It’s more complicated than that.

I’ve watched this play out across multiple initiatives and organisations. Even when the tools are free and the training is available, most people don’t engage. Cost and access matter, but they’re not the whole story. Not even close.

So what’s missing?

I keep arriving at the same answer: it’s culture. Whether an organisation has created the conditions for people to actually use these tools, talk about them openly, experiment without fear, and keep showing up to the conversation week after week.

Without that, everything else struggles to stick.

What I keep seeing at conferences

I spoke at one of the largest fundraising conferences in Australia a few months ago. I taught people about projects in ChatGPT, voice mode, using AI as a strategic thought partner. The room was electric. It was voted one of the best sessions at the conference.

And I walked out feeling confused.

Because none of what I showed them was new. Projects have been around for over a year. Voice mode even longer. These weren’t cutting-edge features I was demoing. They were table stakes. The kind of thing people should have been comfortable with twelve months ago.

I sat in on another AI workshop at a different conference a few weeks back. Same material. Prompting, projects, voice. Same reaction from the room. Genuine amazement. And the same quiet frustration on my end, because these organisations should not be learning this in the middle of 2026.

The gap between what’s possible and what most nonprofits know is possible isn’t just wide. It’s accelerating. Every month these organisations wait, they fall further behind. Not only because the technology is moving too fast, but because nobody inside their organisation is creating the conditions for people to keep up.

The leadership problem nobody wants to name

At that conference, someone asked me a question I’ve heard versions of a dozen times now: “What do we do when our leadership isn’t prioritising this?”

I didn’t have a good answer. I still don’t.

You can suggest they talk to the board. You can point out that funders are starting to ask about AI capability. You can make the case that a newer, leaner organisation could emerge tomorrow and do everything they do with a fraction of the funding and a fraction of the staff, and that funders will notice.

But if leadership doesn’t prioritise AI, the organisation stays stagnant. One or two passionate individuals might experiment in their own time. They become shadow AI users, getting enormous value while their colleagues don’t even know what voice mode is. And the gap inside the organisation mirrors the gap outside it.

I had a conversation recently with someone who’d never heard of voice mode. Not because they’re not smart or curious. Because their organisation had never created the space for them to discover it. I’d take that as a failure of the organisation, not the individual. Voice mode is a feature. How you use it has to be figured out by each person. But if you don’t even know it exists, you can’t walk down that path.

This is what I mean when I say culture is the problem. It’s not about whether individuals are interested. Most are. It’s about whether the organisation has made AI visible enough, safe enough, and normal enough that people can actually engage with it.

What worked

I worked with one organisation for about eight months. Weekly meetings, an hour or so each time. We went through every component of what I think of as an AI strategy: policy, tool selection, surveys, training, a maturity model, and the culture practices to hold it all together.

What made them different wasn’t budget or technical skill. It was commitment. Leadership decided this mattered, blocked the time, and showed up every week. The team did the work between sessions. They took the maturity model I’d developed, which was deliberately broad, and customised it to their own tools, their own goals, their own strategy. They created something that told each person exactly where they were and what progression looked like for them specifically.

By the end, they didn’t need me anymore. That’s the goal. They had the frameworks, the shared language, the habits, and the confidence to keep going on their own.

I’ve since worked with another organisation on a shorter engagement, focused mainly on policy and tool usage. Different scope, different needs, same principle. Each organisation needs to find their own version of this, but the underlying structure is remarkably consistent.

The components that matter

I don’t think culture is something you build with a single initiative. It’s a set of small, recurring practices that keep AI visible and conversational across the organisation. Here’s what I’ve seen work.

A standing agenda item. Five minutes at the end of every team meeting. Someone shares something they tried with AI that week. It doesn’t have to be positive. “I tried using it for grant research and it hallucinated three funding bodies that don’t exist” is just as valuable as a success story. Maybe more, because someone else in the room might have a suggestion, or you note it and revisit it in a few months when the models have improved. The value is the conversation, not the outcome.

A shared channel. Slack, Teams, whatever you use. Dedicated to AI use cases, questions, wins, and failures. It creates ambient visibility. People see what their colleagues are experimenting with even when they’re not actively seeking it out.

Regular policy review. I’ve written about this before, but the AI policy isn’t just a governance document. It’s a conversation starter that keeps regenerating. Every couple of months, you come back to it: is there anything we need to update? Any new tools people want to explore? Any use cases that push against the current guardrails? That review process forces the conversation to keep happening.

The policy hackathon. Take your AI policy, the document everyone was supposed to read, and give teams an hour to transform it into something else using AI. A song. A video. An infographic. A web app. I keep recommending this because it does three things at once: people actually engage with the policy content, they build practical AI skills, and they have to proof the output for hallucinations. One organisation I spoke with had tools monitoring how long staff spent reading their policies. The average was fifteen seconds. You can’t understand a policy in fifteen seconds. But you can understand it when you’ve spent an hour turning it into a podcast.

AI champions. Not everyone needs to be an expert. But you need people inside the organisation who are driving this forward, keeping the audits happening, surfacing new features in existing tools, and holding colleagues accountable in a supportive way. These might be people already doing it informally. Give them formal permission and time.

None of these practices require budget. None require technical expertise. They require a decision from leadership that this matters, and the discipline to keep showing up to it.

Why information isn’t enough

I want to be direct about something I’ve been circling for a while in my own thinking. I used to believe that the main barrier to AI adoption was access to information and tools. That if we could just educate people and give them access, adoption would follow.

I don’t believe that anymore.

Information is abundant. It’s essentially free. There are courses, tutorials, webinars, communities, entire platforms dedicated to teaching AI skills. I run one myself. And yet the foundation experiment showed me sixty organisations with free access and free resources, and almost none of them engaged.

The missing piece was never information. It was never cost. It was culture. It was whether someone inside those organisations was making this a priority, keeping the conversation alive, creating space for experimentation, and building the kind of environment where people felt safe to try, fail, share, and try again.

In systems thinking terms, information and tools are the most visible parts of the system, which is exactly why organisations default to them. They’re tangible. You can point to a subscription, a training session, a resource library. But the leverage point is deeper. It’s in the norms, the habits, the recurring rhythms of how an organisation talks about and engages with change. Shift those, and adoption follows. Leave them untouched, and no amount of free tools will make a difference.

The cost of waiting

I keep coming back to the conferences. Rooms full of smart, mission-driven people, genuinely amazed by features that have been available for a year or more. The enthusiasm is real. The gap is real too.

Every month an organisation doesn’t build these cultural practices, the distance between where they are and where they could be grows. Not because the technology is inherently hard, but because fluency compounds. The organisation that starts having these conversations now, even imperfectly, even for five minutes a week, will be in a fundamentally different position twelve months from now. They’ll have shared language, shared confidence, and a track record of small experiments that build on each other.

