The limits of Copilot and Power Apps: when off-the-shelf AI isn't enough

Most charities in the Microsoft ecosystem have a version of this conversation at some point. Someone suggests AI could help. Someone else says "we already have Copilot" or "can't we build that in Power Apps?" And often enough, the answer is yes. Copilot summarises the meeting. Power Apps collects the form data. The job gets done.

This post is about the jobs that don't get done. The ones where charities try Copilot or Power Apps, hit a wall, and either give up on the idea or assume AI isn't ready for what they need. In most cases, AI is ready. The tool just isn't.

We should be clear: this is not an argument against Microsoft. Copilot does useful things. Power Apps solves real problems. If your charity is using them and getting value, keep going. The argument is about recognising where those tools stop, and understanding that the work beyond that boundary is often the work that matters most.

What Copilot is good at

Copilot handles personal productivity well. Meeting summaries in Teams are potentially useful, and for charities running dozens of meetings a week, they save real time. Email drafting and thread summarisation in Outlook works. Document search across SharePoint surfaces things that would otherwise take 20 minutes to find. We wrote about this when Copilot launched with nonprofit pricing last year, and the assessment hasn't changed much. For individual tasks where one person needs to process, draft, or find something within Microsoft's tools, Copilot is worth the cost.

Power Apps fills a different gap. If you need a simple data collection form, an approval workflow, or a basic internal tool that connects to your existing Microsoft data, it can do that without hiring a developer. For charities that previously relied on spreadsheets passed around by email, that's a genuine step forward.

Where charities hit the ceiling

The problems start when you move beyond individual productivity and simple forms.

Copilot is a one-to-one relationship between a person and their data. It helps you with your emails, your documents, your meeting notes. It doesn't help your organisation see patterns across departments. A fundraising team using Copilot and a service delivery team using Copilot are still operating in silos. Copilot doesn't connect their work. It doesn't notice that the supporters writing to the fundraising team are raising the same concerns that service users are reporting through a different channel. That kind of cross-team insight requires something that can see across the whole organisation, not just one person's inbox.

Data boundaries matter too. Copilot works with data inside the Microsoft ecosystem. If your charity uses Salesforce as a CRM, runs programmes through a case management system that isn't Microsoft, or collects beneficiary feedback through an external survey tool, Copilot can't touch that data. The most valuable analysis often involves combining data from multiple systems, and that's exactly what Copilot can't do.

Scale is another issue. Processing five meeting transcripts is a Copilot job. Processing 2,000 beneficiary feedback forms into a themed analysis with outcome categories mapped to your theory of change is not. Copilot wasn't designed for batch processing at scale. It's a personal assistant, not a data pipeline.

Power Apps hits similar walls from a different direction. It works for straightforward logic: if this, then that. When the logic gets complex, when you need conditional routing based on multiple factors, integration with external systems, or processing that involves judgement rather than rules, Power Apps becomes painful to build in and fragile to maintain. We've seen charities spend months building Power Apps solutions that a purpose-built tool could handle more reliably in weeks.

Neither tool does consumer-facing work. If your charity needs a beneficiary-facing service, a supporter portal, or a public-facing tool, you're outside what Copilot and Power Apps were designed for.

There's also a structural issue. Microsoft's tools are built around permissions hierarchies. Who can see what, who can edit what, who can approve what. That's appropriate for governed document management. It's less helpful when you're trying to enable fluid collaboration, rapid prototyping, or the kind of cross-functional working that good AI adoption requires.

The interesting middle ground

The useful framing isn't "Copilot vs custom" as a binary choice. It's understanding that different problems need different tools.

Use Copilot for the everyday governed tasks where it works: meeting summaries, email management, document search, first-draft writing. These are useful, and the fact that they happen inside tools your team already uses means adoption is straightforward.

But recognise that the problems sitting above that ceiling are often the ones with the biggest organisational impact. Processing thousands of feedback forms. Building a grant-matching pipeline against external databases. Automating a reporting workflow that pulls from your CRM, your finance system, and your programme data. Creating a tool that your beneficiaries or supporters actually interact with. These need something designed for the specific problem, working with your specific data, built around your specific processes.

The economics of this have shifted. Custom development used to mean six-figure budgets and 18-month timelines, which ruled it out for most charities. AI-assisted development has changed that equation significantly. A bespoke solution that solves a specific organisational problem can cost less than a year of Copilot licences spread across 50 users, and it solves the problem that Copilot can't.

One detail that says something about the state of play: reports from within Microsoft suggest that not all of its own engineering teams use Copilot regularly. If the people who built it find it situationally useful rather than indispensable, charities should feel comfortable reaching the same conclusion.

If your charity is paying for Copilot, audit what's actually being used. Meeting summaries and email features tend to stick. The rest often doesn't. That audit will tell you where Copilot is earning its cost and where you're paying for features that don't match how your team works. For the problems where Copilot and Power Apps aren't enough, the question isn't whether AI can help. It almost certainly can. The question is whether you need something built for the purpose, and increasingly, the answer is yes.