The Hidden Costs of Your Nonprofit's AI Tools
- Mar 5
- 4 min read
A new Stripe feature is making it easier for software vendors to mark up AI costs and pass them on to customers, often invisibly. Here's what nonprofit leaders need to know before it affects their budget.
If your nonprofit uses any AI-powered software (a donor CRM, a grant writing assistant, a chatbot for constituents, or even a productivity tool like Notion), there's a cost happening behind the scenes that you may never see on an invoice. It's called a token cost, and a recent development from Stripe is about to make it both more common and more profitable for the vendors you pay every month.
Understanding what's changing, and why it matters for organizations operating on tight budgets, is exactly the kind of digital governance literacy that can protect your mission.
First: What Is a Token Cost?
When AI tools like ChatGPT, Google Gemini, or Claude process text (reading a donor email, drafting a grant section, answering a constituent question), they charge the software company for each small unit of language processed. Those units are called tokens.
Let's say you have Claude create a 200 word email for you. That's roughly 276 tokens. Want it to create a 800 word blog post? That will be another 1,067 tokens. That 10 page report you couldn't find time to write -- a whopping 6,667 tokens.
Every time an AI feature runs inside a tool you use, tokens are consumed. The software vendor pays OpenAI, Google, Anthropic, or another model provider for those tokens, and then decides how to pass that cost on to you.
"The more their customers use their agents, the more tokens they consume from the underlying model provider, making pricing and business model decisions especially critical." TechCrunch, on Stripe's new AI billing feature
What Stripe Just Changed
Stripe, the widely-used payment infrastructure company, recently released a preview of a new billing feature designed specifically for AI software vendors. It does two things that nonprofit leaders should pay attention to:
It automates token tracking. Vendors no longer have to build their own systems to measure how much AI their customers are using. Stripe does it for them, across multiple AI model providers.
It automates markup. A vendor can set a profit margin (say, 30%) on top of whatever the raw token cost is, and Stripe will apply it automatically. As Stripe described it: "Say you're building an AI app: you want a consistent 30% margin over raw LLM token costs across providers. Billing automates the process."
What this means in practice: vendors now have a simple, scalable way to turn your AI usage into a revenue stream rather than just a cost they absorb. And you may have no way of knowing it's happening.
Agentic AI: The Highest-Risk Category
The token cost problem is most acute with a category of AI tools called agents: software that takes actions autonomously on your behalf, rather than just answering a single question. Agents don't wait for a human to prompt them. They run multi-step processes, researching, drafting, sending, updating records, and looping back, consuming tokens at every step.
Many nonprofits are already using agentic tools, often without realizing that's what they are.
Agentic tools common in the nonprofit sector
Intercom Fin AI Agent
Handles constituent inquiries end-to-end without human involvement, escalating only when it can't resolve the issue.
Constituent Services
Zapier AI Agents
Automates multi-step workflows: a new donor triggers research, a thank-you draft, and a CRM update, all automatically.
Operations
Virtuous AI
Automates personalized donor outreach sequences based on behavior, deciding when and how to contact each donor.
Fundraising
Microsoft Copilot Studio
Nonprofits are building custom agents for benefits navigation, intake screening, and FAQ handling.
Program Delivery
Instrumentl
Monitors grant opportunities, matches them to your organization's profile, and flags deadlines automatically.
Grant Research
Make (Integromat)
Builds agentic workflows connecting multiple nonprofit tools without requiring technical staff to maintain them.
Operations
A nonprofit with 500 constituents running through an agentic intake workflow could burn through thousands of tokens a day, without any staff member initiating a single prompt. Under a usage-based pricing model with a vendor markup, that's an expense that arrives without warning.
The Compounding Problem: AI Tool Costs You Can't See It
Most nonprofit staff have no visibility into what data is flowing to AI providers or how much AI activity is happening inside the tools they use every day. Because the cost has historically been invisible, bundled into a flat subscription, the AI usage itself has been invisible too.
The Stripe development doesn't create this problem. But it accelerates it. As token billing becomes easier to implement, more vendors will implement it. And most will not advertise that they're doing so in their pricing pages.
This is a digital governance issue, not just a budget issue. It sits at the intersection of financial oversight, data privacy (constituent data flowing through AI models), and organizational accountability. These are three areas where nonprofits are particularly vulnerable and particularly scrutinized.
Questions to Ask Your Software Vendors Now
You don't need a technical background to start this conversation. You need the right questions.
If a vendor can't answer these questions clearly, that's important information about how much they've thought about their customers' data governance and how much you should rely on their tool for sensitive work.
What This Means for the Sector
Stripe's new feature is a business infrastructure story, but its downstream effects land squarely on organizations with the least financial cushion and the highest accountability to their communities. Nonprofits serving constituents who are already navigating systems of inequity cannot afford to absorb surprise technology costs, or to unknowingly expose the people they serve to data practices they didn't consent to.
The technology decisions being made right now, by vendors, by payment processors, by AI model providers, will shape what tools are available to community organizations and at what cost. Staying informed is the first step toward staying in control.
Need Help Navigating This?
ATL DataLab helps nonprofits and community organizations understand the technology systems shaping their work — translating complex developments like this into practical guidance you can act on. If you're unsure whether your current tools put your budget or your constituents' data at risk, we can help you find out.
Comments