The one that waits will still be amazed by voice mode at next year’s conference.

I know that sounds harsh. I don’t mean it to be. The people in those rooms are doing important, difficult work with too few resources. But that’s precisely why this matters. The organisations doing the hardest work with the fewest people are the ones that stand to gain the most from building this into their culture. And they’re the ones that can least afford to wait.

If you don’t know where to start, start with the smallest thing. Add five minutes to your next team meeting. Ask people what they’ve tried. Listen to what comes back. That’s culture. Everything else builds from there.

Kyle Behrend is an AI, Automations & Systems Specialist with 15 years in the nonprofit sector, based on the Central Coast of NSW, Australia. He is the founder of AI Impact Hub and writes The Leverage Point on Substack — exploring how small shifts create transformative change for mission-driven organisations.

P.S. If you’re looking for a maturity model to help structure your team’s progression, I have a free one available on my website. It’s designed to be customised to your organisation’s tools and goals, not used as-is.

So I asked the obvious next question: what are they doing with that time?

The answer was a long pause, followed by something like, “I mean, they’re just getting through more of their to-do list.”

This is the trap. And almost every organisation I talk to falls into it.

Someone saves two hours on a task that used to take an afternoon. Great. So now they spend those two hours on the next pile of admin. The workload stays the same. The pace stays the same. The only thing that changed is the tool. And three months later, leadership wonders why morale hasn’t shifted and nobody seems particularly excited about AI anymore.

The reward for doing more work in less time should not be more work.

I know that sounds obvious when you read it on a screen. But in practice, it’s the default in almost every organisation I’ve worked with. Not because anyone decided it should be that way. Because nobody decided it shouldn’t be.

What happens with saved time is a leadership decision

This is one of those areas where the absence of a decision is itself a decision. If leadership doesn’t explicitly define what saved time gets used for, it gets absorbed. It fills back up with whatever was next on the list. Emails. Reporting. Data entry. The kind of work that expands to fill whatever time is available to it.

I think there are roughly three categories of reinvestment worth considering, and the best organisations I’ve seen do all three in some proportion.

The first is experimentation. This is probably the most important, and the hardest to protect. Once people have some fluency with AI tools, they start noticing opportunities that nobody planned for. A better way to onboard volunteers. A dashboard that could tell a story about program impact. A process that’s been manual for years but doesn’t need to be. Those ideas only surface when people have the time and permission to explore. Without protected time, they stay in someone’s head, filed under “when things slow down.” Things never slow down.

The second is human connection work. This is the one I think gets overlooked most often in the AI conversation. Consider a fundraiser who’s been spending four hours a week drafting donor emails. AI brings that down to one. The question is whether those three hours go back into the admin pile or into picking up the phone and actually talking to donors. Building relationships. Having the conversations that no AI tool can replicate. In a sector built on trust and connection, the ability to reinvest efficiency gains into more human work is genuinely transformative. Not in the buzzy, marketing sense of that word. In the practical sense that donors who get phone calls give differently to donors who get emails.

The third is the work that was never a priority. Every nonprofit I’ve ever worked with has a backlog. Projects that matter but aren’t urgent. The onboarding guide that’s been on someone’s list for three years. The impact report that could unlock new funding if someone had the time to do it properly. The volunteer training materials that everyone knows are outdated but nobody has capacity to fix. That backlog is a graveyard of good intentions, and AI-driven efficiency is the first real opportunity most organisations have had to actually work through it.

The backlog reframe

I keep hearing a version of this concern in conversations: if AI makes us more efficient, does that mean we need fewer people?

It’s a fair question, and I don’t think dismissing it helps anyone. But I also think it’s the wrong framing for most nonprofits.

Think about it this way. What nonprofit organisation is saying, “I don’t have enough things for people to do”? They’re all saying the opposite. We need more people. We need more resources. We need more hours in the day. The backlog isn’t a nice-to-have list. It’s full of work that would genuinely advance the mission if anyone had the capacity to do it.

AI doesn’t remove the need for people. It removes the excuse for not getting to the work that matters most. But only if someone makes the decision to direct that capacity toward it.

Will AI impact hiring and how many people organisations need over time? Almost certainly. But that’s true across every sector, not just ours. And for most nonprofits, the more immediate reality is that they have more work than they can handle and a team that’s stretched thin. The efficiency gains from AI aren’t a threat to those teams. They’re a release valve, if leadership treats them that way.

The start-up thought experiment

One exercise I’ve found useful with organisations is deceptively simple. Ask yourself: if we were founding this organisation today, from scratch, with the tools that exist right now, how would we do things differently?

I was talking to someone about a European nonprofit that did something close to this. They had a team of around 30 people, and they pulled four of them out and said, see if you can rebuild our whole organisation just using AI. Not as a threat. As a genuine experiment. What would we look like if we started fresh?

I love that, because it forces you out of the incremental mindset. Most AI adoption conversations are about taking existing processes and making them slightly faster. This is a different question entirely. It’s asking whether the processes themselves still make sense.

We’ve all been doing things the same way because that’s how they’ve always been done. And I think there’s an opportunity, with the time that AI creates, to step back and ask whether “how we’ve always done it” is still the right answer. Usually, it’s not. But you never get to that question if every saved hour just gets swallowed by the inbox.

This is a leadership conversation, not a technology one

I want to be direct about something. None of what I’ve described here requires any technical knowledge. It doesn’t require a new tool or a bigger budget. It requires one decision from leadership: we are going to be intentional about what happens with the time AI gives us back.

That decision is the leverage point. Without it, your team saves time and nothing changes. The work just compresses. People might even burn out faster, because now there’s an expectation that everything should take less time, but nobody reduced the volume.

With it, you create space for the work that actually moves your mission forward. The relationships that need building. The ideas that need exploring. The projects that have been waiting years for someone to have the capacity to take them on.

Your team is going to get faster with AI. That’s almost inevitable at this point. The question that matters is what your organisation does with the difference. And that question belongs to leadership, not to the tools.

Kyle Behrend is an AI, Automations & Systems Specialist with 15 years in the nonprofit sector, based on the Central Coast of NSW, Australia. He is the founder of AI Impact Hub, and writes The Leverage Point on Substack — exploring how small shifts create transformative change for mission-driven organisations.

P.S. Try this at your next team meeting: ask each person to name one task where AI has saved them time, and then ask what they did with that time. The answers will tell you whether your organisation is capturing the value or just compressing the work.

A new AI app drops, someone mentions it in a group chat or a webinar, and I’m already signing up. Sometimes I even pay for it. I tell myself it’s research. And then, almost without fail, I find myself drifting back to the same two or three tools I use every day. The ones that actually do the work.

