Applied AI in Humanitarian Operations: UN Case Studies
These are applied AI patterns from UN humanitarian operations: voice access for refugee services, AI-enabled feedback, data integrity, decision support, and controlled workflows. Each case starts with the operational constraint, because a model has no value if people cannot reach the service, staff cannot trust the output, or the organisation cannot stop it safely.
What humanitarian AI actually requires
Humanitarian environments are not standard enterprise settings. They involve fragmented data, multiple languages, limited infrastructure, high accountability requirements, and populations who are often vulnerable. AI in these settings must be designed with governance, safety, and operational usefulness as first principles — not afterthoughts.
What these cases show
Applied humanitarian AI works best when it removes an operational burden without transferring a consequential decision to a model. A voice service can make an appointment process reachable. A feedback classifier can help staff find patterns. A data-integrity workflow can identify records that need review. In each case, a person remains accountable for the outcome.
This page focuses on applied systems and delivery lessons. For the broader guide to selecting, governing, and assessing AI use cases in humanitarian organisations, read the humanitarian AI guide.
Where Shahzad has applied AI
Digital AAP
An AI-enabled accountability platform that strengthens two-way communication with affected populations through structured feedback intake, analysis, escalation, and management reporting.
AI-Enabled Voice Feedback System
A voice feedback solution that captures spoken messages, transcribes them, classifies them, and routes them for follow-up — improving accessibility for low-literacy users.
AI-Powered IVR Appointment System
UNHCR's first global IVR system for refugee appointment booking, supporting 700,000+ refugees and replicated across five countries.
AI for Policy and Decision Support
AI and data solutions that convert fragmented operational information into structured insight for senior management planning, prioritisation, and oversight.
Agentic AI for Sensitive Reporting
An agentic AI workflow designed to support confidential intake of sensitive protection-related reports through familiar channels such as WhatsApp, with privacy, safeguarding, and controlled routing.
AI-Enabled Knowledge Platforms
Knowledge management tools that use AI to organise, surface, and connect institutional knowledge for faster decision-making and operational continuity.
Key design principles
Privacy and data protection
All AI workflows must respect purpose limitation, lawful basis, access control, retention rules, and auditability.
Human-in-the-loop decision making
Human oversight is placed where judgment, rights, protection concerns, or material decisions are involved.
Multilingual accessibility
Systems must support multiple languages and low-literacy users to reach the most vulnerable populations.
Resilience in low-connectivity settings
AI solutions must function under infrastructure constraints common in humanitarian field operations.
Governance and controlled escalation
Clear ownership, documented use cases, risk review, and escalation paths are required before deployment.
Operational usefulness over novelty
AI is adopted only when it improves services, decisions, or accountability — not for innovation signalling.
What leaders should avoid
Deploying AI without governance, data protection review, or clear accountability. Treating AI as a standalone innovation exercise rather than an operational tool. Scaling pilots before validating them against real operational risks. Ignoring the human oversight requirements that sensitive humanitarian contexts demand.