Why we wrote a playbook for charities to use AI

The language coming out of Silicon Valley about AI is alienating. Not because it's wrong, exactly, but because it's written for a world that doesn't look like the charity sector. The breathless announcements about foundation models and emergent capabilities and artificial general intelligence are hard to translate into "should we use AI to process our beneficiary feedback, and if so, how do we do it without breaching data protection?"

That gap between what the tech industry talks about and what charities need to know has been bothering us for a while. We've worked with Oxfam, RNIB, Cancer Research UK, Breast Cancer Now, and others on making sense of AI. In every engagement, we saw the same disconnect: capable professionals who wanted to understand AI but couldn't find guidance that started from their reality rather than the technology industry's.

So we wrote the AI Playbook for Charities. It was an absurd thing to do. We're a small agency. The time it took was time we weren't spending on client work. Almost everyone we talked to about it suggested we shouldn't bother. They were probably right about the economics. But we thought it was important and nobody else was doing it well.

What writing it taught us

The playbook started at 50,000 words. Both of us could agree that was too long. We cut it to just over 11,000, which is still either too long or too short depending on your perspective. The cutting was harder than the writing. Deciding what to leave out forced us to think about what actually matters versus what's merely interesting, and that editorial process was more valuable than we expected.

Selfishly, writing the playbook was incredibly useful for us as an agency. It forced us to map where AI is genuinely useful for charities right now and where it's still immature. It made us more precise about what we think and more honest about what we don't know. The playbook made us sharper at the work we do with clients, which wasn't the reason we wrote it but turned out to be one of the best reasons we did.

AI didn't begin with ChatGPT in November 2022. Our team has worked with machine learning and deterministic AI since 2017. That gives us a longer vantage point than people who discovered AI through a chat interface, though a much shorter one than researchers who've been at this for decades. We think it's long enough to have a useful perspective on what's real and what's marketing.

The bridge problem

The playbook is an attempt to bridge two worlds. Not to simplify AI for people who can't handle complexity. Charity professionals handle extraordinary complexity every day in service delivery, safeguarding, fundraising governance, and impact measurement. The bridge goes the other way: recontextualising what's happening in AI for people whose expertise is in a different domain.

When we said "sometimes the best AI tool for this task is ChatGPT, sometimes it's Claude, sometimes it's a spreadsheet formula," people were relieved. They'd been getting sales pitches disguised as guidance and were tired of it. Vendor-neutral advice shouldn't be unusual, but it is.

Where the playbook reaches its limit

Writing 11,000 words about what charities can do with AI also made something else very visible. For every section where we could honestly say "here's how to approach this yourself," there were areas where the honest answer was "this needs someone to build something."

Data integration across disconnected systems. Feedback analysis at scale that accounts for the specifics of your programmes. Automated workflows connecting your CRM to your reporting. These aren't things you solve by reading a guide, however comprehensive. They're build problems.

The playbook tries to show what's possible and help you ask better questions. But writing it also made very visible where the boundary sits between "you can do this yourself" and "this needs someone to build something." We didn't set out to write a case for our own services, but that boundary turned out to be very real. A playbook tells you what's possible. It doesn't build the thing.

The playbook is at aiplaybookforcharities.com. We'd suggest reading it in sections relevant to your role rather than cover to cover. If we got things wrong, we'd rather hear about it than repeat it in the second edition.