I mention this because I think it reveals something about how most of us, myself included, approach AI. We get drawn to the new thing. The next tool, the next feature, the next capability. And in that rush, we often miss the more important question: what kind of value are we actually chasing?

Because in my experience, AI offers nonprofits two fundamentally different kinds of value. And most organisations only ever see one of them.

The first bracket: efficiency

This is where almost everyone starts, and for good reason. You’ve got tasks that eat hours every week. Drafting emails, summarising meetings, writing thank you letters, pulling together board reports. The kind of work that’s necessary but not exactly what gets people out of bed in the morning.

AI can compress a lot of that. Voice dictation alone is worth pausing on. We speak roughly three times faster than we type. If you’re spending two hours a day on emails and you switch to speaking them, you’ve just found time that didn’t exist before. Not theoretically. Actually.

There’s a useful way to think about which tasks to start with. Map them on two axes: how often you do them, and how much time each one takes. The sweet spot is in the corner where frequency and time are both high. Emails, meeting notes, reports, donor acknowledgements. These aren’t glamorous wins. They’re not going to make anyone’s conference highlight reel. But they are the tasks where the return is immediate and measurable.

I ran a webinar recently where I asked people to share in the chat which AI features they were already using in their existing tools. The answers were telling. Rewriting emails for different literacy levels. Converting paper records to digital databases. Using project management tools that auto-generate subtasks from a description. One person mentioned that the AI in their HR platform had become genuinely useful for surfacing policy information that nobody ever read in document form.

None of that is flashy. All of it is valuable. And most of it was happening inside tools these organisations were already paying for.

The second bracket: imagination

Here’s where it gets interesting. And where most nonprofits stop short.

The efficiency bracket is about doing existing work faster. The imagination bracket is about doing work that was never possible before. Not because the ideas weren’t there, but because the cost of execution was prohibitive.

I was on a call with a funder recently, walking through some basics. Partway through, I pulled up their website and started rebuilding it. Not a mockup. The actual site, functional, in real time, while we were chatting. When I said we could go live with it in a couple of minutes, the silence on the other end told me everything.

That’s not an efficiency gain. That’s a category shift. The gap between having an idea and having a working version of it has collapsed. Time from ideation to something usable is now measured in minutes, not months.

I know a small nonprofit overseas that couldn’t find a CRM that suited their workflows. Nothing on the market quite fit the way they operated. So they built one. Using AI to write the code. When I spoke to them, they hadn’t even considered that they could potentially sell it to similar organisations in their sector as a revenue stream. The possibility hadn’t occurred to them, because six months earlier, building software wasn’t something organisations like theirs did.

This is what I mean by imagination. It’s not about doing the same things with less effort. It’s about looking at problems you’ve walked past for years, the ones that felt too expensive or too complex or too far outside your capabilities, and realising the barrier has dropped.

The fundraising series nobody had time to write. The onboarding guide that’s been on someone’s list for three years. The campaign landing page you couldn’t justify paying a developer to build. The data that could tell a powerful story if anyone had the hours to shape it.

Why most organisations get stuck in bracket one

There’s a natural gravity toward efficiency. It’s concrete. It’s measurable. You can point to hours saved and feel confident something happened. Leadership can report on it. It fits neatly into existing ways of thinking about productivity.

The imagination bracket is harder to justify because it requires a different mindset entirely. You have to stop thinking about what you currently do and start thinking about what you could do. And that means stepping outside the constraints that have defined how your organisation operates, probably for years.

I see this all the time. Organisations that are genuinely under-resourced, running on tight budgets with stretched teams, and they approach AI as a way to make the current situation slightly less painful. Which it can do. But they never get to the point of asking the bigger question: if we could do anything, what would we build?

That question feels indulgent when you’re drowning in operational demands. It’s not. It might be the most strategic question your organisation can ask right now.

The rabbit hole, though

I’d be dishonest if I didn’t flag the other side of this. Because the imagination bracket has its own trap, and I fall into it regularly.

A few weeks ago, some friends in a group chat were arguing about which coding tool was better. Harmless enough. Somehow this turned into me using AI image generation to create Mortal Kombat-style fighting game portraits of their faces. One of them told me later he’d spent most of that evening turning himself into a full character instead of doing the work he was supposed to do.

It’s genuinely easy to get pulled down the rabbit hole. Every new capability is interesting. Every new feature invites experimentation. And before you know it, you’ve spent three hours on something that was fun but produced nothing of value for your mission.

The discipline isn’t in avoiding the imagination bracket. It’s in being intentional about which problems you point it at.

A practical way to hold both

One exercise I’ve been recommending to organisations lately is dead simple, but I think it captures the right way to approach this.

Create two documents. The first is an efficiency register. For each team member, list the repetitive tasks they do regularly and estimate how much time AI could save on each one. Emails, reports, meeting summaries, data entry, whatever it is. Be specific. This becomes your baseline for tracking actual time savings over the coming months.

The second document is an imagination list. And this one requires a different conversation entirely. Ask your team: if our organisation could do anything, with no constraints on time, budget, or technical capability, what would we build? What would we try? What problems have we walked past because solving them felt impossible?

Write everything down. Don’t filter it. Don’t assess feasibility yet. Just capture what people wish they could do.

Then look at that list through the lens of what’s now actually possible. You’ll be surprised how many of those items have moved from fantasy to feasible. Not all of them. But enough to change what your next twelve months look like.

The leverage point

Most organisations approach AI as a productivity tool. Get things done faster, free up some hours, maybe reduce a bit of the operational burden. That’s real value, and I don’t want to diminish it. For teams that are genuinely stretched, even saving a few hours a week can be the difference between burnout and breathing room.

But if that’s where the conversation stops, you’re capturing half the value at best. The other half lives in the imagination bracket, in the work that becomes possible once your team starts seeing AI not just as a faster way to do what they’ve always done, but as a way to do what they never could.

The efficiency register will save you time. The imagination list might change what your organisation becomes.

Both documents take an hour. Start this week.

Kyle Behrend is an AI, Automations & Systems Specialist with 15 years in the nonprofit sector, based on the Central Coast of NSW, Australia. He is the founder of AI Impact Hub and writes The Leverage Point on Substack — exploring how small shifts create transformative change for mission-driven organisations.

So they do what most organisations do next. They start planning training.

It’s a sensible instinct. And it’s exactly where most organisations go wrong.

Not because training isn’t important. It is. But because training without strategy is just an event. People show up, they nod along, they learn a few things. Three weeks later, nothing has changed. Leadership wonders why adoption isn’t sticking. And somebody starts googling the next workshop.

I’ve watched this pattern play out for three years now. The organisations that get value from training aren’t the ones that book the best facilitator. They’re the ones that do the unglamorous work beforehand — the work that makes training land.

Training is step four, not step one

In the last two pieces, I’ve talked about policy and assessment. There’s a reason they come first, and it’s not because I like a tidy sequence. It’s because training doesn’t work without them.

