Case study 5: Chatbot answer to beneficiaries
Scenario
Your organization pilots a beneficiary-facing chatbot for FAQ about registration hours and document requirements.
Approved FAQ source F1: “Registration at Site South is Tuesdays 09:00 to 14:00. Bring national ID or referral letter from Partner Gamma.”
Beneficiary question: “Can I register on Monday at Site North?”
Chatbot reply: “Yes. Registration at Site North is open Monday to Friday 08:00 to 17:00. Bring any photo ID.”
Your task (10 minutes)
- Compare the reply to F1.
- Who could be harmed?
- Should the chatbot answer stand without human review?
- What monitoring or stop rule would you recommend?
Model answer
| Element | F1 | Chatbot | Problem |
|---|---|---|---|
| Site | South only in F1 | North | Wrong location |
| Day | Tuesday | Monday-Friday | Wrong schedule |
| Documents | ID or Gamma referral | Any photo ID | Wrong requirements |
Harm: Exclusion (travel to wrong site), denial of service (wrong documents), loss of trust.
Decision: Red for auto-send. Human review required for any answer outside exact FAQ match. Stop pilot if wrong-site answers exceed agreed threshold (e.g., any confirmed wrong location).
Common errors
- Assuming the chatbot only uses the FAQ file without testing edge questions
- Measuring success by volume of chats rather than accuracy
- No escalation path to a human caseworker
Principle taught
Humanitarian chatbots: Beneficiaries may treat answers as authoritative. Wrong logistics information causes real harm. Test and monitor continuously.