Sarah Osman
Every few weeks, a new AI tool lands in our inboxes with the implicit message: adopt this, or fall behind. For social impact leaders, the pressure is intense. Funders are asking about it and teams are experimenting with it without a clear overview of what value it is adding (if any). The question is not “which AI tool should we use?” but “what does good AI adoption actually look like for us?”. That is a behavioural question.
Before your organisation adopts any technology, it is worth acknowledging that technology does not change what people do. People change what people do.
This is a central lesson from decades of applied social and behavioural science and technology for social impact. Distributing improved seed varieties to farmers does not guarantee adoption. Providing information about health risks does not reliably change health behaviours. And deploying AI tools across a team does not guarantee that they will be used well, used ethically, or used at all.
The questions are do we need it, and what will our people do with it, given how our organisation currently works?
If you look back at 70 years of AI development (I’ll be sharing a breakdown soon), the same pattern repeats itself: bold promises attract investment, reality falls short, trust collapses. It happened in 1974. It happened again in 1987. Each cycle was preceded by innovation and followed by overpromises.
We are in a moment of significant AI capability. Generative AI tools have real utility. But the behavioural science practitioner in me reads the current landscape with some caution: the same hype cycle mechanics are at play, and social impact organisations are being asked to make strategic decisions in the middle of lots of noise and unknowns.
The organisations that navigate this well are the ones that ground their thinking in their mission, not in the headlines.
1. What value is AI going to add in terms of intervention design?
This is the question I ask at the start of every programme design engagement, and it applies here too. What specific tasks do we want people to do differently, and why are they not doing them now? How exactly will AI help?
Instead of using the behavioural diagnosis process and designing interventions that address behaviour, some organisations are rushing to build AI-powered tools and interventions without this diagnosis. Yes the projects are cool, but what value is AI adding in the context of the issue being addressed?
2. Who will be using the AI tool, and what do we know about their needs and pain points?
Good programme design starts with the audience. AI adoption is no different. Your team members come to this technology with different levels of digital fluency, different risk tolerances, different prior experiences with tools that promised to save time but created work instead.
AI may be able to help your team, but it needs to be grounded in understanding how it adds value to their roles. This requires a process of consultation and research within your organisation and amongst teams.
3. What social norms are shaping how AI is discussed in our team?
Social norms are powerful drivers of behaviour. If the dominant narrative in your organisation is scepticism (“AI will replace our jobs”), adoption will face invisible resistance that no implementation guide can resolve. If the dominant narrative is uncritical enthusiasm, you risk skipping the quality controls that responsible use requires. Understanding the normative environment around AI in your team is essential.
4. How will we know if it is working?
This is consistently the weakest area in AI adoption conversations. Measurement gets treated as an afterthought, yet it is precisely the question that distinguishes strategic adoption from reactive adoption.
Define your success indicators before you start. Are you measuring time saved? Quality of output? Team confidence? Staff wellbeing? Without this, you cannot course correct, and you cannot build the evidence base that will inform your next decision.
5. Are we designing for sustained behaviour, or for a first-time trial?
This distinction matters enormously in behavioural design. Getting someone to try something once is a different challenge from embedding a new practice into how an organisation routinely operates. Many AI pilots succeed on the first dimension and fail on the second.
Sustained adoption requires structural support: adjusted workflows, peer learning, regular reflection loops, and leadership role modelling. It requires treating the integration not as a project but as an organisational culture shift.
There is a specific tension that social impact leaders need to address. Our field’s strength is its human-centred, culturally grounded approach. The communities we work with and for are rarely the communities whose data trained the models we are using. The risks of bias amplification, of false confidence in outputs, and of substituting AI synthesis for genuine community engagement are not hypothetical. They are already showing up in practice.
Responsible AI use in social impact work is an ethical position and a quality standard. The question to ask is who bears the cost if it produces something wrong?
There are also valid ethical reasons why teams may not want to use AI, given its negative social and environmental impacts. Leaders need to make room for these conversations.
Organisations would benefit from assessing their team’s current relationship with AI tools. This includes candid conversations on their views of the political economy around AI. And it means grounding every decision in your specific mission and the specific people you serve.
AI will be most valuable to your organisation as an assistant. The questions above help to give that direction.
I have developed a framework and process to guide organisations in navigating these questions and more, please DM, email me on sarah@osmanadvisoryservices.com or book a call with me here.