Why charity meeting notes are deceptively hard for AI

Every charity we work with has the same question somewhere on their AI wish list: can we automate meeting notes? It seems like it should be easy. Teams and Zoom both offer built-in transcription. There are dozens of dedicated tools. The technology exists, so the problem is solved, right?

Not quite. There's a real gap between "a transcript exists" and "meeting notes that are actually useful to the organisation." For charities specifically, that gap is wider and more interesting than most people expect.

We've spent time on this problem, and what we've found is that meeting transcription is a neat case study in why surface-level AI adoption often disappoints. The technology works. The transcript is accurate enough. But the transcript isn't what anyone actually needs.

What a transcript gives you versus what you need

A raw transcript gives you a wall of text. It captures everything that was said, including the five minutes of small talk at the start, the tangent about the car park, and the moment someone's dog barked. What a charity team actually needs after a meeting is different: clear action items assigned to specific people, decisions recorded with context, follow-ups linked to the right project or funder, and anything safeguarding-relevant flagged for the appropriate lead.

Getting from one to the other is where the difficulty lives.

Start with speaker identification. In a small internal team meeting, it's manageable. But charity meetings often include external partners, trustees, beneficiary representatives, local authority contacts, funders. Teams transcription can usually identify people with Microsoft accounts in your organisation. Everyone else shows up as "Speaker 3" or gets attributed to the wrong person. For a multi-agency safeguarding meeting or a trustee away day with guest speakers, the transcript can be close to useless for attribution.

Then there's the language problem. Charities use acronyms and sector-specific terms constantly. GDPR and CRM are common enough, but what about ToC (theory of change), RoI in the charity sense (return on investment, but calculated very differently from the commercial version), or organisation-specific shorthand? One client we worked with referred to a major programme internally as "the lighthouse project," which every transcription tool dutifully recorded but no AI summariser could make sense of without context. Generic transcription handles generic language well. Charity meetings aren't generic.

The problems that only appear when you look closely

Action items are where things get properly tricky. In a commercial meeting, "John will send the report by Friday" is a clear, extractable action. In a charity meeting, you get: "We need to follow up with the funder about the timeline." Which funder? Which grant? Whose job is it? The answer is obvious to everyone in the room and completely opaque to an AI reading the transcript. Useful action extraction requires organisational context that no off-the-shelf tool has.

Confidentiality is a serious concern that gets too little attention. Charity meetings routinely discuss individual beneficiaries, sometimes by name. Case conferences, service reviews, safeguarding discussions: these contain sensitive personal information that shouldn't be processed by generic cloud transcription services. Where is that audio being sent? Is it stored? Could it be used for model training? For a charity discussing a vulnerable person's circumstances, these aren't abstract data protection questions. They're duty-of-care questions.

And then there's what we think of as the multi-meeting problem. The real value of meeting intelligence isn't capturing what happened in one meeting. It's tracking decisions and actions across a series of meetings. Did the follow-up from the June board meeting actually happen? Has the risk register item from three meetings ago been resolved? Is the same issue being discussed for the fourth time without progress? This kind of longitudinal tracking is where meeting notes become genuinely valuable to an organisation, and it's well beyond what any current transcription tool offers out of the box.

Where does this leave charities?

The gap between "Teams can transcribe" and "we have meeting intelligence that works for our organisation" is real. It's not a gap that gets filled by buying a better transcription tool. It gets filled by thinking carefully about what your organisation actually needs from its meetings and building something that accounts for your context, your language, your confidentiality requirements, and your workflows.

That might mean a transcription tool connected to your CRM so action items land in the right place. It might mean a custom prompt layer that knows your organisation's acronyms and project names. It might mean a policy decision that certain meeting types are never transcribed by cloud tools, full stop. It probably means some combination of all three, tailored to how your organisation actually works.

We find this kind of problem interesting precisely because it looks simple on the surface. "Just transcribe the meeting" is the kind of request that sounds like a quick win, and it can be, if all you want is a transcript. But if you want meeting notes that save your team time, feed into your systems, and respect the sensitivity of what gets discussed in charity meetings, you're probably looking at some bespoke work to get it set up properly, not an off-the-shelf purchase.

That's not a reason to avoid the problem. It's a reason to approach it properly. And for what it's worth, the charities that have invested in getting this right tell us it's one of the highest-value AI applications they've implemented. Not because the technology is impressive, but because meetings are where decisions happen, and decisions that get lost in inboxes and forgotten notebooks are expensive in ways that never show up on a budget line.