The real cost of AI for charities - and why custom is cheaper than you think

The AI industry's pitch to organisations is simple: AI saves money. And it can. But the claim is often made without accounting for the full picture, and the full picture includes costs that AI vendors prefer not to mention.

We've worked with enough charities on AI implementation to have a reasonable view of what AI actually costs, not in the abstract, but in the day-to-day reality of charity operations. Here's what we've seen.

The visible costs

Subscriptions are the costs you can see on an invoice.

• Microsoft Copilot: roughly £22/user/month with nonprofit pricing. For 50 users, that's around £13,000/year.
• ChatGPT Plus: roughly £16/user/month. For 10 heavy users, around £2,000/year.
• Google Workspace with Gemini: free for nonprofits (basic tier), or from roughly £3/user/month for premium features with 75% nonprofit discount.
• Canva Pro: free through Canva for Nonprofits.

API costs come into play if you're building automated workflows (processing feedback forms, generating reports from data, running batch analyses), you pay per use. API costs vary by model and volume. For modest use, this might be £20-50/month. For high-volume processing, it can be hundreds.

The invisible costs

This is where the honest accounting gets uncomfortable.

Staff time for implementation. Getting AI tools working in your organisation takes time. Someone has to evaluate options, set up accounts, configure settings, write usage guidelines, and troubleshoot problems. For a medium-sized charity, budget 40-80 hours for initial setup and rollout. That's one to two weeks of someone's time.

Training. AI tools are easy to start using and hard to use well. The difference between a team member who pastes text into ChatGPT and one who knows how to write effective prompts, choose the right tool for the task, and critically evaluate outputs is significant. Budget at least a half-day per team member for initial training, and ongoing time for support. We run these sessions for charity clients and they make a measurable difference to adoption quality.

Reviewing AI output. Every piece of AI-generated work needs human review. A grant application drafted by AI still needs someone to check accuracy, add organisational context, and ensure it actually answers the funder's questions. A data analysis still needs someone who understands the data to verify the conclusions. This review time is real and ongoing. It doesn't appear on any AI vendor's ROI calculation.

Data preparation. AI works better with good data. Most charity data isn't good. CRM records with inconsistent formatting, spreadsheets with merged cells and colour coding, programme data scattered across platforms. Cleaning this up enough for AI to work with it takes time and often money. It's the most valuable investment you can make, and the least visible.

Policy development. Your AI policy won't write itself (well, AI could draft it, but humans need to think it through). Developing governance that covers data protection, acceptable use, risk management, and accountability takes time from senior staff and trustees.

Error correction. AI makes mistakes. When those mistakes make it into a published report, a donor communication, or a board paper, someone has to find and fix them. The cost of errors is unpredictable but real.

When AI saves money

AI produces genuine savings when the task meets these conditions: it's high-volume, it's repetitive, the inputs are well-defined, and acceptable accuracy is less than 100%.

Processing 500 feedback forms into themed analysis. Categorising a year of transactions. Summarising 50 grant reports for a board paper. Drafting 200 personalised donor acknowledgments. These are tasks where AI can do in hours what previously took days.

We've seen charities reclaim 20+ hours a month from a single well-chosen AI application. At staff cost rates, that's a meaningful saving.

When AI is cost-neutral

Many AI applications shift work rather than eliminating it. You spend less time drafting but more time reviewing. You spend less time searching for information but more time evaluating what AI finds. The total time might be similar, but the nature of the work changes.

This isn't necessarily bad. If AI handles the tedious parts and your staff focus on the parts that require judgement, that's a better use of everyone's time even if the hours are the same.

When AI costs more

This happens more often than anyone admits. If the task is low-volume (writing one report a month), the setup and learning time exceeds the time saved. If the task requires high accuracy (financial reporting, safeguarding records), the review time can exceed the drafting time. If your data is in poor shape, you spend more time cleaning it than the AI saves.

The worst case: buying expensive AI subscriptions, spending weeks on setup and training, and then watching adoption fizzle because the tools don't solve problems people actually have. The AI Playbook's "buy, build, assemble, or wait" framework is useful here. The worst option, as we note there, is usually buying something expensive and then not using it, which is what happened to several charities that invested in Copilot licences before their internal data was ready to support it.

For any AI tool you're considering, estimate the full cost: subscription, setup time, training time, and ongoing review time. Compare against doing the task manually. If AI isn't clearly cheaper when you include everything, reconsider. Start with free tools: ChatGPT's free tier, Claude's free tier, and Gemini through Google for Nonprofits are all capable enough for most charity use cases. Don't pay for AI until you've established that free tools can't do the job. And if you adopt a tool, track the actual time saved over three months. Not the theoretical saving. The actual, measured hours reclaimed.

The economics of build vs subscribe

There's one more cost comparison most charities don't consider: building something custom versus subscribing to something generic.

The conventional assumption is that custom development is expensive and subscriptions are cheap. That used to be broadly true. It's become less true as AI tooling has made development faster and cheaper. A bespoke solution that solves your specific problem (a feedback analysis pipeline built around your data, an automated reporting workflow that pulls from your actual platforms) might cost less than a year of per-seat licences for a generic tool that doesn't quite fit.

We've seen charities spend £12,000/year on 50 Copilot licences where a purpose-built solution costing a fraction of that would have solved the actual problem. The maths is more interesting than charities might expect, and it's worth running before defaulting to the subscription model. When is a one-off build cheaper than an ongoing subscription? More often than you'd think, and increasingly so as the cost of development continues to drop.

We build AI solutions for charities, so we have an incentive to be optimistic about custom development. We're being honest anyway: sometimes the subscription is the right answer, sometimes building is, and sometimes the answer is free tools with a thoughtful workflow. The charities that come out ahead are the ones that compare all three options with clear eyes.