AI module · complements the 12 weeks · about 4 hours

Leading a humanitarian organisation with artificial intelligence

Humanitarian Mini MBA

Module objectives
  • Distinguish the leadership tasks where generative AI brings a real gain from those where it creates risk.
  • Write and enforce an AI use policy suited to a humanitarian organisation.
  • Steer AI adoption with indicators, a budget and clear data protection rules.

This module complements the 12 weeks of the programme. It adds no exam: certificates and the diploma are earned as before. Tools are described as known in November 2026; they change quickly.

1What generative AI does, and what it does not do

Generative AI assistants (Claude, Gemini, ChatGPT, Copilot and others) produce text, tables or code from a written request. They predict a plausible answer: they do not "know" whether it is true. Three limits matter for a director:

  • Errors stated with confidence ("hallucinations"): an invented figure, reference or clause, presented in the same tone as accurate information.
  • Bias: models know heavily documented, English-language contexts better than the realities of a rural district or a little-written language.
  • Confidentiality: what you paste into a tool may be stored, reviewed or used depending on the terms of service and the type of account.

Leadership rule: AI prepares, people check, decide and sign. A decision that commits the organisation (recruitment, disciplinary action, contract award, choice of an intervention area) is never delegated to a tool.

2Useful use cases for leadership

The clearest gains, provided you work on non-sensitive or anonymised documents:

  • Summarising: an 80-page evaluation report, a donor's country strategy or cluster minutes, then checking the key passages in the original text.
  • Preparing: listing the likely questions from a board, a donor or a local authority before a negotiation.
  • Drafting: a first draft of a strategy note, a message to the team or an internal communication plan, which you rewrite with your knowledge of the field.
  • Working translation: understanding a document in a language you do not read; for an official document or one intended for communities, review by a speaker remains mandatory.
  • Scenario analysis: testing a strategic plan against unfavourable assumptions (a 30% drop in funding, loss of access to an area).

Measure the gain: time saved per task, number of errors found on review. Without measurement, it is easy to overestimate what the tool contributes.

3An AI policy for the organisation

Without a written rule, everyone uses whatever tool they like with their personal account. A short, workable policy covers:

  • Authorised tools and the type of account: preferably professional accounts whose contract excludes the use of your data to train models and specifies the retention period.
  • Data classification: public (may be used), internal (only with a professional account), confidential and sensitive (never: personal data of affected people, protection, health or PSEA case files, security information, named salaries).
  • Prohibited uses: individual decisions about people, candidate assessments delegated to AI, misleading content, generated images or voices presented as real.
  • Transparency: disclose when a document was largely drafted with AI, if the donor or recipient asks for it.
  • A focal point (often information management or data protection), basic training and a channel for reporting incidents.

4Affected people's data, ethics and humanitarian principles

Reference frameworks exist: the Inter-Agency Standing Committee (IASC) Operational Guidance on Data Responsibility (updated 2023), the Handbook on Data Protection in Humanitarian Action by the ICRC and the Brussels Privacy Hub, the Principles for the Ethical Use of Artificial Intelligence in the United Nations System (2022), and national data protection laws (GDPR in Europe, African, Asian or Latin American laws).

Points of vigilance for a director:

  • No identifying data in a consumer tool. Be careful: a village name, an age and a disability can be enough to recognise a person in a small community.
  • Do no harm: a leaked list of vulnerable families can expose those families to reprisals.
  • Impartiality: a prioritisation tool trained on biased data can exclude groups that are already marginalised.
  • Accountability: communities have the right to know how their information is used and to complain.

5Governing adoption: pilot, costs and skills

Serious adoption proceeds in stages:

  1. Inventory: who already uses what, for which tasks, with which data? "Hidden" uses are common.
  2. A three-month pilot on two or three non-sensitive tasks (report summaries, first drafts of minutes), with indicators set in advance.
  3. A decision based on the results: scale up, adjust or stop.

The real cost includes subscriptions, training time, review time and account management. AI also has an energy footprint: favour uses that replace real work rather than systematic content generation. Finally, AI changes jobs: plan training for national teams so that the gain benefits all staff, not only headquarters.

Exercise: a one-page AI policy

Situation. You lead an NGO with 120 staff in three countries. An internal survey shows that 70 staff use an AI assistant, 52 of them with a free personal account. A protection officer pasted a table of 40 individual cases into a tool to "save time".

Questions. 1) What share of users work with a personal account? 2) What are your three immediate measures? 3) What headings do you put in the one-page policy?

Answer key. 1) 52 / 70 ≈ 74%. 2) Immediate measures: handle the protection incident (assess what was shared, follow the data breach procedure, inform the data protection focal point, assess the risk for the people concerned); issue a written reminder that no personal data may go into an unapproved tool; open professional accounts for authorised uses. 3) Headings: authorised tools, data classification (public, internal, confidential, sensitive), prohibited uses, transparency, focal point and reporting, mandatory training, annual review.

Key points

  • AI prepares; people check, decide and sign, especially for decisions that affect individuals.
  • The best gains: summarising, preparing, first drafts and working translation, on non-sensitive data.
  • A short AI policy: authorised tools and accounts, data classification, prohibited uses, focal point, training.
  • Affected people's data: never in a consumer tool; refer to the IASC, the ICRC and the applicable law.
  • Adopt through a measured pilot, counting the real cost and the training of national teams.

Self-assessment quiz

6 self-assessment questions, corrected immediately. They do not count towards the exams.

1Why can an AI assistant give a reference to a report that does not exist?
2Which decision can be entrusted to an AI assistant without human control?
3In an AI policy, individual protection case files belong to the category:
4Out of 70 users, 52 have a free personal account. Share:
5Which reference document deals with data responsibility in humanitarian action?
6First step of a serious AI adoption:
Going further

The Mini MBA AI and Humanitarian Action (Claude and Gemini) covers these topics in depth over 12 weeks.

See the AI Mini MBA