Learnings from a Pilot Collaboration with ADRA Canada
In 2024 and 2025, Fortell Impact partnered with ADRA Canada to pilot the use of artificial intelligence to improve monitoring and evaluation in humanitarian food assistance programming across four countries.
Background
ADRA Canada conducted a pilot study in partnership with Fortell Impact to evaluate the potential of artificial intelligence to streamline and strengthen its data collection, analysis, and reporting processes in food assistance programming. ADRA Canada's food assistance programs are built around a standardised Performance Measurement Framework (PMF). This framework tracks key indicators through baseline and endline assessments, which serve as the foundation for understanding the project's impact on the food security of recipients over the duration of the response, ensuring accountability and driving learning.
Adapting to diverse humanitarian contexts
Food assistance programmes span vastly different operating environments—from security-constrained regions like Syria and Yemen to remote areas of Somalia and Haiti—each requiring tailored approaches to data collection and analysis.
Syria and Yemen did not utilise avatars for data collection due to concerns from local authorities regarding the use of unfamiliar technologies by NGOs, particularly in sensitive political environments. Human enumerators were used instead, but AI was applied post-collection for analysis and reporting.
Localised avatars and automated reporting
The pilot placed particular attention on adapting both the accent and physical appearance of the avatars to reflect local populations—including developing a Somali-speaking avatar with a regionally appropriate accent and an avatar in Haitian Creole to ensure authentic engagement with participants.
Results by country
Combining AI efficiency with local expertise
AI analysis surfaced patterns and trends across large datasets quickly, while local staff contributed contextual knowledge that enriched interpretation. This combination—automated insight generation paired with on-the-ground expertise—enabled faster, more actionable reporting.
"Report generation time dropped significantly. The actual time varies by implementing partner M&E capacity, and is less important than the fact that it is available within 48 hours after the data is uploaded and is therefore more useful for real-time project management decisions."
"Participants responded positively to AI avatars, requiring little support from the accompanying enumerators, confirming their viability and potential to reduce costs and security risks in regions where in-person visits are logistically challenging."
Outcome
Report generation time dropped significantly, with analysis available within 48 hours after data upload—enabling more real-time reflection and mid-course corrections. Participants responded positively to AI avatars in Somalia and Haiti, requiring little support from accompanying enumerators, confirming their viability and potential to reduce costs and security risks in logistically challenging regions. The system also made it easier to track changes from baseline, helping project teams make attribution claims with confidence.
Decision Impact
Findings supported rapid decision-making and mid-course corrections under tight timelines.
Have a similar challenge?
We'd be happy to discuss how we might approach your specific context and requirements.