AI-Assisted Beneficiary Deduplication in Humanitarian Assistance (WFP)
EvalCommunity Academy · AI Case Study
AI-Assisted Beneficiary Deduplication in Humanitarian Assistance: The WFP Enterprise Deduplication Service (EDS) Case Study
How the World Food Programme is using artificial intelligence to identify potential duplicate beneficiary records and support the management of food assistance.
Case at a glance:
Organisation: World Food Programme (WFP)
System: Enterprise Deduplication Service (EDS)
AI application: AI-assisted beneficiary deduplication and data-quality support
Context and Source: Humanitarian food assistance – Every meal counts: How WFP is using AI to reach more people, faster | World Food Programme
1. What is the World Food Programme?
The World Food Programme (WFP) is a United Nations organisation that provides humanitarian assistance and works on food security and related needs in crisis-affected settings.
Its humanitarian operations require the management of information about people receiving assistance. In difficult operating environments, beneficiary information can become complex because populations may move, records may be incomplete, and information may be recorded differently across systems.
2. What problem is WFP addressing?
WFP explains that duplicate beneficiary records can occur in humanitarian settings for several reasons, including displacement, incomplete records, differences in spelling or language, and systems that are not synchronised.
Duplicate records can create operational problems because the same person may appear more than once in beneficiary information. WFP is using its Enterprise Deduplication Service to help identify possible duplicate records more efficiently.
3. What is WFP’s Enterprise Deduplication Service?
WFP’s Enterprise Deduplication Service (EDS) uses artificial intelligence and photo-based technology to analyse large volumes of information and identify possible duplicate beneficiary records.
According to WFP, the system analyses information including names and photographs. WFP also describes the possibility of using additional biometric information, such as fingerprints, in the future.
When the system identifies records that may belong to the same person, it flags them for staff review. A staff member then reviews the information and makes the final decision.
The AI identifies potential duplicates; staff make the final decision.
WFP explicitly states that the tool does not decide who receives assistance. Its role is to support the identification and review of potential duplicate records.
4. How does the AI-assisted workflow work?
Based on WFP’s description, the process can be represented as:
Beneficiary Data → AI-Assisted Deduplication → Potential Duplicate Flagged → Staff Review → Final Decision
WFP also reports that the system can identify data-quality issues, allowing staff to address problems in beneficiary information.
5. What results does WFP report?
WFP reports that a pilot of EDS in Mali in 2025 saved more than US$431,000 over six months by reducing duplicated assistance.
WFP projects that the system could save at least US$4.7 million in 2026 as it is scaled globally.
WFP states that, using an estimated cost of approximately US$0.70 per meal, the projected US$4.7 million saving would be equivalent to approximately 6.7 million additional meals.
These figures should be distinguished. The US$431,000 figure is a reported saving associated with the 2025 Mali pilot, while the US$4.7 million figure is a projection for 2026.
WFP also reports that tasks that previously took weeks can now be completed in hours.
According to WFP, the system can also be up to 50 percent cheaper than other biometric solutions, partly because it uses open-source AI models without licensing fees.
6. Data protection and safeguards
WFP identifies privacy and data protection as important considerations in the use of the system.
WFP reports that EDS has been reviewed by WFP experts and through external audits to assess compliance with relevant data protection and privacy requirements.
WFP also states that when the system is introduced into new contexts, privacy and data-protection assessments are conducted.
WFP reports that its face-recognition technology can work without requiring people to remove culturally important items such as veils or turbans.
7. Where is WFP using EDS?
WFP reports that the system has been deployed in Mali and piloted in:
- Afghanistan
- Burkina Faso
- Cameroon
- Mozambique
- Niger
- Somalia
- Uganda
8. What does this case show about AI in humanitarian operations?
The WFP case provides a documented example of AI being incorporated into an operational humanitarian process.
In this case, AI is being used to help process and compare beneficiary information and identify potential duplicate records. The resulting flags are then reviewed by staff.
The case therefore illustrates an AI-assisted workflow in which technology supports an operational task while human staff remain responsible for reviewing potential duplicates and making final decisions.
9. What can M&E professionals learn from this case?
EvalCommunity perspective: The following section is an M&E-oriented interpretation of the WFP case. It is not presented as WFP’s own evaluation framework or methodology.
For M&E professionals examining AI-assisted systems, this case provides an opportunity to consider the performance of the wider process rather than looking only at the AI technology itself.
An evaluator could examine questions such as:
- How accurately does the system identify potential duplicate records?
- How often are people incorrectly flagged as duplicates?
- How often are genuine duplicates missed?
- How are AI-generated flags reviewed by staff?
- What happens when staff disagree with an AI-supported result?
- Does the system reduce the time required for the process?
- Does it improve the quality and consistency of beneficiary information?
- Are errors or limitations concentrated in particular groups, locations or data conditions?
- What effects does the process have on assistance delivery?
- Are privacy and data-protection safeguards functioning as intended?
10. From AI performance to programme effects
Measuring the technical performance of an AI system is only one possible part of an evaluation.
An M&E assessment could examine the broader chain:
Data → AI-Assisted Identification → Human Review → Decision → Assistance → Programme Results
Evidence about processing speed is different from evidence about duplicate-detection accuracy. Both are different from evidence about effects on assistance received by programme participants.
Keeping these levels of evidence separate helps prevent an operational improvement from being interpreted automatically as a programme-level impact.
11. What does the published WFP case not establish?
WFP’s published account provides information about the system, its use, reported savings, projected savings, operational changes, locations of deployment and data-protection measures.
The published account does not provide a complete evaluation of false-positive and false-negative rates, differential error rates across population groups, or causal effects on beneficiary outcomes.
These would require additional evidence and evaluation. They should therefore be treated as questions for further investigation, rather than findings established by the WFP case.
12. Key takeaway
WFP’s Enterprise Deduplication Service provides a documented example of AI being incorporated into a humanitarian operational workflow.
The system is designed to help identify possible duplicate beneficiary records and data-quality issues. Human staff remain responsible for reviewing flagged records and making final decisions.
From an M&E perspective, the case can be used to explore how AI-assisted processes should be assessed across the wider chain from data and AI-supported analysis to human review, operational decisions and programme results.
The important evaluation question is not only whether the AI works, but what happens when the AI becomes part of a real operational process.
Explore AI evaluation resources
EvalCommunity Academy provides practical resources for professionals interested in evaluating and using AI in monitoring, evaluation and development work.
Explore the AI Tool Evaluation Protocol for M&E →
Related EvalCommunity resources
