DevelopMetrics

Shout out to Matiné Esmailian Matiné did an outstanding job hosting AI for Development & Impact 2025. She kept the audience engaged, introduced each speaker with energy and clarity, and helped the sessions flow seamlessly. #AI4D2025 #AIforDevelopment #DevelopMetrics #AIforGood Devex TechChange Humanity Link

Great recap from our media partner Devex on the AI for Development & Impact (AI4D) conference. From efficiency gains to real-world impact, the piece highlights how AI is reshaping global development. Featuring insights from Nasim Motalebi, Ph.D., Lindsey M., Mara Delia Puacz, and Javier Ideami. Check out the key takeaways below. #AI4D #AIforDevelopment #ImpactTech #ResponsibleAI #DigitalDevelopment #DataForGood

AI4D 2025: Building the AI for Development Alliance Thank you to all the people who joined us, both online and in person, for this year’s AI for Development & Impact (AI4D) conference. This year was not about the next big model drop. It focused on something more practical and ambitious: linking our data, models, and experience into a shared AI ecosystem for development. AIDA: The Ecosystem for Shared Intelligence At DevelopMetrics, we introduced the AI for Development Alliance (AIDA), a global effort to connect development datasets, models, and case studies into a living map of what is working, where, and why. When organizations operate in similar spaces, AIDA helps them connect data responsibly, avoid duplication, and coordinate faster. Some partners share data openly, others license their models, all contributing to a common index of evidence and impact. From Hype to Implementation Nasim Motalebi, Ph.D. Motalebi, AI Lead at WFP, showed how domain-specific models trained on clean, contextual data often outperform general ones. Her message was clear: start from real workflows, understand your people and data, and then build. Real-World AI: GIZ’s NegotiateAI A highlight came from Robin Nowok and Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ) GmbH, whose NegotiateAI tool supports Global South delegations in UN plastics treaty talks. Using multilingual search and retrieval-augmented generation, it analyses draft texts and country positions so smaller delegations can negotiate on equal footing. This is exactly the kind of model AIDA aims to link across development domains. Evidence Over Internet Exhaust Elia Gandolfi and Sasha Tanghe @The Agency Fund reminded us that the development sector’s greatest strength is its evidence base. When we build on mission-aligned data, we get AI that reflects real human context, not internet noise. 5️⃣ How DevelopMetrics Fits In Our Development Evidence Large Language Model (DELLM) is trained on decades of verified evaluation data. It helps practitioners learn from what has already worked, using AI that is transparent and grounded in evidence. Integrated with AIDA, it allows governments, NGOs, and investors to search and reason from shared data in real time. The focus is shifting from models to methods, from isolated pilots to shared intelligence. Through AIDA, we are helping build the infrastructure to make that shift possible. Thank you to Devex, TechChange and Humanity Link for partnering with us to bring this global community together, and to everyone who joined from around the world. Stay tuned If you would like to learn more or get involved, reach out. We are happy to talk with anyone working to make AI work for development. #AI4D #AIDA #AIforDevelopment #DataForGood #ImpactTech #ResponsibleAI #DELLM #GIZ #NegotiateAI #DigitalDevelopment

My colleague, Dan, and I were in Barcelona this week representing Elrha at the AI for Development and Impact conference, hosted by DevelopMetrics. It was pretty inspiring to be surrounded by researchers, practitioners, innovators, entrepreneurs, and funders who are all actively trying to translate #AI into real-world impact. Some key recurring themes that stood out: 🤝 𝗖𝗼𝗹𝗹𝗮𝗯𝗼𝗿𝗮𝘁𝗶𝗼𝗻 𝗶𝘀 𝘁𝗵𝗲 𝗳𝗼𝗿𝗰𝗲 𝗺𝘂𝗹𝘁𝗶𝗽𝗹𝗶𝗲𝗿: no single actor can do this well alone. Embedding responsible AI at scale depends on standardisation and open sharing of methods, data and evaluation (eg DevelopMetrics’ AidInsight platform (Lindsey M.), OpenAI's Academy (Emmanuel Lubanzadio), or The Agency Fund’s AI evaluation framework (Elia Gandolfi) showcase how structured collaboration is already changing the landscape). 💾 𝗘𝗾𝘂𝗶𝗽 𝗽𝗲𝗼𝗽𝗹𝗲 𝗶𝗻𝘁𝗲𝗻𝘁𝗶𝗼𝗻𝗮𝗹𝗹𝘆: If you’re early in the journey of embedding AI in your organisation, training staff on AI should not be the first step! Start more basic than that by running a digital skills audit. You must know the baseline to be able to train meaningfully. The big question is not ‘who knows AI’ but ‘how can technology really help our work instead of creating further issues down the line’? 📎 𝗔𝗜 𝗿𝗲𝗮𝗱𝗶𝗻𝗲𝘀𝘀: This seemed to be the hot topic these last two days. Individual AI adoption outpaces organisational readiness with 9 in 10 interviewed humanitarians using or having tried AI tools while only 8% organisations reporting wide AI integration (finding from the 2025 Humanitarian AI Survey, a global initiative co-led by the Humanitarian Leadership Academy and Data Friendly Spaces). Some other great insights shared by Daniela Weber (NetHope) and José Cobos Romero from Data Friendly Space (DFS), who are working with us on supporting three grantees building AI solutions for humanitarian response. 🔍 𝗧𝗲𝘀𝘁 𝘁𝗲𝘀𝘁 𝘁𝗲𝘀𝘁, 𝗹𝗲𝗮𝗿𝗻, 𝗶𝘁𝗲𝗿𝗮𝘁𝗲: ongoing experimentation is key to ensuring AI tools work and are fit for purpose in often messy, real-world conditions. ✅ 𝗥𝗲𝗴𝘂𝗹𝗮𝘁𝗶𝗼𝗻 𝗺𝘂𝘀𝘁 𝗽𝗿𝗼𝘁𝗲𝗰𝘁 𝘄𝗶𝘁𝗵𝗼𝘂𝘁 𝗱𝗲𝘁𝗲𝗿𝗿𝗶𝗻𝗴 𝗶𝗻𝗻𝗼𝘃𝗮𝘁𝗶𝗼𝗻: Everyone agreed (how often does this happen nowadays?) regulation is needed. However, the major challenge is how to design regulations so that impact innovation is not unintentionally stifled. Lastly, an area that could have been unpacked more is the impact of AI on climate. While large models have historically consumed substantial energy (especially during training), new research shows much more energy-efficient model architectures with smaller models becoming now a viable alternative. I think more conversations are needed on this as there’s still significant room for innovation to reduce the negative environmental impact of both model training and deployment. #AI4D2025 #ResponsibleAI #Innovation #HumanitarianAI

Grateful to have joined the #AI4D2025 Conference virtually. It was an inspiring and insightful event bringing together voices from across the world to discuss AI, data, and development impact. The sessions on responsible AI, measurement, and governance were especially relevant to ongoing efforts to make technology work for everyone. Appreciation to the organizers DevelopMetrics and speakers for an excellent conference and thoughtful discussions. #AI4D2025 #ResponsibleAI #DataForDevelopment #AIforGood

A big thank-you to TechChange and Nick Martin for powering this year’s AI for Development and Impact Conference with Kolab! We were thrilled to collaborate on the virtual networking and lightning talk sessions bringing together hundreds of participants across sectors to connect, share ideas, and experience what truly engaging digital convening can feel like. The TechChange team has built something remarkable with Kolab, and it was the perfect fit for fostering collaboration and creativity throughout the conference. #AI4D2025 #AIForGood #globaldevelopment #Kolab #TechChange