Thoughts on what makes AI strategies useful

A growing number of charities are commissioning AI strategies. That's a good thing. But we've noticed that "AI strategy" can mean very different things, and some versions are much more useful than others.

The version we see most often goes like this: a senior leader decides the organisation needs an AI strategy. A consultant is brought in. They produce a document. The document gets presented to the board, approved, and filed. The people who were supposed to change how they work either never read it or read it and shrugged. Six months later, very little has changed.

We've delivered strategies too, and we're not going to pretend we've always got it right. But we've started to form views about what separates the strategies that actually move an organisation forward from the ones that don't.

Strategy as a verb

The word "strategy" sounds like a noun. A thing you produce. A document. In practice, the useful ones behave more like a verb. They're a process of figuring things out, involving people, making decisions, and keeping those decisions alive as things change.

That means an AI strategy shouldn't just be written by consultants and presented to staff. The people who will use AI in their work need to be involved in shaping the strategy. Not just consulted at the end, but engaged from the start. What are the problems they're actually dealing with? Where are they wasting time? What do they think would help? What are they worried about?

That bottom-up input is essential because AI adoption is fundamentally a change management challenge (we wrote about that in August). A strategy that ignores how people feel about AI, what they're anxious about, and what their day-to-day constraints look like will produce a document that sounds right on paper and changes nothing in practice.

The straitjacket problem

AI strategies have a specific problem that other technology strategies don't. The tools change fast enough that a strategy written in January might need revising by June. A strategy that locks the organisation into specific tools or specific assumptions about what AI can and can't do becomes a straitjacket.

We've seen strategies that said "we will use Microsoft Copilot for X and ChatGPT for Y" and within months, both tools had changed significantly. Better strategies describe the problems the organisation is trying to solve and the principles for evaluating tools, rather than specifying which tools to use. They say "we need to process beneficiary feedback at scale" rather than "we will use Claude to summarise feedback forms." The former survives a model release. The latter doesn't.

This is uncomfortable for governance. Boards want certainty. They want to approve a plan and know what's happening. An AI strategy that says "we'll figure it out as we go" won't pass a board paper. But there's a middle ground: clear about the problems you're solving and the guardrails you're operating within, flexible about the specific tools and approaches. That gives the board something to approve while giving the delivery team room to adapt.

Involving the right people

The best AI strategies we've been involved with had senior leadership in the room for the framing sessions and operational staff in the room for the detail. Not one or the other. Both.

Leadership brings the authority to actually change things, the budget, and the organisational context. Operational staff bring the reality of what the work actually looks like, where the pain points are, and where AI might help or cause new problems. When either group is missing, the strategy is weaker. Leadership-only strategies tend to be ambitious but disconnected from reality. Staff-only strategies tend to be practical but too small, because they can't commit resources or change processes.

There's a fear dimension too. People are worried about AI. Worried about their jobs, worried about making mistakes, worried about looking stupid in front of colleagues who seem to get it. A strategy process that doesn't make space for those concerns will produce polite nodding in meetings and quiet resistance afterwards. The organisations making real progress are the ones where someone has said, honestly, "this is new for all of us and we're going to learn as we go."

What a useful strategy looks like

We're wary of prescribing a template because the whole point is that strategy should fit the organisation. But the useful ones tend to share some characteristics:

They start with problems, not technology. "Our fundraising team spends 30 hours a month on gift aid reconciliation" is a better starting point than "we should explore how AI can help with fundraising."

They're honest about where the organisation is. If your data is in twenty different spreadsheets with no consistent formatting, an AI strategy that assumes clean, integrated data is fiction.

They include actual experiments, not just recommendations. "We'll run a four-week pilot using AI to process supporter feedback" is better than "we recommend exploring AI for supporter feedback analysis."

Someone is responsible for making them happen. Not a committee. A person. With time allocated.

And they expect to be revised. An AI strategy written in late 2024 that hasn't been updated by mid-2025 probably isn't being used.

We're still learning what works. The charity sector is early in this, and so are we. But we've seen enough strategies gather dust to have strong opinions about what makes the difference.