AI Adoption In Social Impact Sector: What Do We Mean?

AI is transforming the way organizations operate, and nonprofits and government agencies are no exception. The potential of artificial intelligence in these sectors is vast—enabling more efficient processes, more informed decisions, and innovative solutions to complex challenges. There has been a lot of chatter about the need to begin using AI, but what does AI adoption look like for these organizations, and how far might they go in embracing these technologies?

AI adoption by nonprofits and government organizations can be envisioned across a spectrum, with varying levels of engagement. From leveraging intuitive AI tools to developing sophisticated, customized models, the journey will differ depending on the needs, scale, and program mix. Adoption falls into three broad categories:

1. Consumer Use: Use of consumer AI tools to help with personal and team productivity engaging generative AI vendors, AI within established applications, and increasingly pre-built AI agents. 

2. Advanced Use: Use of advanced AI skills such as prompt engineering with fine tuning and RAG to build out models for specific use cases on top of existing Large (and Small) Language Models and increasingly build out Agents on top of existing AI tools and applications.

3. Builders: Purpose-built models and productization of AI tools that will be used by other social impact organizations.

I will say at the outset, that I would not place value on one level of AI adoption over another. Indeed, even within a particular organization, the adoption levels might look quite different between program staff and engineering staff for example.  The aim is to provide context so that social impact leaders can level set their expectations for themselves and their teams.  In short, not everyone is going to build their own Large Language Model, just like everyone doesn't build their own Constituent Relationship Management (CRM) system. And that's okay. Social impact organizations should adopt the technology that is appropriate for their context.

Consumer Use: 40-60% of Total Adoption

The majority of AI adoption by social impact organizations will be driven by front-end tools that seamlessly integrate AI capabilities into the everyday workflows of nonprofit and government staff. Tools like Microsoft Co-Pilot, ChatGPT, Google Gemini for Workspaces, and others will become an integral part of the toolkit. These applications make AI accessible to people who aren't necessarily technical experts, allowing them to benefit from automation, enhanced decision-making, and improved productivity.

This will mean that around 40-60% of staff and organizations will start utilizing AI by interacting with these user-friendly, built-in AI functions. Tasks like drafting reports, organizing schedules, analyzing data, and communicating with stakeholders will be augmented by generative AI capabilities, helping teams work faster and with greater insight. Moreover, this phase will see existing tools like CRMs enhanced by AI, as well as AI tools developed by other nonprofits being shared across the sector.

Crucially, we’re beginning to see the next evolution—AI agents. These agents will soon take on even more complex tasks that require multiple steps or sustained engagement, complementing the abilities of human staff and freeing them up for more meaningful, strategic work. This blend of generative AI and task-oriented agents will be a major step forward in terms of how nonprofits and government organizations use AI to amplify their impact.

Advanced AI Skills and Use Cases: 25-40% Adoption

The next group of nonprofits and government organizations—around 25-40%—will move beyond the front-end tools and start engaging with more sophisticated AI technologies. For these staff and organizations, AI adoption will extend to advanced skills like prompt engineering, where staff learn to effectively communicate with large language models to derive specific insights or produce customized results.

This group will also begin using approaches like Retrieval-Augmented Generation (RAG), where they enhance existing large language models (LLMs) like ChatGPT, Claude, or open-source models such as LLaMA with their own data. This allows them to create AI solutions that are tailored to their specific needs—whether that’s creating detailed analysis of social impact data, developing interactive educational content, or building customized chatbots that reflect their organization’s values and priorities.

These organizations may also dip into fine-tuning existing LLMs to better serve specific operational needs, using their proprietary data to ensure that the AI's output is as relevant and helpful as possible. This type of customization can significantly enhance efficiency and alignment to specific missions, offering deeper, more precise ways to serve beneficiaries and manage internal processes.

Purpose-Built LLMs for Specialized Needs: A Small, Niche Group

Finally, there will be a small but important subset of nonprofits and government agencies and staff from those organizations that go all-in, building their own purpose-built large language models. These organizations may have specific requirements that call for such a tailored approach—such as managing highly sensitive data, ensuring robust compliance with regulatory frameworks, or delivering mission-critical services that cannot rely on externally provided AI models.  For example, government entities that need to manage classified information, or nonprofit organizations handling sensitive health data for vulnerable populations, may choose to develop proprietary models to ensure that all data is kept in-house and that the AI behaves in exactly the way they need. 

These organizations may not particular requirements, but have ambitions to deliver tools at scale beyond their own organization, either by sharing the models they build with others or productizing their AI technology to deliver with in multiple contexts.  Some of the work in the early AI for social impact accelerators are seeing early candidates for these kinds of scale.  

This approach requires substantial resources, both in terms of technology and expertise, and will likely only be viable for  organizations with deeper technical expertise, or those with very specific needs that can't be met by off-the-shelf AI solutions.

The Road Ahead: Adapting and Growing with AI

AI adoption is not a one-size-fits-all process. The unique missions and values of nonprofits and government organizations will drive their decisions on how best to leverage AI. The exciting part is that we are already seeing a diverse range of adoption strategies, from integrating existing AI tools to developing entirely bespoke solutions.

Regardless of where an organization falls on this spectrum, the key is to approach AI with an open mind and a focus on how these tools can enhance their mission. The potential for generative AI, agents, and fine-tuned models to elevate the work of these organizations is massive—and the real opportunity lies in using these technologies to further positive social impact.

Are You Ready for AI?

AI adoption can feel overwhelming, but it’s clear that the tools are getting more accessible and relevant to nonprofit and government work every day. Whether you're excited to start using tools like ChatGPT for daily tasks or considering how to customize AI for your organization's specific needs, there’s a path for everyone.

Where do you see your organization on this spectrum? Are you ready to take the next step with AI?

Author's Note: I wrote this blog in conjunction with Chat-GPT. Transparency in the use of AI is an important principle in the ethical use of AI.