How charities can leapfrog digital transformation with AI

Here is the conventional wisdom: before you can do anything interesting with AI, you need to fix your foundations. Clean your CRM. Digitise your paper processes. Standardise your data. Get everyone onto one system. Do the painful, multi-year digital programme. Then, and only then, think about AI.

We think that advice is mostly wrong.

Foundations do matter. But the economics of building digital solutions have changed so dramatically that the traditional sequence no longer holds. Charities that skipped the digital wave don't necessarily need to go back and swim through it. Some of them can jump straight over.

The mobile phone principle

In the 1990s and 2000s, developing countries faced a familiar problem: they had no landline infrastructure. The conventional wisdom said you needed to build the phone network first, then build services on top of it. Instead, mobile phones arrived and entire countries skipped the landline phase entirely. Banking, communication, government services -- all built on mobile, without ever building the intermediate step.

The analogy to charities isn't perfect. But the principle is sound: when a new technology changes the cost equation, the old sequence of steps doesn't always apply.

For charities, what's changed is how computers work. Until recently, computers were rigid. They dealt in ones and zeros, and they needed data formatted in exactly the right way. That rigidity is why digital transformation was so painful. To get a computer system to work for your organisation, you had to reorganise your organisation to work for the computer. Standardise your data. Get everyone onto one platform. Enforce consistency across teams. Digital transformation required a kind of totality and hegemony in order to function, and neither of those things sit well within charities, where teams are small, missions are specific, and one-size-fits-all has never fitted anyone.

Large language models changed this. They can work with messy data, inconsistent formats, and natural language. They don't need your spreadsheet to be perfectly structured. Agentic AI goes further: it can do things for people using computers, where previously people had to help computers by doing awful things like copying data from one spreadsheet to another. The flexibility is genuine and it changes the economics of building digital tools.

Where something that would solve a specific problem for your charity might previously have cost half a million pounds, it might now cost £50,000. We've dropped an order of magnitude. That is a massive shift compared to digital transformation, where costs were exorbitant and projects were multi-year behemoths. We can now have quick planning and building cycles that actually create value for organisations.

You can see the market catching up with this. The crash in valuations of narrow SaaS companies like Figma and Asana in early 2026 reflects a dawning realisation: it's now relatively straightforward to build high-quality, bespoke tools that match what an organisation actually needs. The era of buying an enterprise product and spending six months configuring it to roughly fit is ending.

Why this matters more for charities than anyone

Digital transformation was particularly ill-suited to charities because charities aren't factories. They don't need totality. They need nimble, small products that work for specific teams within their specific context, at the time they need them. A housing charity's service delivery team has different needs from its fundraising team. A hospice needs different workflows from a campaigning organisation. Off-the-shelf software forced everyone into the same moulds. The bespoke alternative was too expensive to consider.

That's no longer the case. The cost of building something tailored to how your charity actually works has dropped to the point where it's often cheaper than buying and configuring an enterprise product. Not for everything. Many challenges still require complicated software. But the range of problems where a targeted, bespoke solution makes economic sense has expanded enormously.

Where this breaks down

This approach has real limits and we should be upfront about them.

Data quality still matters. If your underlying data is wrong, a targeted AI tool will produce wrong results faster. Our post on data quality as the real AI strategy still holds. You don't need pristine data across your whole organisation, but you need good-enough data for the specific problem you're solving.

Some problems genuinely need foundations first. If your charity literally has no way to record beneficiary interactions digitally, you need a basic system before you can build on top of it. The leapfrog works when you have digital data in some form, even messy form, not when you're starting from paper.

Targeted solutions can become their own kind of mess. Build ten separate tools without thinking about how they connect and you've created a different problem. The discipline is to solve specific problems while keeping an eye on how the pieces fit together.

And this isn't free. "Cheaper" doesn't mean "no cost." Building targeted AI solutions still requires someone with the skill to build them, or a partner who can. The economics are better than they were. They're not zero.

Too many charities are stuck because they believe they need to complete a digital foundation programme before they're allowed to use AI. They look at their messy CRM and their scattered spreadsheets and conclude they're not ready.

The charities that skipped the digital wave aren't as far behind as they think. The cost of catching up just dropped by an order of magnitude, and the new tools don't demand the rigid, organisation-wide conformity that made digital transformation so painful. The question isn't "have we done enough foundation work?" It's "what's the one problem we'd most like to solve?"