Think about what happens when you skip straight to a workshop. You’ve got a room full of people at wildly different capability levels. Some have been experimenting with AI for months. Some tried ChatGPT once and found it underwhelming. Some are quietly terrified. And you’re pitching to the middle, hoping something sticks for enough of them to call it a success.

Meanwhile, you might not have agreed on which tools people should actually be using. You haven’t defined what good looks like at different levels. You haven’t established a shared vocabulary — and this one matters more than people think. I see it constantly: someone calls something a “project,” someone else calls it a “folder,” people throw around the word “agents” when they mean something else entirely. If your team can’t talk about AI in a common language, they can’t learn about it together in any meaningful way.

And underneath all of that, there may not even be a policy that tells people what’s okay to do and what isn’t. Which means even if someone walks out of training excited and ready to experiment, they hit a wall of uncertainty the moment they sit back down at their desk.

Policy. Assessment. Tools. Then training. Not because it’s a rigid framework, but because each step creates the conditions for the next one to actually work.

The problem with generic

Here’s the thing I keep coming back to: information is not the problem anymore. If someone wants to learn how to prompt, they can go to YouTube. There are courses, tutorials, webinars, entire academies dedicated to teaching AI fundamentals. The information is abundant and mostly free. You can even access a whole range of these on the AI Impact Hub.

So when an organisation spends money bringing someone in for training, the question should be: what are we getting that people can’t get on their own?

The answer has to be specificity. Training built around the actual tools your organisation uses, the actual tasks your people do, the actual capability levels in the room. Not “here’s how to write a good prompt” in the abstract, but “here’s how to use this tool for the specific challenges your fundraising team faces every week.”

That’s a fundamentally different offering. And it requires all the groundwork I just described — knowing your tools, knowing your people, having a strategy that connects the training to something larger.

Without that, you’re paying for something people could get for free. And in the nonprofit sector, that’s not a small thing.

Shared language is the unlock most people miss

I want to stay on this point about vocabulary for a moment, because I think it’s one of the most underrated barriers to AI adoption.

It can be genuinely intimidating for people to talk about AI if they don’t feel they have the correct understanding or the correct terminology. Especially in a room where a few colleagues are clearly further ahead. People go quiet. They stop asking questions. They nod along in meetings and then go back to doing things the way they’ve always done them.

One of the main goals of training is to enhance the communication around AI within your organisation. You can’t really do much unless you’re talking about it. Unless people share their concerns, their experiments, the things that worked and the things that didn’t. And that only happens when everyone feels they have the language to participate.

This is partly a training problem and partly a policy problem. Your policy can include a glossary of terms, clear definitions that everyone works from. Your training can reinforce those definitions and make sure that from this point forward, when someone says “agent” or “hallucination” or “prompt,” everyone in the room means the same thing.

It sounds basic. It’s not. It’s the difference between a team that can learn together and one where a few people race ahead while the rest quietly disengage.

The case for training your own people

On a recent call, I did something I probably shouldn’t admit to doing regularly: I talked myself out of potential work.

The person I was speaking with was looking for external training and consulting. And look, consultants absolutely have value, especially around strategy, frameworks, the architecture of how all these pieces fit together. But when it comes to the ongoing work of training and supporting a team through AI adoption, I keep coming back to the same conclusion: you’ll often get more value from someone internal.

The people who understand your organisation best are the ones inside it. They know the workflows, the politics, the unspoken constraints, the actual problems that need solving. If you’ve got someone who’s already showing initiative, someone with a natural pull towards this technology and a willingness to help others, that person is worth more to your AI adoption than any external consultant doing a fly-in engagement.

I know an organisation overseas, a small nonprofit, where the founder used AI to compress his own work down to three days a week. He now spends the other two days supporting his staff and even training similar organisations in what he’s learned. That’s not a consultant’s framework being applied from outside. That’s someone who figured out what works in their context and is sharing it from lived experience.

So before you default to looking outside, consider what’s already in front of you. Could someone’s role be pivoted, even part-time, into an AI implementation or support function? Could your existing champions be given formal permission and time to do what they’re probably already doing informally?

There’s a real problem right now with hiring for AI roles externally. Everyone can claim expertise. It’s genuinely difficult to verify. But the person who’s been quietly experimenting in your team for the past six months? You already know what they can do.

Make it an experience, not a lecture

One idea I keep coming back to is the internal hackathon.

Not the intimidating, technical kind. Something simpler. Take your AI policy, the document everyone was supposed to read but probably didn’t, and give teams an hour to transform it into something else using AI. A song. An infographic. A slide deck. An interactive website. A video.

What happens in that exercise is remarkable. First, people actually engage with the policy, because they have to understand it to transform it. Second, they’re building practical AI skills, prompting, iterating, working with different tools, without it feeling like a training session. Third, and this is the part I think matters most, you ask them to proof the output at the end. Because AI hallucinates. The content might be wrong. And that proofing habit, the instinct to check, to verify, not to trust the output blindly, is arguably the single most important skill in AI literacy.

Three outcomes from one exercise: policy comprehension, practical AI skills, and the critical thinking habit that makes all of it safe. And you’ve surfaced capability levels organically, because you can see where people are by watching what they build, without anyone having to self-report on a survey.

You don’t need an external facilitator for this. You need a few internal champions who understand the basics, a clear brief, and permission to have fun with it. If you’ve got more advanced people on the team, let them take the outputs further. Show what’s possible at the next level. Let the room see the gap between where they are and where they could be, and let that gap be exciting rather than threatening.

The leverage point

Here’s where this connects to the bigger picture.

Training is where most organisations put their energy when they decide to “do something about AI.” It’s visible, it’s actionable, it feels like progress. And it can be, if it’s embedded in something larger.

But training in isolation is an intervention at the wrong point in the system. It’s like trying to improve water quality by cleaning the water at the tap, when the problem is upstream. The upstream work is policy, assessment, tool decisions, shared language. Get those right, and training becomes transformative. Skip them, and training becomes an event people attended once.

The leverage point isn’t the training itself. It’s the strategy that makes training specific, relevant, and connected to everything else your organisation is trying to do with AI. Without that, you’re paying for something people could find on YouTube. With it, you’re building genuine capability that compounds over time.

If you’re planning AI training for your team, pause before you book anything. Ask yourself: Do we have a policy? Do we know where people are? Have we decided on tools? Do we have a shared vocabulary? If the answer to any of those is no, that’s where your time and money should go first.

The training will be better for it. And so will everything that comes after.

Kyle Behrend is an AI, Automations & Systems Specialist with 15 years in the nonprofit sector, based on the Central Coast of NSW, Australia. He is the founder of AI Impact Hub and writes The Leverage Point on Substack — exploring how small shifts create transformative change for mission-driven organisations.

