A humanitarian chatbot needs somewhere real to send the difficult case

Andre Heller’s account begins with repetitive but urgent questions: where to register, which documents to bring, and how to find assistance. Answering those questions at scale can free caseworkers for problems that require investigation and intervention.

He describes Signpost’s move from tools that help staff draft responses toward bounded assistants using vetted local material. Related projects provide teacher support through WhatsApp and help newcomers navigate unfamiliar services. The intended service includes escalation, not simply a fluent answer.

The distinction becomes clear in a case involving a refugee’s missing food-assistance record. Information was useful because people could investigate the problem and reach the office capable of correcting it. A chatbot alone could not restore the entitlement.

This is a program director’s account, including selected examples and reported reach, rather than an independent impact evaluation. It nevertheless makes the underlying product requirement concrete: the answer must connect to a functioning service with someone responsible for the next step.