Focus area
AI & Automation
We build AI that supports the expertise of development and humanitarian teams, and we run it with them.
Why it matters
Development and humanitarian teams work in crisis with limited time, sensitive data and a duty to be accountable to the people they serve. If properly set up, AI can help on all three, above all with the volume of data: feedback, surveys, media and social conversation already exceed what a small team can go through by hand.
Yet, without meaningful assessment, AI produces outputs that might look convincing, but might not be the actual answers communities need.
Use of AI tools is already widespread in the sector while few organisations have internal policies or training behind it, so tools are adopted without scoping or review, and errors that are hard to spot reach reports and recommendations, leaving organisations and countries with less expertise the most exposed.
The question is therefore where automation can really support sector expertise, so that teams can verify the results, stay responsible for them and make faster and better decisions to meet the needs of the communities they serve.
Why the way AI is built matters
Each tool is adapted to the technical and analytical capabilities of the team and works with the systems it already uses, through an API, a dataset upload or a browser extension, and a chat assistant lets any colleague question the data in plain language, so that findings are not limited to analysts.
Tools that fit the team
We follow which false or misleading health content circulates, in which languages and on which channels. We separate misinformation, which is shared without the intent to mislead, from disinformation, which is created and spread on purpose, because the response is not the same.
Your own document as the core
Our assistants answer from the organisation’s own research and guidance library, past campaigns and evidence, in the format the team reports in, so that outputs sound like the organisation and rest on what it already knows.
Science-based
Our tools follow the same methodology and rigour as any research study: categories are defined with the team, our practitioners, analysts and developers test the automation before it is scaled, and every result shows what it rests on, from the explanation beside each data point to the sources behind a persona and the patterns behind a risk score.
Science-based
Analysts review and correct the categorisation, the team sets the thresholds for alerts, and each change suggested by our editorial assistant is presented on its own for a person to accept or leave, so that the team has the last word.
Data protected
Each dataset is visible only to the people who own it.
Adapted to offline use
Where the topic, the country or the organisation requires it, the tools can be adapted to run offline.
Our work in practice
Research for Purpose has built AI with development and humanitarian teams so that they spend less time preparing data and more time acting on it. For the World Bank, we automated the analysis of open ended survey answers, which saved tens of hours per survey and gave more accurate results than manual coding. For the Pacific Community, we run an alert system for 17 diseases with its analysts, who now spend more time designing interventions.
We bring the same experience to helping teams understand AI and to listening to the communities they serve. We trained World Bank communications and external affairs staff on how generative AI works, its risks and opportunities, and the Bank used it for its next steps on guidelines and capacity building. For UNICEF, the community listening platform we built is used by more than 20 of its offices, with analysts validating the results.
In every project, practitioners and AI engineers work alongside your team, and we document the method so that your team can run it. Our AI Listening & Insights Platform, Misinformation Virality Detector, AI Personas, Editorio and Opinion Research & Survey Insights Platform support this work.