P.S. If you’re at the stage where training is next on the list, start by checking the foundations. I’ve got a free maturity model available on my website that can help you figure out where your people are, which is the precondition for training that actually lands.

This is the step most organisations skip.

They write the policy (or they don’t, but let’s assume they did). Then they jump straight to training. They book a workshop, sign up for a webinar series, maybe bring someone in for a lunch-and-learn. And the training is fine. The facilitator is engaging. People nod along. But three weeks later, nothing has changed, and leadership wonders why adoption isn’t sticking.

Here’s what happened: they trained at the wrong level. They pitched to the middle of a room they’d never measured. Half the staff were already ahead of the material. The other half weren’t ready for it. And the people who might have been champions, the ones with a natural magnetic pull towards this technology, never got identified, so they went back to experimenting quietly on their own.

All of that is avoidable. But only if you start by finding out where people actually are.

Why this matters more than it sounds

I know “do an assessment” isn’t the most exciting advice. It sounds like bureaucracy. Another form, another process, another thing on the to-do list for an already overstretched team.

But think about what you’re working with when you skip it. You don’t know who’s already using AI, and for what. You don’t know who’s curious but hesitant. You don’t know who’s been experimenting with tools the organisation hasn’t sanctioned. You don’t know who’s terrified of it. And you don’t know who’s quietly brilliant at it.

If you don’t know where people are and what they’re using, it increases your risk and can diminish what support you can provide. That’s not abstract. It means your training misses. Your policy has gaps you can’t see. Your champions go unnoticed. And the people who need the most support don’t get it because you didn’t know they were there.

The assessment isn’t paperwork. It’s the precondition for every other step actually working.

What good assessment looks like

The most common starting point is a survey, and that’s a decent one. But not all surveys are equal. What you’re trying to do is measure people against some kind of maturity model, a framework that defines what capability looks like at different stages.

What does a level one user look like? Maybe they’ve tried ChatGPT a few times for personal use but haven’t applied it to their work. Level two might be someone using AI regularly for drafting or brainstorming but sticking to basic prompts. Level three could be someone who understands structured prompting, knows how to work with markdown outputs, can use conversational techniques to refine results, and has explored the deeper functionality of whatever tools the organisation provides.

You need a combination of quantitative and qualitative questions. The quantitative side gives you the data: specific skills, courses completed, tools used, terminology understood. The qualitative side gives you the texture. Questions like:

• How do you currently apply AI to your role?

• What opportunities do you think AI will present for your work in the future?

• Is there a tool or capability that’s piqued your curiosity, something that opens up ways of doing things that weren’t possible before?

• How would you identify which parts of your work could potentially be automated?

That blend, the measurable and the open-ended, gives you something you can actually work with. You get both a snapshot of current capability and a sense of where people’s heads are at.

I’ve got a downloadable maturity model on my website that you can use as a starting framework, or you can build your own based on the specific tools your organisation uses and what advancement looks like in your context.

Spotting your champions

One of the most valuable things assessment reveals is who your AI champions are, or could be.

A champion isn’t necessarily in a specific role. They’re not always in IT or operations. They’re the person who takes the initiative without being prompted. The one who’s been learning about AI on their own time, who gets genuinely excited about what’s possible, who’s already helping colleagues figure things out. They have a natural drive towards this technology and towards helping others use it.

You can’t always manufacture that. Sometimes nobody pulls ahead, and that’s a different situation entirely. In those cases, you’ve got three options: wait and see if someone emerges, hire specifically for an AI-focused role, or have someone in leadership step forward and take the reins.

Of those three, there’s no universal right answer. It depends on your size, your budget, your capabilities. But I’ll say this: it’s always helpful to have someone in leadership actually learning and getting excited about this technology, because that has a genuine flow-on effect through the organisation. Show, don’t tell. When staff see a leader who’s visibly curious and experimenting, not just issuing directives about innovation, it shifts the culture. People feel permission to explore. That’s not structural, it’s cultural. And culture is where adoption actually lives.

What you do with the data matters more than collecting it

Data is only as good as what you do with it. That might sound obvious, but I’ve seen enough surveys disappear into shared drives to know it needs saying.

Once you’ve got the results, you should be using them to make informed decisions. Who needs foundational training? Who’s ready for advanced capability building? Who should be given room to lead? Where are the gaps between what people know and what your policy expects of them?

Store the data somewhere accessible, because if you run a follow-up assessment in six months (and you should) you want to be able to track movement. Progress becomes visible. Gaps that persist become priorities.

And here’s a practical note: you can use AI to analyse the results. Anonymise the names, feed the data in along with your policy and any strategic goals, and ask for a proposed set of next steps. It won’t give you a perfect plan, but it’ll give you a starting point that’s grounded in your actual data rather than generic advice.

Reimagining the form

This is where I want to push a little further, because I think we’re leaving opportunities on the table.

A survey doesn’t have to be a form. It could be an interactive chatbot, a text-based experience that adapts as someone answers, adjusting the complexity of its questions in real time and landing the user at a capability level by the end. It could be a dynamically generated assessment app where the questions change based on previous answers. It could be a mini hackathon where you hand teams your AI policy and challenge them to transform it into different media: a song, an infographic, a video, an interactive web app. That exercise reveals capability levels without anyone having to self-report, because you see where people are by watching what they build.

I’ll be honest. This is the advanced end. If your organisation is just getting started, a well-designed survey is the right move. But if you’ve identified a champion or you’re ready to invest, these experiential approaches are the direction worth heading in. We have mediums available to us now that were previously gatekept behind the expense of developers. Why should an assessment look the same as it did five years ago?

I don’t think this space is thoroughly enough explored yet. But that’s exactly why it’s worth experimenting with, even if the experiment itself is the point.

The smallest viable step

Here’s what I’d actually like you to do after reading this. Not just nod along and file it away for later.

Open up whatever AI tool you have access to. Describe your organisation: your size, your sector, your current level of AI adoption, what tools you’re using. Tell it what you want to measure and why. Ask it to help you frame an assessment that fits your context.

This is a genuinely useful exercise for two reasons. First, you’ll end up with a draft assessment you can actually use. Second, the process of building it is itself a practical exercise in using AI, which means you’re developing your own capability at the same time as designing a tool to measure everyone else’s.

It’s one thing to read about this. It’s another to actually make it happen.

Kyle Behrend is an AI, Automations & Systems Specialist with 15 years in the nonprofit sector, based on the Central Coast of NSW, Australia. He is the founder of AI Impact Hub, and writes The Leverage Point on Substack — exploring how small shifts create transformative change for mission-driven organisations.

P.S. If you want a head start, I’ve got a free maturity model available for download on my website. Use it as-is or as a jumping-off point for building something tailored to your organisation.

