It’s nine o’clock and the coordinator is still at her desk, the only one left in a building that ran a program earlier and will run it again tomorrow. She is the program. She answers the phone, unlocks the door, sits with the people who come in, and locks up after them. Tonight she is doing none of that. She’s writing a grant acquittal, matching what the money did against what the application said it would do, line by line, because the funder needs it by Friday and there is nobody else to write it.

Down the hall the program room is dark. The people it’s for went home hours ago, or didn’t come at all, because the hours the grant paid for got spent on the report about the hours.

That’s the quiet arithmetic almost nobody in the sector says out loud. Every donated dollar that goes into proving how the last dollar was spent is a dollar taken off the mission. The donor gave it for the work. It ended up funding the paperwork about the work. And the person doing that paperwork is usually the same person who was supposed to be doing the work.

What the paperwork is really costing

Wages are the single biggest line in the charity sector’s costs, about 55 per cent of the total, and they’re rising close to 10 per cent a year, the steepest climb on record (ACNC, Australian Charities Report, 11th edition, 2025). That’s the number a board sees. What the board doesn’t see is where a real slice of that wage bill actually goes: not to the mission, but to assembling reports, cross-matching acquittals, and clearing an inbox that never empties.

None of that is waste in the lazy sense. Acquittals have to be honest and funders have a right to know. But the compliance load is now heavy enough that reporting has become the sector’s most-wanted capability. When Infoxchange surveyed 824 organisations, “data and reporting for evidence-based decisions” came out as the number one digital priority at 44 per cent, up from just 17 per cent in 2023 (Digital Technology in the NFP Sector, 10th edition, 2025). The sector isn’t asking for reporting help because it likes reports. It’s asking because reporting is eating the week.

And the demand at the other end hasn’t paused to let anyone catch up. Ask Izzy, the service-finder a lot of Australians reach for when things go wrong, logged 8.5 million searches for help in 2025, its highest ever, after a 30 per cent jump in hardship searches the year before (Infoxchange Annual Report, 2025). More people are knocking. The coordinator is at her desk writing about last quarter.

Two reports, one grind

Look closely at what she’s actually doing tonight and it’s the same motion twice.

The acquittal cross-matches program spending against the restricted categories the grant came with, so she can show every dollar landed where it was promised. Next month she’ll write the funder impact report, turning case notes and attendance into a story about outcomes. Different documents, same underlying work: taking what happened on the ground and turning it back into the shape the money asked for.

The regulatory side doesn’t make it simpler. The Productivity Commission called the charity framework itself “confusing and burdensome,” spread across multiple Commonwealth and state regulators (Future Foundations for Giving, Report no. 104, May 2024). The same events get re-shaped again and again for different audiences, by hand, by the one person who knows where everything is.

It isn’t a discipline problem

Here’s the part that matters, because the sector keeps mis-diagnosing this as a people problem. The coordinator isn’t slow. She isn’t disorganised. She’s carrying context between systems that were never built to talk to each other. The donation lands in the payment rail. The donor sits in the CRM. The spend sits in the accounting ledger. The outcome sits in a case note, or in her head. Nothing joins those up on its own, so she does, every reporting cycle, from memory and spreadsheets.

That’s why “just buy better software” hasn’t fixed it, and it’s where most of the honest story about AI lives too.

The part the vendors won’t tell you

Your donor CRM has an AI feature now, and so does your accounting package and your email tool. Salesforce has Agentforce, Blackbaud has Copilot, Xero has Just Ask Xero, QuickBooks has Intuit Assist, Mailchimp and Bloomerang have their own. Some are genuinely useful. But nearly all are sealed in-product: they work inside their own four walls and don’t open a door for anything outside to reach in. On the current landscape, the number of those vendor AIs an outside agent can actually call is, roughly, none (software-landscape review, June 2026, checked API by API).

So the mission dollars you’re losing to admin don’t come back through the buttons your vendors are lighting up. They come back through a plainer door: the data connections your systems already expose. The reachable rails are the ones you’d expect — Microsoft 365, Stripe, Xero, QuickBooks Online, MYOB — and the donor CRMs with proper interfaces: Salesforce Nonprofit, Blackbaud Raiser’s Edge NXT, Neon, Raisely, Funraisin, plus Mailchimp for engagement. That’s where an agent can read what it needs and hand back a draft.

