AI advisory · Humanitarian organisations
I work with NGO leadership teams who want to move forward on AI while staying in control: deciding where it serves your mission, setting your principles, writing the rules that govern its use, choosing the right tools, and training the teams who will rely on them. With one fixed point: every use must remain compatible with the protection of the people you serve.
In most organisations I come across, staff are already using AI tools: to translate, summarise a report, rework a proposal or code a questionnaire. Often on a personal account, and without any written rule.
The day someone raises a concern, two reflexes kick in. Banning it, and the use carries on from personal phones, out of sight. Borrowing a corporate policy, and you end up with a tidy document that says nothing about beneficiary data, consent in a crisis, or accountability.
A humanitarian organisation carries a constraint few others share: the people whose data it processes have no bargaining power. They do not choose the provider, they do not read the terms of use, and they often do not know that any processing is taking place. The Do No Harm principle applies to your digital tools as well, and it is up to leadership to say how.
Where AI can serve your mandate, where it has no place, and in what order to move. Work with the leadership team that starts from your programmes and your constraints, and ends with priorities you can defend before your board and your donors.
What your organisation stands by when it uses AI. These principles are built with you, from your mandate, the humanitarian principles and your commitments to communities. Everything else follows from them.
Which data may leave the organisation, which uses are allowed and under what conditions, which tools are approved, where human review is mandatory, and who decides when something unexpected comes up. A short text that makes decisions, written with the teams who will apply it.
Commercial tools, AI features already built into your software, sector-specific tools or an in-house solution: I compare the options against your data, your budget and your capacities, and recommend what fits each use.
A policy needs someone to keep it alive. I help you set out who decides, how a new tool is assessed before it is adopted, whether bought or built in-house, and how the rules are reviewed as practices change.
Sessions built on your real cases: what AI does well, what it does badly, and what your policy allows. The goal is practical: everyone should know what to do when faced with a tool, a sensitive file and a deadline.
Few organisations need all of this at once. A policy already on paper may lack governance; an AI project already under way may benefit from an outside view before it goes further. I build the assignment around what is most pressing, and around what your teams can absorb.
An AI policy flows from principles, and those principles belong to your organisation. I build them with you from what you already stand for: your mandate, the principles of humanity, neutrality, impartiality and independence, your commitments to accountability and protection. Once set, they also settle the cases the policy did not foresee.
The policy itself is short, and it decides. It answers six questions explicitly. When one is missing, that is exactly where a team gets stuck on the day it opens the document to find out what to do.
| Question | What the policy must say |
|---|---|
| Data | What may leave the organisation, and what never may. This is the rule a colleague will remember under pressure. |
| Uses | Described by role. Translating an internal memo and pre-selecting households for a distribution do not carry the same risk. |
| Tools | A short list of assessed tools, and a procedure to have others assessed. Without that procedure, the list is out of date within months. |
| Human review | Where it is mandatory, who does it, and what it commits them to. |
| Decision-making | A named person, and the route to follow for the case nobody anticipated. |
| Informing people | What communities and partners are told about the use of AI. This is the part most often forgotten. |
A rule that bans without offering a solution pushes people to work around it. So before writing the list of approved tools, I compare what is available to you: consumer commercial tools, the business offers of those same providers, AI features already built into the software you use, sector-specific tools, or an in-house solution hosted under your control.
Each option is assessed against the same criteria:
| Criterion | The question asked |
|---|---|
| Data | Which data the tool will process, where it is hosted, under which jurisdiction, and whether it is used to train a model. |
| Fit | What the tool does well for your roles, and what it does badly. |
| Cost | The real cost across the organisation, including licences, training and administration time. |
| Field conditions | Connectivity in your offices, the languages of your teams, the devices available. |
| Capacity | What your IT staff can run and maintain over time. |
| Lock-in | What you lose if the provider changes its terms, its prices or shuts down, and how you get your data back. |
The result is a reasoned recommendation, use by use. It often combines several options. If a solution I designed is part of the comparison, I tell you, and it is judged on the same criteria as the others.
The order matters as much as the content. A policy written before looking at actual practice describes an imaginary organisation, and ends up in a shared folder nobody opens.
The work rests on the reference texts, so that your choices hold up before a donor, an auditor or a partner:
To apply these frameworks day to day, I boil them down to five requirements, which I call TRUST. They can be used to assess any use of AI, a purchased tool as much as an in-house project.
The five requirements in detail, and how I apply them to my own tool, are set out on the TRUST page.
I have spent eight years in humanitarian organisations, across six countries: monitoring and evaluation in the Democratic Republic of Congo, Sudan and Ukraine, coordinating a programme portfolio in the Central African Republic, and today coordinating development and quality in Burundi. I know the trade-offs a country director makes, what donors expect, and what happens to a rule once it reaches the field.
Alongside that, I designed Opti', an AI-assisted monitoring and evaluation platform used by more than 800 professionals in over ten countries. I apply to it the rules I recommend: report figures are calculated outside the language model, every generated document is reviewed before release, and the most sensitive data can be processed by OptIA, a private AI that goes through no commercial provider. So I know what these principles demand, in time as well as in technical choices.
I also run webinars on AI ethics in humanitarian work, notably with the MEAL Afrique network.
Every assignment is tailored: its scope depends on the size of your organisation, how AI is used, and what is already in place.
Thirty minutes is enough to know where to start.