But here’s what I keep running into when I talk to nonprofit leaders about AI: they tell me they know their staff are using it. They just don’t know how.

That sentence should be a red flag, and in most organisations, it isn’t being treated as one.

In the last article, I laid out the zoomed-out map — the full picture of how AI adoption fits together for nonprofits. I said the first step isn’t signing up for ChatGPT. It’s having the conversation. This piece is about the vehicle for that conversation, and why the humble AI policy is the most underestimated tool in the entire process.

The scaffolding, not the cage

When I talk about AI policy with organisations, I’m not talking about a compliance document that tells people what they can’t do. I’m talking about scaffolding — something that gives people a framework to operate within so they can actually grow.

A good AI policy provides clarity. It tells your team when and how they can use AI, when they shouldn’t, what data they can feed into these tools, and what stays out. It creates shared language across the organisation so that everyone, from the CEO to the newest caseworker, is operating from the same understanding.

That clarity matters more than most leaders realise, and not just for efficiency reasons. It matters for trust.

Consider this scenario. Your organisation uses predictive AI to show different donors different suggested donation amounts on a page, based on their previous giving history. A donor calls and asks whether you use AI. One staff member says no. Another says yes. You’ve just created a trust problem — not because of how you used the technology, but because your people weren’t on the same page about it.

Donors are going to start asking these questions. Funders already are. And in a sector built on trust, particularly in philanthropy and fundraising, having inconsistent answers isn’t a minor issue. It’s a reputational risk you can prevent with a conversation and a document.

The living document

Here’s where the framing matters. An AI policy isn’t something you write once and shelve. The technology moves too fast for that. Features that didn’t exist when you drafted your policy will emerge within weeks. New integrations will change what’s possible — and what’s risky.

I use a specific example when I talk about this. Early on, many organisations adopted the sensible-sounding rule: don’t put any private data into AI tools. Reasonable enough. Then the major platforms rolled out connectors — the ability to plug in your Google Drive, your Gmail, your SharePoint — and suddenly people were linking their entire organisational filing systems to AI tools without anyone stopping to ask: does this go against our policy?

Nobody reviewed it. Nobody flagged it. Because the policy was treated as a finished thing rather than a living one.

I recommend reviewing your AI policy at least every couple of months. Not because you need to rewrite it each time, but because the landscape shifts that quickly, and your policy needs to reflect how your organisation is actually using these tools, not how it was using them six months ago.

What goes in it

I’ve developed a template that’s freely available on my website, and I walk organisations through a version of it regularly. The skeleton of a good nonprofit AI policy covers several key areas.

Definition and scope. What are you actually talking about when you say “AI”? This policy might focus on generative AI specifically, but you may also need to address predictive AI separately. Getting clear on what’s in scope prevents confusion later.

A curiosity mindset within guardrails. The policy should encourage exploration, not shut it down. The goal is responsible experimentation — creating the conditions for people to try things while managing risk.

Data privacy and protection. This is the big one. What data can be used with AI tools? What absolutely cannot? Are your team members on paid plans that opt out of using your data to train models? If they’re on free tiers, do they understand the difference?

Human oversight. Every output needs a human in the loop. AI drafts; people decide. Quality control, fact-checking, and final judgement stay with your team. This needs to be explicit, not assumed.

Transparency and accountability. Will you disclose when AI has been used in your content? Under what circumstances? And critically — who is accountable when something goes wrong?

Training and review. Who is responsible for making sure people actually know how to work within this policy? How often will the policy itself be reviewed? And who is the go-to person for questions about compliance or concerns?

None of these sections need to be lengthy. Some of the most effective AI policies I’ve seen are a few pages at most. What matters is that they’re clear, shared, and actually used.

The real value is the conversation

Here’s what I’ve found organisations don’t expect when they start building their AI policy: the process itself is where the magic happens.

When you say “we’re developing our AI policy,” it opens the door to questions that most organisations have been quietly avoiding. Where are people already using AI? What tools are they using? What’s working? What’s not? Where do they think the organisation should be using it? What makes them uncomfortable?

Those conversations surface things no leader could discover through a survey or an all-staff email. They reveal the shadow use that’s already happening, the fears people haven’t voiced, the ideas people have been sitting on because they didn’t know if they were allowed to explore them.

The policy gives people permission — not just to use AI, but to talk about it openly. And that visibility is the foundation everything else gets built on. You can’t train people effectively if you don’t know where they are. You can’t manage risk if you don’t know what’s happening. You can’t build a culture of experimentation if people feel like they need to hide what they’re doing.

The leverage point

In systems thinking terms, this is an intervention that changes the information flows in your organisation. AI goes from something happening invisibly across dozens of desks to something the organisation understands, guides, and learns from collectively. That shift — from invisible to visible, from individual to shared — is where the disproportionate impact lives.

It’s not expensive. It doesn’t require a consultant, though having someone walk you through it can help. It doesn’t require technical expertise. It requires a decision to start, and a commitment to keep the conversation going.

If you read the last article and thought “right, but what do I actually do on Monday morning,” this is it. Start the policy conversation. Not because the document itself will transform your organisation, but because it creates the conditions for everything that comes after.

The scaffolding goes up first. Then people can build.

Next week, I’ll dig into what comes after the policy — understanding where your team actually is with AI, and why that assessment changes everything about how you approach training and tool access.

Kyle Behrend is an AI, Automations & Systems Specialist with 15 years in the nonprofit sector, based in NSW, Australia. He is the founder of AI Impact Hub, and writes The Leverage Point on Substack — exploring how small shifts create transformative change for mission-driven organisations.

P.S. I have a free AI policy template on my website that you can use as a starting point. It’s not meant to be copied and pasted — it’s meant to be adapted through exactly the kind of conversation I’ve described here.

What has changed is my understanding of the answer. Three years ago, I probably would have started with the tools. Now, after hundreds of these conversations across organisations of wildly different sizes, sectors, and levels of readiness, I’ve developed what I think is a pretty clear picture of how all the pieces fit together. I say “I think” deliberately — anyone claiming to be an expert in AI is naive at best. The landscape shifts every few weeks (or sometimes days). But the patterns I keep seeing in how organisations approach this? Those have been remarkably consistent, and they’re what I want to share here.

This is the zoomed-out view. The map I wish I could hand every nonprofit leader before we sit down and talk about any specific tool or tactic. Over the coming weeks, I’ll dig into each piece properly. But if you want to see how it all connects, start here.

The gap that keeps showing up

There’s a moment I’ve watched play out dozens of times now. I’ll show something practical, like loading a nonprofit’s strategic plan and past fundraising emails into a ChatGPT project and watching it draft something 90% of the way there, and the room shifts. People aren’t amazed so much as they are struck by the gap between what they assumed AI required and what it actually took. No coding. No specialist. Just documents they already had and a tool that costs less than a monthly phone plan.

