Evaluation for programmes where the evidence is hard to get.
An independent practice. Theory-based, mixed-methods evaluation for UN agencies, INGOs and bilateral donors, delivered in access-constrained and conflict-affected settings.
Discuss an evaluation →Senior evaluator and lead report author on a UNICEF country programme evaluation awarded the highest possible rating under UNICEF's Global Evaluation Reports Oversight System, and published on its global evaluation database.
Evaluations for Mercy Corps, UNICEF, IOM, UNMAS, ACTED, Expertise France, the British Council, Right to Play and the Danish Refugee Council, against donor frameworks including FCDO, AFD, BHA, the US Department of State, ECHO, UNTFHS, the EU, KOICA and Global Affairs Canada.
A mid-term evaluation of a USD 96 million cash consortium reaching 1.9 million people across all 18 states of Sudan. A multi-stage cluster endline of 2,617 households in conflict-affected Borno State. Fieldwork in Kenya, Zimbabwe, Malawi, Tanzania, Ethiopia and Jordan; remote delivery in Sudan, Somalia, Syria, Lebanon, Nigeria and the Philippines.
Theory of Change reconstruction, contribution analysis, process tracing and outcome harvesting, designed against OECD-DAC criteria and UNEG ethical standards, alongside reproducible quantitative pipelines and standard humanitarian indicators.
Evaluation design
Inception, evaluation matrices, sampling strategy and instrument suites.
Quantitative and qualitative analysis
Survey pipelines and indicator analysis; thematic coding with traceable evidence chains.
Reporting and quality assurance
Report drafting, citation and consistency checks, validation workshops and briefs.
MEL tooling and fieldwork support
Indicator dashboards, transcript processing, real-time fieldwork tracking.
Underneath the analysis. Never in front of it.
AI does the mechanical work: document review at scale, first-pass coding, citation and consistency checks. Interpretation, ethical judgement and accountability for every claim stay with the evaluator. Each method has a specific way it breaks when a model is applied to it carelessly, and every deliverable discloses how these tools were used.
How it works →