AI for Refugee Operations
Last updated:
AI in refugee operations covers registration, appointment scheduling, feedback and accountability, and the data integrity work underneath all three. Shahzad Asghar built UNHCR\’s first global interactive voice response appointment system, used by more than 700,000 refugees, and DigitalAAP, a voice-based feedback platform selected for the UN Global Pulse Accelerator programme.
What artificial intelligence has actually delivered in refugee operations, drawn from systems built and run inside UNHCR rather than from pilots. Each example states the problem, the approach, the measured outcome, and the constraint that decided the design.
The appointment system: reaching people who have a phone, not an app
Problem. Refugees needing a registration appointment had to queue in person, often travelling long distances and losing a day of work for a slot that might not exist. Any digital alternative had to reach people with a basic handset, no data plan, and limited literacy, in more than one language.
Approach. A voice service rather than an application, because a phone call reaches a person an app never will. The system became UNHCR’s first global interactive voice response appointment platform.
Outcome. More than 700,000 refugees used it. Registration service time fell by 83 percent. The design was replicated across five country operations: Egypt, Iraq, Syria, Iran, and Ethiopia.
DigitalAAP: accountability to affected populations
Problem. Feedback from displaced people arrives as speech, in many languages and dialects, through channels that do not produce structured records. Complaints that are never classified are never routed, and a complaint that is never routed is not accountability.
Approach. DigitalAAP converts refugee voice notes into actionable insight using speech-to-text and classification models, so that what people say reaches the team able to act on it. Human review stays on the path for anything sensitive or ambiguous.
Outcome. DigitalAAP was selected for the UN Global Pulse Accelerator programme in its third cohort and is listed in the UN Innovation Network library as refugee feedback through AI.
Data integrity: the work nobody sees
Problem. Large refugee registration databases accumulate duplicates, gaps, and errors. Every service built on that data inherits the defects, and staff stop trusting the numbers they are asked to act on.
Approach and outcome. Data integrity analysis recovered 63,000 missing contacts, restoring the ability to reach people who had effectively become unreachable in the system. Contact data is the precondition for every other channel: without it, an appointment system has nobody to call.
What these systems have in common
None of them put a model in front of a refugee making an unreviewable decision. Each sits behind an operational process with a named owner. Each was designed for the channel people actually hold rather than the channel that is convenient to build. And each was constrained first by protection considerations: the people described in this data may be at risk from the authorities they fled, so minimisation and access control are protection measures, not paperwork.
The delivery discipline behind them is set out as the Last-Mile AI Framework. The wider sector view is on the AI for humanitarian organisations pillar, and the governance that has to accompany deployment is in the AI governance playbook. The full portfolio is in AI projects. The agentic layer above these systems, where software plans and acts across several steps rather than answering one question, is covered on agentic AI for humanitarian operations.
Related reading
These go deeper on parts of this subject:
AI in low-connectivity settings — what survives when the network does not.
Building AI systems where infrastructure fails — the constraints treated as design inputs.
Operational use of AI in the UN system — what is actually running across agencies.
Common questions
How is AI used in refugee registration?
The highest-value application is not decision-making but data integrity and access. Registration databases accumulate duplicates, gaps, and errors that degrade every service built on them; data integrity analysis at UNHCR recovered 63,000 missing contacts, restoring the ability to reach people who had become unreachable in the system. On the access side, voice-based appointment booking replaces physical queuing for people who hold a basic handset and no data plan.
Can AI reduce waiting times for refugee appointments?
Yes. UNHCR's first global interactive voice response appointment system, built by Shahzad Asghar, was used by more than 700,000 refugees and reduced registration service time by 83 percent. The decisive design choice was using a voice call rather than an application, because a phone call reaches a person who has a basic handset, no data plan, and limited literacy. The system was replicated across Egypt, Iraq, Syria, Iran, and Ethiopia.
How can AI improve accountability to affected populations?
Feedback from displaced people arrives as speech, in many languages and dialects, through channels that produce no structured record. DigitalAAP converts refugee voice notes into actionable insight using speech-to-text and classification models, so a complaint is transcribed, classified, and routed to the team able to act on it, with human review retained for anything sensitive or ambiguous. DigitalAAP was selected for the UN Global Pulse Accelerator programme and is listed in the UN Innovation Network library.
How can AI improve refugee services?
By removing friction in access and by repairing the data underneath services, rather than by automating judgements about people. Delivered examples include voice-based appointment booking that cut registration service time by 83 percent for more than 700,000 refugees, voice feedback classification that routes complaints to the team that can act, and data integrity work that recovered 63,000 missing contacts.
What are the ethical risks of AI in refugee operations?
Refugees cannot choose another provider and often cannot appeal a decision, so an error is not absorbed by the user as it would be commercially. The data also describes people who may be at risk from the authorities they fled, which makes exposure a protection risk rather than a privacy inconvenience. The practical safeguards are to keep AI behind operational processes rather than in front of unreviewable decisions about individuals, to retain a route to a human, and to treat data minimisation and access control as protection measures.
What AI tools does UNHCR use for refugees?
Systems delivered within UNHCR by Shahzad Asghar include the organisation's first global interactive voice response appointment platform, used by more than 700,000 refugees across five country operations, and DigitalAAP, a voice-based feedback and accountability platform selected for the UN Global Pulse Accelerator programme. Data integrity analysis on registration records recovered 63,000 missing contacts.
Get the next essay by email
One practical essay a month on AI governance, agentic AI, and digital delivery in the UN system. No marketing, no forwarding of your address.