At the Fundraising Institute of Australia conference recently, that reaction was palpable. People were blown away by how simple it was. And then, almost on cue, the conversation moved to the same place it always does: So where do we start?

Here’s the pattern I’ve noticed across all of these interactions. The aha moment is never the problem. People get it. They see the value almost immediately. The problem is what happens between that moment and anything actually changing.

People go back to their desks. They’re busy, and in this sector, “busy” doesn’t begin to cover it. The insight fades. They might try a prompt or two, get a mediocre result, and quietly move on. Not because they don’t see the value, but because nobody has created the conditions for them to do anything with it.

That gap between the spark and the sustained change is where I’ve come to focus almost all of my attention. And it turns out it’s not a technology gap at all. It’s a leadership one.

The thing most leaders don’t realise

Something that catches a lot of leaders off guard when I bring it up is that their team is almost certainly already using AI. They’re drafting emails with it, summarising documents, brainstorming ideas. They’re just not telling anyone.

This is shadow AI, and it’s far more common than most organisations realise. The irony is hard to miss. The organisations that haven’t developed a policy because they think “we’re not really using AI” are usually the ones where it’s most unregulated.

I heard of one nonprofit where a staff member was feeding confidential client information into ChatGPT, and because there were no guardrails in place, no policy, no guidance of any kind, the organisation’s response was to implement a blanket two-year ban on AI for the entire staff. One person’s misuse, and the whole organisation lost access to tools that could have been transforming their work for the next two years.

Think about what that actually costs. Not just in efficiency, but in learning, in capability, in the growing distance between what’s possible and what your team knows how to do. That distance widens every month right now, and a ban like that doesn’t solve the underlying problem. It just makes AI invisible again.

Start with the conversation, not the tool

If I could change one assumption I keep running into, it would be this: the first step with AI is not signing up for ChatGPT. It’s having the conversation.

A policy might sound like bureaucracy, but in my experience it’s the opposite. It’s the single most important first move because it forces the discussion most organisations have been quietly avoiding. What do we actually think about this technology? How might we use it? How should we not? What are we comfortable with? What worries us?

That conversation does something no tool can do on its own. It makes AI visible. It brings the shadow use into the open, surfaces fears and assumptions and opportunities in the same room at the same time, and it means that when people do start using tools with intention, they have guardrails built from shared understanding rather than restrictions born from fear.

I keep coming back to a simple analogy here. You wouldn’t hand everyone in your organisation a power drill before figuring out whether you need one, what it’s for, and what the safety basics are. The conversation comes first. The tool comes second.

The flywheel

Once the policy is in place and the conversation has happened, there’s a sequence I’ve seen work, and a sequence I’ve seen organisations skip, usually to their regret.

It starts with understanding where people actually are. Not everyone is in the same place. Some staff are already experimenting quietly, some are anxious about the whole thing, and some have never opened an AI tool in their life. You need to know this before you invest in training, or you’ll pitch it at the wrong level and wonder why it didn’t land.

From there, get people access to proper tools. Free tiers are fine for a first look, but they’re limited in ways that create a misleading impression of what AI can actually do. A paid subscription is less than most organisations spend on a single team lunch, and the return on that investment, even from basic use, is almost immediate.

Then train for where people are, not where you wish they were. Not a one-off workshop. Not a recorded webinar someone watches at double speed. Practical, hands-on training that gives people wins with work they’re already doing. Drafting that grant report, summarising board minutes, turning messy data into a narrative someone actually wants to read.

Getting those quick wins matters more than most leaders realise. Before anyone talks about transformation or strategy, people need to feel the difference in their own daily work. When someone saves two hours on something that used to take a full afternoon, they don’t need convincing anymore.

And then, only then, you create space for experimentation. This is where it gets genuinely interesting, because the time saved from those quick wins becomes the space for exploration. That’s where people start discovering uses nobody planned for. Things no training programme could have predicted, because they come from the people closest to the work.

This is the flywheel. Each step creates the conditions for the next. The wins create the time. The time creates the experimentation. The experimentation creates new wins. Skip a step, and the wheel stalls.

The drill, the Plover, and why this matters

I think about all of this through a lens that might seem oddly domestic.

When I put together flat-pack furniture at home, I can use a screwdriver or I can use a power drill. The drill gets it done in a tenth of the time. That’s the efficiency argument for AI, and it’s real. If you’ve got access to power tools and you’re still reaching for the hand tools, you’ve got an opportunity cost whether you see it or not.

But the interesting part isn’t speed. It’s what happens once you’re comfortable with the tool and start seeing opportunities you wouldn’t have considered before. I’ve used my drill to put up cat shelves on the walls, mount a TV, build hammocks in the outdoor cat run. Projects I never would have attempted with a hand tool, not because they were impossible, but because the effort would have been prohibitive.

We had a Plover nest in the middle of one of our paddocks recently. Plovers are Australian birds, and this one had set up right where the sheep walk, completely exposed to wind and rain. So I grabbed some leftover wood, cut it with the power saw, screwed on a piece of perspex we had lying around, and made a little shelter over the nest. Twenty minutes, maybe. No budget, no plan, no trip to the hardware store. I just saw a problem, had the tools, had some scraps, and built the thing.

That’s what I want nonprofit leaders to understand about AI. Yes, it does your existing work faster. But once your team is fluent with it, they start seeing problems they can actually solve. Problems they would have walked past before because the cost of solving them felt too high. The fundraising series nobody had time to write. The onboarding guide that’s been on someone’s list for three years. The impact data that could tell a powerful story if anyone had the hours to shape it.

The tool doesn’t just make existing work faster. It makes new work possible. But only if people get the chance to reach that point.

Where the leverage actually is

After three years of these conversations, and with the significant caveat that I’m still learning alongside everyone else, I’m more convinced than ever that the leverage point here isn’t the technology. It’s leadership making AI visible and creating the conditions for people to learn.

That means starting the conversation. Developing the policy. Understanding where your people are. Investing in tools and training. Protecting time for experimentation. Building a culture where people share what they’re discovering, the wins and the failures both.

None of that is expensive. Most of it isn’t even particularly hard. But it requires a decision, a deliberate choice to prioritise this now rather than next quarter or next year.

I want to be honest about why I think the timing matters. It’s not because AI is going to replace anyone’s job next Tuesday. It’s because the gap between what’s possible and what most nonprofits know is possible is growing wider every week. The organisations that start now, even imperfectly, even slowly, will be in a fundamentally different position a year from now than the ones that wait.

Your staff deserve access to these tools. Your mission deserves the multiplied impact. The people you serve deserve an organisation operating with every advantage available to it.

The power drill is sitting right there. Pick it up.

