AI powered Data Synthesis Framework for Evaluations

UNICEF faced challenges summarizing information related to the same evaluation questions when data came from different sources and in inconsistent formats. To help UNICEF make better use of this wealth of information, rowsquared built a data synthesis framework that powers an interactive HTML report with dynamic visualisations. As a result, evaluators now have a comprehensive overview of programme performance and contexts across countries.

We collaborated with seven country offices in Latin America and the Caribbean to analyse data from both UNICEF’s internal databases and external sources. rowsquared developed a data pipeline that integrated structured and unstructured data. We applied AI techniques, including classification, summarisation, and synthesis, for processing unstructured content. We linked quantitative and qualitative sources through shared metadata, allowing us to combine statistical trends with contextual understanding across countries and programmes.

To provide an accessible interface to stakeholders, we developed an interactive HTML report with dynamic visualisations and analytical text, using the Quarto framework. Quarto was chosen for its ability to integrate narrative, code, and outputs within a single document. The reports were developed iteratively, incorporating client feedback to refine the analysis and improve usability.

As a result, existing data can now be more easily and effectively used. UNICEF evaluators can quickly identify focus areas, understand programme context, and interpret evaluation results from a diverse set of countries. The established methodology is reproducible for future evaluations.