Donors and foundations
Compare evidence across programmes and funding options, with assumptions and uncertainty available for review.
The client problem
The evidence exists. The connections take work.
A programme decision rarely rests on one report. Research, evaluations, operational records, and local expertise may describe different contexts or point in different directions. Teams need a way to compare that material while keeping its sources, gaps, and limitations visible.
Who the service is for
Development organisations
Bring research, project evaluations, and implementation experience together around a practical programme question.
Evaluation and learning teams
Organise findings across sources and identify where further evidence or specialist interpretation is needed.
Inputs and source material
Start with the evidence behind your question.
Agree the scope and permissions first. The material can combine published research with the records and contextual knowledge your team is authorised to use.
- Research and publications
- Academic articles, evidence reviews, policy papers, and relevant public research.
- Evaluations and learning
- Project evaluations, lessons learned, programme reports, and institutional knowledge.
- Operational data
- Monitoring records, indicators, implementation documents, and supporting datasets.
- Local knowledge
- Field observations, stakeholder perspectives, and contextual expertise shared with appropriate permission.
The workflow
From a defined question to a checked synthesis.
The workflow connects source preparation, analysis, verification, and human review. The scope and review responsibilities are agreed for each engagement.
Frame the question
Agree the decision, intended users, scope, and criteria for useful evidence before choosing a model or generating an answer.
Agree the source material
Review available documents and data, record their context and dates, and agree access and publication permissions.
Structure and compare
Organise relevant passages around themes, interventions, outcomes, and context. Compare findings without hiding disagreement or missing evidence.
Check the source trail
Review findings against the original documents and passages. Separate what sources report from interpretation, and record caveats and evidence gaps.
Review before use
Subject specialists and the responsible team check the synthesis, revise the interpretation, and decide what is suitable for the intended use.
Expected outputs
Outputs your team can inspect and use.
Select the deliverables that fit the question and intended use. A brief, a matrix, or an evidence map should carry its supporting sources and limitations with it.
Source-linked synthesis
A written brief that brings findings together with the references needed to inspect them.
Comparative evidence matrix
A structured comparison of themes, findings, contexts, and areas of disagreement.
Evidence map
An organised view of the material considered, its coverage, and the questions it leaves open.
Review and limitations record
Documented assumptions, evidence gaps, reviewer decisions, and constraints on interpretation.
Source-verification controls
A finding should lead back to its evidence.
Verification is part of the work, not a label applied to the generated text. Reviewers need to inspect what a source says and whether it supports the interpretation.
Keep the original source in view
Retain document and passage references so a reviewer can check whether a finding is supported in its original context.
Separate evidence from interpretation
Distinguish a reported result, an analytical inference, and an unanswered question. A citation alone does not establish that a claim is correct.
Make uncertainty explicit
Record differences in source quality, dates, methods, languages, and settings rather than presenting every source as equally applicable.
Human-review points
People remain responsible for interpretation and use.
Agree who reviews the evidence, who brings subject and local expertise, and who is accountable for the final decision.
Before analysis
The commissioning team agrees the question, source boundaries, permissions, and review criteria.
During synthesis
Subject and context specialists check extracted evidence, challenge interpretations, and flag missing perspectives.
Before a decision
The responsible decision-maker reviews the findings and limitations and approves how the output will be used.
An example from the existing portfolio
Education evidence in Nepal
The existing Mercy Corps Nepal case study describes bringing academic articles, project evaluations, and local data together to inform education programme planning. It provides context for how evidence synthesis can support a specific programme question.
Connected capabilities
Evidence synthesis within OutcomeLine.
OutcomeLine connects evidence preparation, model evaluation, responsible-AI controls, workflow delivery, and monitoring. DELLM provides structured, source-linked evidence, while AidInsight is the workspace for workflows and review.
Limitations
Know what the synthesis can and cannot tell you.
A useful output makes its boundaries visible. Scope, evidence quality, and expert judgement determine how findings should be interpreted.
- AI can omit material, misread a passage, or produce an unsupported interpretation. Findings need source-level checks and specialist review.
- The synthesis reflects the material available and agreed for the task. Missing, outdated, biased, or poorly documented sources limit what can be concluded.
- Evidence from one setting may not transfer to another. Local knowledge and methodological judgement are needed to assess relevance.
- A synthesis does not establish causality, replace a formal evaluation, or transfer accountability for a consequential decision to an AI system.
Related reading
Explore the work and the methods behind it.
Case studies
Mercy Corps Nepal
Education Evidence AnalysisFCDO / Sri Lanka
Trilingual AI for Economic Policymaking
Publications
Land Use Policy
Integrating human-centered AI for land use policyStanford Social Innovation Review
When USAID Shut Down, Its Lessons Nearly Vanished. AI Helped Recover Them
Start with one evidence question.
Tell us about the decision, the available material, and the people who need to review the result. We can discuss a useful scope and a practical next step.
Discuss your evidence needs