The further you move toward the Australian-built grants and case-management tools, the more the door narrows. SmartyGrants is read-and-export only. Tools like Bonterra Penelope, SupporterHub and GiveNow are effectively closed, no callable surface at all. You don’t wire an agent straight into those; you reach them side-on, through the CRM, finance and payment rails they already feed. It’s not glamorous. It’s just where the recovery actually is.

What actually changes on that desk

Picture the same nine o’clock, a year on.

The acquittal isn’t written from a blank page. An agent has already read the spend from the ledger, mapped it against the restricted-grant categories, and produced an exceptions list — the handful of lines that don’t sit cleanly where they should — so the coordinator reviews a same-day draft instead of building it from scratch over half a day. The impact report is drafted from the case notes and the CRM, every figure carrying the note it came from, so she’s editing a first cut rather than assembling one. The shared inbox has been triaged by intent and urgency, with values-aligned replies drafted for her to approve or bin.

Two rules make this safe rather than reckless.

First, a named person decides everything that touches a beneficiary or goes to a funder. The AI drafts, cross-matches and triages; the coordinator signs. Where a case note is thin, the draft comes back thin, with visible gaps marked for confirmation, not a confident paragraph invented to fill the silence. Every figure traces to its source. If something looks like a crisis disclosure in the inbox, it escalates to a human ahead of any auto-drafting, erring toward over-escalation. The point isn’t a charity that runs itself. It’s a coordinator who gets her hours back for the mission.

Second, the data stays yours. This is built inside your own tenancy, onshore in Australia, never fed to train someone else’s model, with role-based access. De-identification sits at the point of synthesis, where the outcome summary is written, rather than as a blanket gate — because thanking a donor by name or routing a participant’s enquiry needs the real name, and pretending otherwise just breaks the work. That matters, because about half of NFPs name security, privacy or data sovereignty as their top concern about AI, and only 14 per cent have an AI policy at all (Infoxchange, 2025). The answer to that isn’t a promise. It’s an architecture you can see.

Honest limits, and honest proof

Two things worth saying plainly.

The gains here are real but they’re shaped by the software, not by wishful thinking. As an illustration of the pattern, not a promise about your numbers: a funder impact report that took days to weeks drafted in about half the time; supporter enquiries that sat 24 to 48 hours triaged in minutes; a restricted-fund acquittal that ate half a day per grant returned as a same-day exceptions list; a grant round where the panel read 200-plus applications instead reading a scored shortlist of around 30. Those are illustrative benchmarks, re-baselined against your real numbers before anyone commits to them.

And on proof: Real Minds AI does not yet have an Australian own-client NFP result to show you. The eye-catching figures floating around this space — tens of thousands of hours saved at a Spanish foundation, working days handed back at a US nonprofit — are third-party and offshore, and we’ll only ever quote them labelled as exactly that. We’d rather tell you that than dress up someone else’s case study as ours. Your pilot is designed to become the first Australian proof, not to be measured against a borrowed one.

Give the coordinator the program back

The efficiency worth caring about here isn’t a leaner org chart. Nobody’s mission got bigger by cutting the one person who runs it. It’s the coordinator who doesn’t spend her evening turning last quarter into a document, because the draft was waiting for her to check instead of build. It’s the program room with the lights back on at the hour it’s meant to have them. It’s the dollar a donor gave for the work going to the work, instead of to the report about the work.

That’s the whole promise, and it’s a small, human one: stop making the mission’s money spend its time counting itself.

Where to start

You don’t need a transformation programme to find out whether this is real for you. You need one honest look at where a reporting cycle actually goes — which of your systems an agent can reach, which are closed, and how many of your coordinator’s hours are being spent proving the last dollar instead of earning the next outcome.

Two low-risk ways in:

• Read the map first. We’ve written a plain-English brief on the NFP software landscape — which of your tools an agent can genuinely reach, by what mechanism, and what “closed” is quietly costing you. → https://realmindsai.com.au/guides/nfp/
• Book a free 30-minute discovery call. We spend it walking one real path — an acquittal, an impact report, an inbox — to find the highest-value place to hand your hours back; any build that follows is quoted fixed-scope, on your numbers. → https://outlook.office.com/book/Consultation@realmindsai.au/?ismsaljsauthenabled

Somewhere tonight a coordinator is writing about the work instead of doing it, and the people the money was for are on the other side of the paperwork. The only question worth asking is how much of that desk you’re willing to hand back to the mission.