We kicked off our recent London Funders Festival of Learning session with an unusual challenge. How would you describe a social problem you work on in a game of Taboo, where you can’t mention the beneficiaries or the social problem directly?
It sounds like a party game, but it landed a serious point. When we ran a round of “Philanthropy Taboo” in the room, participants quickly discovered that the words we reach for to define a field are also the words that hide it. When describing social cohesion, take away “cohesion,” “belonging” or “community,” and you’re forced to describe activities, relationships and outcomes that take place within and around that field, which we can call the system. That inconvenience captured the whole theme of the morning: funders are trying to act in systems, but the system doesn’t come neatly labelled.
Setting the scene: why landscapes are hard to read
There are three things that make complex fields so difficult to interpret:
- Social issues are interconnected — poverty, health, housing and inequality overlap rather than sitting in silos.
- The funding landscape is fragmented — organisations describe similar work in very different language.
- And funders now face more information and demand than they can realistically process, especially as AI makes it easier to generate polished proposals without adding real clarity about fit, need or duplication.
Hence, funders need a way to see the system to understand what organisations actually do rather than what they call themselves, and to spot where activity is concentrated, thin, duplicated or disconnected.
The method we shared: from fragmented language to landscape intelligence
We then walked the room through our approach, which combines four ingredients: publicly available data (charity registers, grants databases, websites, reports); a clear taxonomy defining the field; large language models and semantic similarity tools that compare meaning rather than exact words; and human interpretation to sense-check and apply judgement.
The idea that resonated most was the shift from keyword search, “does this text use the words I searched for?”, to semantic similarity, “is this talking about a similar idea, even in different words?” In our methodology, a machine learning model places every organisation on a “map of meaning,” so two charities doing near-identical work sit close together even if one says “cohesion” and the other says “belonging.”
The case study: social cohesion
To make it concrete, we shared our social cohesion mapping. We built a six-domain taxonomy and applied it to the Charity Commission register and 360Giving data. This surfaced 14,560 charities and 70,816 grants, many from organisations that never use the word “cohesion.” These included lunch clubs, community gardens and local sports teams that build trust without ever describing themselves that way.
Our analysis found that areas with more social cohesion charities per 10,000 residents tended to have fewer mission-critical neighbourhoods - a term from the Independent Commission on Neighbourhoods (ICON) for areas with especially urgent social needs and significant gaps in local services and cohesion. In other words, places with a denser base of community organisations were, on average, less likely to fall into that most-at-risk category. Though this is a correlation, not causation.
We also analysed the top 100 funders in the social cohesion space, grouping them by the organisations they fund. This showed that the statutory funders all had very similar organisations in their recipient portfolios, prompting a question about where philanthropy can genuinely complement public investment rather than reinforce an already well-funded part of the system.
You can read more about how this work began in our blog, Seeing the bigger picture: using a systems approach to guide philanthropy.
The group activity — and what people took away
We closed by inviting groups to build their own taxonomy for a field they care about: defining an outcome, breaking it into domains, and identifying the language organisations might use. The exercise brought the morning full circle — back to Taboo, and the reminder that the words we use determine what we see.
Our honest framing to the room was this: data doesn’t remove complexity, and it won’t tell you how impactful a charity is or where it truly works. But it can help you see the whole field — not just the part that speaks your language — and that is where smarter, more joined-up funding can begin.