For an NGO, association or foundation, artificial intelligence is neither a passing technology trend nor a simple shortcut to greater productivity. It is entering activities that directly affect the mission: producing content, analysing information, handling sensitive data, communicating with donors and documenting complex human situations.

The relevant question is therefore not how many tasks can be automated, but which activities can be improved through AI, within what limits and under whose responsibility. Used methodically, AI can reduce repetitive work and return valuable time to teams. Used without a framework, it can dilute quality, weaken confidentiality and create dependencies that are difficult to control.

Start with an operational decision, not a technology promise

The most useful experiments begin with a clearly identified problem: meeting notes that take too long to structure, information that is difficult to retrieve, content that must be adapted for several audiences or a recurring translation need. The tool comes afterwards. This sequence prevents a collection of subscriptions, disconnected pilots and use cases with no accountable owner.

Every use case should define four elements: the intended outcome, the data involved, the acceptable risks and the person responsible for final approval. A limited, documented and reversible pilot is often more informative than an organisation-wide rollout. It allows the team to test real value before embedding a tool in its operating model.

Define the responsibilities that cannot be delegated

AI can organise, reformulate and suggest. It does not understand the full context, the power relationships involved or the human consequences of a decision. Producing a plausible answer is not the same as exercising judgement. Several responsibilities must therefore remain explicitly human.

  • Decision-making: AI may prepare evidence, but a person must make, document and own the decision.
  • Relationships: engagement with participants, partners, volunteers and donors requires listening, continuity and judgement.
  • Representation: stories, images and testimonials must protect dignity, consent and context.
  • Verification: factual, legal, financial and sensitive information must be reviewed by a competent person before use.
AI can accelerate part of the work. It cannot carry responsibility for a decision or replace the relationship behind it.

Build an explicit validation chain

The quality of an AI use case depends less on the power of the model than on the process around it. A professional workflow clearly separates source material, generated draft, editorial review and approved version. It identifies who intervenes, when they intervene and according to which standards.

For a newsletter, AI might suggest a structure or condense selected passages. The team then rewrites the text, verifies facts and links, adjusts the tone, checks alignment with the communication strategy and approves the call to action. The final output must do more than sound correct: it must be accurate, relevant and faithful to the organisation’s relationship with its audience.

Treat data as a governance issue

Data should never be treated as mere technical fuel. Participant records, donor details, medical, legal or financial information and confidential internal documents require a precise framework. Before any use, the organisation must know what data is transmitted, where it is hosted, how long it is retained and for which purposes it may be reused.

An effective policy relies on rules that people can apply: minimise data, anonymise wherever possible, use approved solutions only and assign clear accountability. Staff, volunteers and suppliers should be able to determine quickly what they may share, through which tool and under what conditions.

Preserve a grounded, credible and recognisable voice

Automatically generated content tends to standardise language. It may be fluent while remaining interchangeable, detached from the field and filled with familiar formulas. To maintain credibility, an organisation should work from its own material: team observations, authorised testimonials, mission language, verified facts and concrete examples.

A short editorial guide can establish the essential boundaries: preferred and prohibited terms, how people should be described, the required standard of evidence and the expected transparency around AI use. The tool then serves as a production assistant for an existing identity rather than acting as a substitute author.

Assess the value created, not only the speed gained

Time saved is useful, but it is not enough. Serious evaluation also considers output quality, the amount of correction required, audience understanding, team satisfaction, detected errors and complaints. Automation that increases production while increasing review work is not an improvement.

The strongest approach remains progressive: a small number of well-chosen use cases, explicit responsibilities, controlled data and regular reviews. An organisation does not lose its human dimension simply because it uses AI. It weakens it when the tool begins to dictate its pace, language or decisions. Technology is valuable when it strengthens attention, clarity and the ability to act — never when it removes the need to exercise them.