rather than trying to build everything from scratch. This has helped us develop a wider view of the actors and ecosystems needed to deliver purpose-built software solutions for the social sector.

We can do this work because we’ve built a team with expertise in user research, product, data pipelines, experiments, and software engineering. Working alongside grantees, we design open-source proofs of concept (POCs) that solve specific problems inside an NGO. We then generalize these solutions, with the long-term goal of modernizing the nonprofit tech stack. So far we have created platforms for data warehousing and experiments automation (i.e., A/B testing across digital and in-person environments), and there is more to come… 

Like a PDP, we have partnered at every phase of product development. There are many areas of engineering and go-to-market that we do not support directly (e.g., design, site reliability, technical support, marketing, sales). For these and other needs, we collaborate with software companies that already know how to deliver and scale low-cost software-as-a-service (SaaS).

What this looks like in practice

A good example is our collaboration with Project Tech4Dev (T4D), a non-profit developer of open-source, social-sector SaaS solutions. We developed the POC for a data-warehousing platform that automates data ingestion, storage, and analysis of diverse NGO data streams – from paper surveys and digital forms to chat services and administrative logs. T4D transformed this into Dalgo, a low-cost product that now serves a dozen organizations (and is rapidly shipping new features).

In partnership with T4D, we also host user community events like “unconferences” and tech sprints. These allow our grantees and their peer organizations to share experiences, make feature requests, and hold us accountable. At a recent sprint, participants brainstormed ways to integrate an LLM-management platform, Langfuse, with existing T4D products like Glific, an open-source chatbot builder. They also gave us feedback on new product ideas, including an A/B test registry and the experiments automation platform. 

Another example is our partnership with Kenya-based Shamiri Institute, which increases access to mental healthcare for youth in underserved regions. In addition to providing Shamiri with grant funding, we led the co-development and deployment of Shamiri Digital Hub (SDH), a back-office platform to streamline data collection and operations for health programs operating in both schools and clinics. This software is now being used by other Kenyan nonprofits, and soon Shamiri expects to open source the platform to others. 

To cultivate an engineering culture at Shamiri that can sustain the SDH product, we assisted in vetting and hiring two key technology roles. We helped establish modern engineering practices and created dashboards for regular review of operational data stored in the SDH. Shamiri now features “charts of the week” at company-wide meetings as a way to foster data-driven decision making. They recently conducted an A/B test demonstrating the impact of a modest operational change on the treatment of youth depression and anxiety. The streamlined operations and data management enabled by SDH have helped Shamiri achieve rapid growth, reaching more than 100,000 additional youth this past school year.

Our role in these products isn’t always obvious: we work quietly in the background – ideating, prototyping, building, and transitioning. While our fingerprints are hard to see, we think we’re playing a critical role in bringing new tech solutions into reality.

A strategic vision

As with the fight against neglected tropical diseases, we think it’s possible for a small group of dedicated people to find creative solutions for missing markets. Through systemic and collaborative efforts, we can stimulate innovation. But we need many such efforts to modernize the tech stack for social services. We need donors who are willing to invest in highly scalable, cost-effective products rather than paying for one-off, custom-built solutions. We need tech companies willing to maintain products that will not generate billions in revenue (but that can be self-sustaining). Perhaps most of all, we need software designed for learning, rather than just monitoring – so that social service organizations can evolve and explore how to generate impact across diverse user groups. 

Our emerging product development strategy is rooted in openness and collaboration: we view our grantees as design partners, we leverage existing open-source and commercial solutions, and we foster partnerships with a variety of software firms. We are advising organizations to reduce their reliance on costly custom-built tech systems and to open source anything they have built (for reuse by others). In this way, we are upholding the public interest and reducing the waste of scarce donor dollars. 

While we are keen to advance this new paradigm for software development, we would value feedback from the community. We believe that this model could yield products and platforms that are financially sustainable and generalizable across diverse social service organizations. But have we put in place the right incentives? Are we right to push back against custom solutions? Is it possible to build an affordable, modern data and tech stack that supports a diversity of social services?