The challenge

Country Strategic Plans (CSPs) define WFP’s priorities and actions in a specific country to address hunger and support food security.

Developing a CSP requires analysis of a wide range of information, including reports, policy documents and lessons learned. This analysis helps country offices identify gaps and make strategic decisions suited to their context.

WFP identifies the manual review of this information as time-intensive. Limited time and access to relevant insights can also constrain strategic planning.

The AI solution

WFP developed an AI-powered system designed to analyse complex documents and support strategic decision-making during plan formulation.

The system can analyse:

• Country Strategic Plans;
• evaluations;
• WFP strategies; and
• assessments.

It is designed to identify relevant outcomes and activities, answer questions across Country Strategic Plans and annual country reports, and generate strategic outcome sections using prompts and information from previous documents.

How the system supports planning

The system brings information from multiple documents into an AI-assisted workflow for strategic planning.

This can help staff locate relevant information without manually searching through every document each time a strategic question arises.

The system also supports the drafting of strategic outcome sections based on prompts and information from previous documents.

Development and reported potential

WFP states that the development of a Country Strategic Plan can currently take up to 18 months.

The AI solution is positioned as a way to significantly shorten this development process by consolidating information from multiple sources.

The first prototype generated multiple requests for expanded functionality. The minimum viable product was completed in March 2025 and presented during UN 2.0 Week.

WFP describes the system as having the potential to generate major efficiency gains and improve accuracy and alignment across strategic planning processes.

Why this matters for M&E

CSP formulation draws on many sources that are also important to Monitoring and Evaluation, including evaluations, assessments, previous plans and lessons learned.

AI-assisted document analysis could therefore change how evidence is located and brought into strategic planning.

For M&E professionals, the key question is not simply whether AI can analyse documents. It is whether easier access to evidence leads to better-informed strategic decisions.

EvalCommunity perspective

The following is an EvalCommunity interpretation and not a set of indicators reported by WFP for this project.

An M&E assessment could examine the entire chain rather than evaluating only the technical performance of the AI.

Potential evaluation dimensions

• Accuracy: Does the system accurately represent source information?
• Completeness: Does it identify important evidence or omit relevant information?
• Traceability: Can AI-generated claims be linked to their source documents?
• Context: Does the system preserve country-specific context?
• Human verification: How do users check and correct AI outputs?
• Efficiency: Does AI reduce the time required to find and analyse evidence?
• Evidence use: Does retrieved information actually influence strategic reasoning and decisions?
• Alignment: Does the system help users connect previous strategies, evaluations and assessments with proposed priorities?

Retrieving evidence is not the same as using evidence

An AI system may successfully locate relevant information without that information necessarily influencing a strategic decision.

This distinction is important for M&E. Evaluation should therefore move beyond asking whether the AI can find information and examine what happens after that information is retrieved.

This helps determine whether AI is simply making documents easier to search or whether it is contributing to evidence-informed strategic planning.

Evaluating AI-generated strategic content

Because the system can generate strategic outcome sections, the generated content itself becomes an important object of evaluation.

M&E professionals could examine whether:

• generated statements accurately represent the source evidence;
• important qualifications and limitations are preserved;
• contradictory findings remain visible;
• claims can be traced to their sources;
• country-specific context is retained; and
• human reviewers identify and correct errors.

These are proposed evaluation considerations, not reported findings about the WFP system.

What a stronger evaluation could examine

• Time: How long does evidence retrieval and review take?
• Accuracy: How accurately does the system retrieve and represent source information?
• Coverage: What proportion of relevant evidence is identified?
• Traceability: Can generated content be linked to its sources?
• Human review: How often are AI outputs corrected or rejected?
• Evidence use: Does retrieved evidence influence strategic decisions?
• Quality: Is there measurable change in the evidence base used during planning?
• Efficiency: Does the system reduce the resources required for documentary analysis?

What the current evidence does not establish

The WFP project description documents the system’s intended capabilities, prototype development, MVP completion and potential benefits.

It does not provide a full impact evaluation demonstrating:

• a quantified reduction in CSP development time;
• a measured improvement in strategic-plan quality;
• a measured improvement in accuracy or alignment;
• how frequently users correct AI-generated content;
• the completeness of AI-generated evidence synthesis; or
• downstream effects on programme or beneficiary outcomes.

These should therefore be treated as areas for further evaluation rather than conclusions that can be inferred from the project description.

Source: World Food Programme Innovation, “AI for Country Strategic Planning.” Project capabilities, development milestones and potential benefits are reported by WFP. The M&E interpretation and proposed evaluation questions are developed by EvalCommunity and are not presented as WFP findings.