Social impact organisations are under growing pressure. Needs are rising, resources are constrained, and teams are being asked to do more for more people, often with less. In this context, artificial intelligence offers real promise. Used well, AI can help organisations unlock knowledge, reduce administrative burden, improve decision-making and extend support to people and communities.
But AI is not a shortcut to impact. It cannot replace trust, relationships, good governance, local knowledge or strong institutions. If used poorly, it can amplify bias, deepen exclusion, spread misinformation and create new risks for the people it is intended to support.
For businesses, funders, governments and civil society, the question is not whether AI will shape social impact. It already is. The more important question is how to ensure that AI strengthens lives, livelihoods and learning in ways that are inclusive, accountable and grounded in real human needs.
Drawing on Business Fights Poverty’s recent insight paper, Staying Ambitious: Harnessing AI for Social Impact, five lessons stand out.
- Start with the system, not the tool
AI’s impact depends less on the model itself than on the system into which it is introduced. A sophisticated tool can fail if the surrounding conditions are weak. A simpler tool can be powerful if it is embedded in trusted relationships, clear workflows and strong institutions.
This matters because many AI conversations still focus on tools, pilots and use cases. These are important, but they are not enough. A health AI tool depends on clinical validation, referral pathways, trained workers, patient trust and regulatory oversight. An agricultural advisory tool depends on farmer registries, weather data, local languages, connectivity and trusted delivery channels.
For business, this points to a more disciplined approach. The starting point should be the social problem, not the technology. What gap is being addressed? Who could benefit? Who might be excluded or harmed? What infrastructure, governance and capacity are needed for the tool to work over time?
Staying ambitious does not mean adopting AI wherever it is available. It means using AI where it can solve a real problem, fit a real workflow and improve outcomes for people.
- Build partnerships with clear accountability
Responsible AI for social impact cannot be delivered by one actor alone. Technology companies bring engineering expertise and infrastructure. Governments bring legitimacy and responsibility for public systems. Civil society organisations bring trust, community relationships and delivery experience. Academia brings evidence and scrutiny. Funders bring patient capital and convening power. Communities bring lived experience and the right to shape what is built for them.
The challenge is not simply to bring these actors together. It is to clarify responsibility. Who defines the problem? Who owns the data? Who maintains the system? Who monitors performance? Who is accountable if the tool causes harm? Who decides whether to scale, adapt or stop?
Without this clarity, partnerships can become fragmented, duplicative or extractive. Too many disconnected pilots can waste scarce resources and weaken learning.
For businesses, the role is broader than providing technology. Companies can act through core business, philanthropy and policy engagement. They can shape how AI is used in operations, value chains, products and services. They can support civil society organisations to use AI responsibly. They can advocate for safe, ethical and inclusive regulatory frameworks.
The strongest partnerships are problem-led, not product-led. They are built around shared goals, transparent governance and a willingness to share responsibility for outcomes.
- Focus on human-augmenting AI
In many social impact contexts, the most useful AI may not be the largest or most advanced model. It may be a smaller, task-specific tool that works reliably within real-world constraints.
For a frontline health worker, AI may help triage urgent cases, support diagnosis, translate information or reduce paperwork. For an agricultural extension worker, it may provide advice in the right language, based on local crops, soil and weather conditions. For a social worker, it may help identify which families need support most urgently, while leaving the relationship-based work to people.
This distinction matters. Across poverty reduction, healthcare, education, financial inclusion and humanitarian response, the strongest uses of AI are often about augmenting people, not replacing them.
AI can also unlock organisational capacity behind the scenes. It can help draft reports, summarise evidence, clean survey data, translate documents, map systems, support proposals and improve knowledge management. These uses may appear less exciting than frontline applications, but they can free up scarce human capacity for higher-value work.
The practical test is whether AI helps people do their work better. Does it strengthen judgement, relationships and agency? Does it reduce low-value tasks so people can focus on trust, care and problem-solving? Or does it simply cut costs while shifting risks onto workers and communities?
The goal should not be AI-first. It should be human-first and AI-enabled.
- Measure outcomes, not adoption alone
AI success should not be measured only by usage, satisfaction or technical performance. A tool may be popular, efficient or impressive without improving social outcomes.
The real test is whether AI improves lives, livelihoods and learning. Does it strengthen access to healthcare? Improve incomes or job quality? Support education and skills? Improve decision-making? Reduce harm? Increase access for people who were previously excluded?
This requires evaluation from the beginning. Organisations need to understand the baseline before AI is introduced: current workflows, costs, waiting times, quality, user experience, risks and outcomes. After deployment, they need to assess whether the tool improves the wider system, not just whether people are using it.
Safeguards are equally important. AI can amplify bias, misinformation, fraud, unsafe decision-making and exclusion. In finance, it can expand access to credit, but also enable opaque or discriminatory decisions. In healthcare, it can support diagnosis, but false outputs or model drift can cause harm. In recruitment, it can speed up screening, but reproduce structural bias.
Responsible AI is not a one-off approval decision. It requires ongoing monitoring, human oversight, transparency and clear routes for redress.
- Treat inclusion as agency, not access alone
Inclusion is not simply about giving more people access to AI tools. It is about who defines the problem, whose knowledge counts, who owns the data, who shapes the design and who benefits from the value created.
Many AI systems are built on data that reflects wealthy, connected, English-speaking and institutionally visible populations. This creates the risk that marginalised communities, low-resource languages, rural populations, women, oral cultures and offline communities are excluded or misrepresented.
Language is one of the clearest examples. If people cannot ask questions, access services or express needs in their own language, they are excluded from shaping the digital future.
Responsible AI must therefore involve community participation in design, testing and governance. It must also recognise that digital inclusion is not always best achieved through apps or smartphones. In some contexts, printed materials, voice interfaces, basic phones, community intermediaries or in-person support may be more appropriate.
For businesses and funders, inclusion means investing in diverse datasets, low-resource languages, accessible design, local research capacity and participatory governance. For civil society, it means protecting the trust they hold with communities and ensuring that sensitive data is handled with care.
A human-centred AI future
AI offers a genuine opportunity to help social impact organisations stay ambitious in a time of crisis and constraint. It can unlock knowledge, reduce administrative burden, improve decisions and extend frontline capacity.
But its value depends on the choices made around it. Businesses, governments, funders and civil society need to invest not only in tools, but in the systems, partnerships, safeguards and community agency that make responsible AI possible.
For business, the opportunity is clear. AI can strengthen lives, livelihoods and learning, but only if ambition is matched by responsibility, inclusion and partnership.
That is the task ahead: not simply to deploy AI for social impact, but to shape it with people, for people, and in service of more equitable outcomes.
