Unlocking livestock investment insights from global open data

We're leveraging machine learning to understand the landscape of investments into livestock development.

Overview

Through our partnership with the University of Edinburgh's Data Science Unit (DSU), we are addressing a critical gap in global development data: understanding the landscape of investments into livestock development. While significant funding flows into livestock systems worldwide, these investments are often obscured within broad agricultural categories in international datasets. By leveraging machine learning to mine the International Aid Transparency Initiative (IATI) datastore, we aim to surface hidden livestock projects, extract valuable lessons from historical investments, and create a searchable evidence base for policymakers and funders.

The Challenge

Global development funding is vast, with the IATI datastore alone containing over 900,000 reported projects. However, livestock-specific interventions are frequently misclassified under generic "agriculture" or "rural development" tags. For major stakeholders like the Gates Foundation and other key funders, this lack of granularity makes it difficult to track where money is going, identify overlapping efforts in specific geographies, and learn from past successes or failures. Manually reviewing hundreds of thousands of project descriptions and associated documents to find relevant livestock data is impossible at scale, leading to fragmented knowledge and potential duplication of effort.

Our Approach: Machine Learning for Project Identification

We are deploying advanced machine learning models to automate the identification and analysis of livestock-related investments. Our methodology follows a two-stage process:

  1. Automated Identification: We train machine learning algorithms to scan IATI metadata and project descriptions, flagging activities that involve livestock systems even when they lack specific tags. This allows us to cast a much wider net than traditional keyword searches, capturing projects that would otherwise remain invisible in the dataset.
  2. Evidence Extraction and Synthesis: Once relevant projects are identified, the system mines associated grey literature—primarily donor investment reports and evaluation documents—to extract specific outcomes. The goal is to collate a robust body of evidence highlighting "what worked" and "what did not" across different contexts. These mined insights are then structured into a queryable database, allowing SEBI-L researchers and partners to ask specific questions about historical performance, funding trends, and intervention efficacy.

Throughout this process, SEBI-L livestock specialists work closely with data scientists to validate model outputs against strict acceptance criteria, ensuring that the identified projects and extracted insights are accurate and contextually relevant.

Strategic Value and Outcomes

This activity transforms opaque, unstructured data into actionable intelligence. By systematically surfacing hidden livestock investments, we provide a clearer, more comprehensive picture of the global funding landscape. This enhances transparency and enables funders to make evidence-based decisions, avoiding duplication and targeting resources where they are most needed. Furthermore, by aggregating learnings from historical projects, we create a living knowledge resource that helps the broader development community understand the drivers of success in livestock interventions. Ultimately, this collaboration demonstrates how data innovation can turn vast, underutilized datasets into powerful tools for improving agricultural development outcomes.

Building on previous efforts

In 2022, SEBI-Livestock created the Livestock Project Portal, which pulled open access IATI data from, presenting it in an easy-to-navigate visual dashboard. 

The dashboard offered funders, project implementers and national governments easy access to key information on livestock projects: where they are, what they are about, and what data they possess. The goal was to support improved coordination and efficiency in livestock investments and interventions.

The first dashboard was primarily a manual effort, and the potential for harnessing large amounts of open access development data. Feedback from the Livestock Data for Decisions (LD4D) community was positive and showed that there was demand for this type data tool. 

Learn about this earlier activity

Next Steps

Current efforts focus on refining the machine learning models to maximize recall and precision in identifying livestock projects within the IATI datastore. Following model validation, the team will proceed with large-scale data extraction from linked investment documents, which are scattered across many repositories. In part, the success of the tool will rely on the number of reports found. The final output will be an integrated, queryable evidence platform accessible to SEBI-L partners, providing unprecedented visibility into global livestock investment trends and outcomes. 

More information

Updates