Who Connects Global Aid? The Hidden Geometry of 10 Million Transactions

The global aid system functions as a complex and evolving ecosystem; yet widespread understanding of its structure remains largely limited to aggregate volume flows. Here we map the network topology of global aid using a dataset of unprecedented scale: over 10 million transaction records connecting 2,456 publishing organisations across 230 countries between 1967 and 2025. We apply bipartite projection and dimensionality reduction to reveal the geometry of the system and unveil hidden patterns. This exposes distinct functional clusters that are otherwise sparsely connected. We find that while governments and multilateral agencies provide the primary resources, a small set of knowledge brokers provide the critical connectivity. Universities and research foundations specifically act as essential bridges between disparate islands of implementers and funders. We identify a core solar system of 25 central actors who drive this connectivity including unanticipated brokers like J-PAL and the Hewlett Foundation. These findings demonstrate that influence in the aid ecosystem flows through structural connectivity as much as financial volume. Our results provide a new framework for donors to identify strategic partners that accelerate coordination and evidence diffusion across the global network.

💡 Research Summary

The paper presents the first large‑scale topological mapping of the global aid ecosystem using over ten million transaction records from the International Aid Transparency Initiative (IATI) spanning 1967‑2025. These records involve 2,456 distinct publishing organisations—governments, multilateral agencies, NGOs, private firms, universities and foundations—operating across 230 countries.

Data and Network Construction
The authors model the aid system as a bipartite graph where one set of nodes represents funders (providers) and the other set represents recipients (implementers). Each financial transaction creates an edge linking a provider to a recipient. This bipartite structure captures the full diversity of funding flows, including grants, loans and equity investments.

Embedding and Dimensionality Reduction
To capture the structural role of each organisation, the bipartite graph is fed into node2vec, a random‑walk‑based embedding algorithm that learns high‑dimensional vectors preserving both local neighbourhoods and broader network topology. The resulting vectors (typically 128‑256 dimensions) are then reduced to two dimensions using Uniform Manifold Approximation and Projection (UMAP). UMAP preserves manifold structure while allowing a clear visual separation of functional groups.

Hidden Geometry
The UMAP projection reveals two orthogonal axes that dominate the aid landscape:

• Horizontal axis – Humanitarian ↔ Development: organisations involved in short‑term crisis response (e.g., OCHA, Médecins Sans Frontières) cluster on the left, while long‑term development financiers (World Bank, USAID) occupy the right.
• Vertical axis – Funders ↔ Implementers: donors sit at the top, implementing NGOs, contractors and field agencies at the bottom.

These axes generate four quadrants (Humanitarian Funders, Humanitarian Implementers, Development Funders, Development Implementers). The space between quadrants is sparsely populated, indicating structural “holes” that impede the flow of knowledge and resources between sectors.

Centrality and the “Solar System”
The bipartite graph is projected into a one‑mode collaboration network to compute HITS Hub Scores, which identify organisations that connect to many distinct partners. Ranking by Hub Score and visualising the top 100 nodes produces a “solar system” diagram with concentric rings reflecting the power‑law distribution of centrality. The innermost ring (top 25) contains expected financial giants (US Department of State, European Commission, World Bank) and a distinct class of knowledge brokers: universities (e.g., J‑PAL at MIT) and foundations (e.g., William & Flora Hewlett Foundation).

Despite handling far fewer deals than large bilateral agencies, these brokers exhibit disproportionately high betweenness centrality, sitting on the shortest paths that link otherwise disconnected humanitarian and development clusters. For example, J‑PAL’s network position enables rapid diffusion of randomized‑control‑trial evidence across sectors, while the Hewlett Foundation’s portfolio connects research organisations (Population Council) with major implementation partners (UN Foundation, Oxfam America).

Policy Implications
The authors argue that influence in the aid ecosystem is not solely a function of budget size. Structural fragmentation—particularly the humanitarian‑development divide—creates coordination risks, duplication of effort, and hampers scaling of proven interventions. By identifying the small set of high‑centrality brokers, donors can design “strategic partnership” policies that:

1. Prioritise funding to organisations that act as bridges, thereby reducing structural holes.
2. Encourage joint programmes between humanitarian and development actors facilitated by knowledge brokers.
3. Use centrality metrics alongside traditional volume‑based assessments to allocate resources more efficiently.

Conclusion
Through a combination of massive transaction data, node2vec embeddings, UMAP visualisation, and hub‑score centrality analysis, the study uncovers a hidden geometry of global aid: a fragmented but bridgeable network where universities and foundations play a pivotal connective role. This new framework shifts the evaluation of aid effectiveness from sheer financial volume to structural connectivity, offering a concrete roadmap for donors seeking to accelerate coordination, evidence diffusion, and progress toward the Sustainable Development Goals.