The Social Sector Has a Context Problem

By Nick Hamlin, Director of Strategy and Innovation

Contents:

• Introduction

• AI has put context in the spotlight

• Context problems are everywhere

• Soft infrastructure is critical for context

• AI makes context infrastructure even more important

AI has made me think a lot about how we study history. In school, we learn how secondary sources can be useful to get a high-level overview or a specific viewpoint about a historical event or idea. But, we’re also taught that if we really want to form our own understanding of that idea, we need to go find and read the primary sources – the original documentation of the event as it occurred – and infer our own conclusions. Trouble is, there’s a lot of detective work needed to make sense of primary sources like letters, records, and notes. The historian’s essential task is trying to understand how they fit together, the underlying context of their creation, who made them, and what unspoken incentives or biases that creator had.

Today, social sector organizations do a reasonable job of creating primary sources by documenting and organizing outputs. However, they frequently force every person outside of a project or organization to take on the role of the historian. Organizations often struggle to capture crucial context: why outputs were useful, why decisions were made, what assumptions formed a strategy, what relationships were critical for a project’s success, what constraints shaped the work, and what people learned informally along the way. But, with better context management, we can shift people from historians to problem-solvers. Context management is crucial for understanding our work and it creates reusable infrastructure that can enable organizations to make decisions, use new AI technology, and drive better outcomes and more efficient operations.

AI has put context in the spotlight

Agentic AI tools have introduced the age-old “garbage in, garbage out” data problem to a new, wider audience. Vague prompts lead to frustrating and unhelpful responses in the same way that messy, incomplete data yields unreliable and imprecise conclusions. Context is the same way: when missing or done badly, different readers will draw different learnings from the same report, new team members won’t spot the key step in a process that only lives in experienced team members’ brains, and organizations lose valuable time to repetitive conversations required to “get people up to speed”.

AI systems work the same way. Context has to be provided intentionally and explicitly or the system won’t know it exists. From there, it needs to be further organized, refined, and sometimes compressed, so the system can use it effectively over time. Just like an organization, AI systems need curated context which is often scattered across different areas of focus, one-off pilot projects, people, meetings, emails, relationships, and grant files. 

Context problems are everywhere

Context problems rarely announce themselves as context problems. Instead, they tend to  surface as frustration with bad information flow, slow processes, and useless repetition. We’ve all had the delightful experience of calling a doctor’s office or a tech support service and fully explaining our issue to the person on the other end, only to be transferred to someone else who had absolutely no context for the call and made us repeat our story from the beginning.

Context problems show up as staffing issues, when only one or a few people understand the history of a grant, program, partnership, or data system. People with this context often come to be seen as indispensable, not because of their problem-solving skills, but because the organization can’t afford to lose their institutional memory. The reliance on them often results in work that only follows this person’s way of working, doing things because they’ve always been done a certain way, rather than finding and trying new ways of solving the same persistent problems. 

Context problems appear as data problems when teams “have numbers”, but don’t understand what those numbers can and cannot say. A dashboard might show whether outcomes for a project are moving, but not whether a change in the data reflects implementation quality, community conditions, reporting differences, eligibility changes, or a shift in who is being reached. 

Context problems also appear as evaluation problems when findings are read without the implementation context that gives them meaning. A program with mixed results may be dismissed as ineffective, when the more useful lesson is that it worked under certain conditions, for certain participants, or with a level of relational support that was never built into the formal model.

There are many other examples, but in each case, the question is often the same: what context was missing, who had it, and why wasn’t it shared? 

Soft infrastructure is critical for context

It’s usually easy for organizations to think about “hard infrastructure”. These are the technical tools and systems (databases, CRMs, software, etc.) that organizations use to capture and move information around. In contrast, organizations tend to have a much harder time analyzing and improving their “soft infrastructure” — the people and process architecture that drives an information ecosystem. Soft infrastructure is how decisions get made, translated into action, and tracked over time. It includes clear project objectives and the mechanisms for ensuring staff have the expertise and resources needed to deliver those objectives. It’s the people and roles, processes and routines, governance, and adaptation mechanisms organizations use to react to change. 

At DARO, we’ve seen countless examples of how context creation suffers when organizations don’t prioritize soft infrastructure. This challenge is common to resource-constrained organizations because it can feel like an extra step that’s never as urgent as delivering services, launching a program, or meeting a deadline. But, context creation and management is an important component that shouldn’t be sidelined. Dedicated processes to document and continually refine and update context artifacts enable organizations to significantly speed up the time it takes to try new experiments, iterate on programs, and learn faster as a team. Otherwise, organizations might be able to solve a few scattered context problems as one-offs, but they’ll be missing the critical systems needed to do it again and again, limiting their ability to solve new problems and update their shared learning over time.

AI makes context infrastructure even more important

Investing in context management via thoughtfully managed soft infrastructure has been a key differentiator for successful social sector organizations already, but AI makes that context even more valuable. Now, many of the artifacts that we might create to capture context and pass it between humans, like process documentation, role definitions, governance policies, etc. can also be used to quickly get AI agents up to speed on how an organization works. Before this shift, context for humans often had to be translated into context for computers, like translating a process into code that captured the corresponding business logic. Now, the same context artifact works in more places, and organizations have more incentive to keep them compact, up-to-date, and discoverable as a result.

Organizations that excel at context creation and management will not only be well-positioned to make use of the newest AI tools, they’ll also be more resilient to the social sectors constantly shifting political and funding environments.  They’ll be able to map their current organizational understanding onto new problems, domains, and challenges much quicker than organizations with poor context artifacts. They’ll also have built strong organizational muscles, which will help them add how and why they are shifting priorities to their context artifacts in a way that is clear and usable.

So, the next time a donor complains that their contact person at your organization knows nothing about their last gift, or a new employee struggles to “drink from the firehose” and learn everything about your programs, or no one knows if that document labeled “data retention policy draft v2 for real this time final FINAL.docx” on your shared drive is actually trustworthy or not, ask yourself: am I actually facing a context problem? And then ask yourself, if you have the right soft infrastructure in place to work on that context problem. If you don’t, it’s probably time to start investing in how you create, manage, update and refine context for your organization.