This is the first in a series. Over the coming weeks, I’ll go deeper into each of these components: policy, tool access, training, experimentation, culture. If there’s one you want me to tackle first, hit reply and let me know.

Kyle Behrend is an AI, Automations & Systems Specialist with 15 years in the nonprofit sector, based on the Central Coast of NSW, Australia. He is the founder of AI Impact Hub, and writes The Leverage Point on Substack — exploring how small shifts create transformative change for mission-driven organisations.

I didn’t come back.

It wasn’t that I couldn’t do it. I’ve built websites for myself and organisations using exactly this approach: page builders, frameworks like Automatic CSS that bake in best practices, section templates you assemble and customise. I’ve done good work this way. But I knew what was waiting for me when I reopened that site: not momentum, but friction. The slow, manual process of dragging elements around a canvas, one section at a time, knowing there were dozens more to go.

So I let it sit. I had client work. Other priorities. The site I wanted — a dynamic portfolio that could showcase my projects, skills, and the work I’d been doing with nonprofits — stayed on the someday list.

Then, a few months later, I built it in under a week.

I’d been using Claude Code for about six weeks at that point, mostly on smaller experiments. Redoing a nonprofit’s donation page. Turning an impact report into something more like an experience than a document. I kept starting projects, hitting a natural pause point, and moving on to the next one. Not because of limitations in the tool — because I was genuinely more interested in the beginning part. The creative phase. The moment where an idea becomes something visible.

Those experiments changed my understanding of what was possible. Not in an abstract, futuristic sense. In a practical, I-just-watched-that-happen sense. I’d describe what I wanted in plain text, and code would appear. Not placeholder code. Working code with animation, with responsive design, with the kind of polish I’d previously associated with hiring a developer.

At some point the obvious thought arrived: I should use the tool that I teach to build the platform that I want to showcase.

So I did.

The first session was fun in a way that the December attempt simply wasn’t. I used voice dictation to describe what I wanted — the same interview-based approach I use for most things. Share ideas out loud, let the AI extract structure from them, review a draft, refine it. Within a few hours I had something visible. Not finished, but real. A thing I could react to, reshape, build on.

I noticed something about the language I was using as I worked. I kept saying “we.” We mapped out the structure. We tried a different approach for the hero. We added Easter eggs. It wasn’t a slip — it genuinely felt like a collaboration. I had creative direction: I wanted the site to feel like a terminal window, with those three coloured dots in the navigation that you could actually click on. I wanted hidden details that rewarded curiosity. Claude Code had the technical competency to make those ideas real, and the capacity to suggest things I hadn’t considered.

You could watch it happen in real time. I had localhost open in Chrome, and as Claude Code wrote changes, the site would shift and adapt right in front of me. It’s a strange thing to watch your ideas materialise that quickly. Not all of it worked first time — there was troubleshooting, there was iteration. But the direction was always forward.

The whole build took roughly eighteen hours of work, spread across less than seven days, squeezed between client projects and after hours. The site that went live — kylebehrend.com — has a portfolio section, project pages, an interactive terminal, a dark mode toggle, and Easter eggs hidden throughout. Built with Claude Code, Next.js, and Vercel. A $200 monthly subscription and my own time.

I want to be honest about why I’m telling this story, because it’s not really about me.

I’ve spent fifteen years in the nonprofit sector, and here’s what I know about development capacity in most mission-driven organisations: there isn’t any. Very few nonprofits have a developer on staff, because developers are expensive. So when an organisation needs a website for a specific campaign, or wants to turn their annual report into something people actually engage with, or needs a custom tool for their team — the answer is usually “we can’t afford that” or “maybe next financial year.”

That constraint has shaped what nonprofits believe is possible. And beliefs about what’s possible shape what people even bother to attempt.

Something has shifted. I’ve been thinking about it as the difference between the efficiency era and the builder era. When conversational AI first arrived — ChatGPT, Claude, and Gemini — most of the value was about doing existing work faster. Drafting content. Summarising documents. Helping with emails. That’s real and useful, but it’s optimisation. You’re still doing the same things, just quicker.

What’s happening now with tools like Claude Code is different in kind, not just degree. We’re not optimising existing workflows. We’re building things that didn’t exist before. Things that weren’t even on the roadmap because the cost and expertise required put them permanently out of reach.

Imagine a nonprofit that can spin up a purpose-built campaign website in a week. Not a template with their logo dropped in — a genuinely crafted experience designed for that specific campaign, that specific audience. Use it, learn from it, build the next one. A two-hundred-dollar subscription instead of months of lead time and tens of thousands in development costs.

That’s not a marginal improvement. That’s a different category of capability.

But here’s where I have to be careful, because I know the reality on the ground. I work with organisations every week who are still building foundational AI skills — learning how to prompt well, how to use AI as a collaborative thought partner, how to do the basics with confidence. Claude Code is, for most nonprofits right now, a long way off.

I don’t think that’s a reason not to talk about it. If anything, it’s a reason to talk about it more clearly, because the gap between what’s possible and what most organisations know is possible is growing fast. And that gap is where opportunity lives — or dies.

The leverage point here isn’t the tool. It’s the conditions.

Organisations that give their people time to experiment, space to fail, and access to training are going to discover capabilities they haven’t even imagined yet. That’s not hyperbole — I literally didn’t know what I was going to build until I started building. The donation page experiments, the impact report prototypes, the portfolio site — none of those were on a project plan. They emerged from having the space to explore.

This is what I want nonprofit leaders to hear: you don’t need to understand Claude Code. You don’t need to become technical. But you do need to create the conditions where the people in your organisation can develop these skills. Give them time. Give them a subscription. Give them permission to experiment without a guaranteed outcome. Because the organisations that do this are going to find themselves filled with opportunities and ways to make impact that they haven’t even figured out yet.

We’ve got so much to figure out. That’s not a problem. That’s the most exciting thing about this moment.

The footer of my new site says “Built with Claude Code.” I thought about whether to include that. It felt important — not as a gimmick, but as a statement. The tools I use to build are the same ones I teach. If I’m going to tell nonprofits that AI can change what’s possible for them, I should probably be willing to show that with my own work.

So here it is. Eighteen hours. One week. A site that I’d put off for months because the old way of building had lost its momentum.

The tool didn’t do it for me. But it did something that matters more than most people realise: it made the gap between what I could imagine and what I could create small enough to step across.

That gap is getting smaller for everyone. The question for nonprofits isn’t whether this technology is ready. It’s whether they’re creating the conditions to find out what is on the other side of possibility.

Kyle Behrend is an AI, Automations & Systems Specialist with 15 years in the nonprofit sector, based on the Central Coast of NSW, Australia. He is the founder of AI Impact Hub, and writes The Leverage Point on Substack — exploring how small shifts create transformative change for mission-driven organisations.

P.S. If you want to see the site for yourself, it’s at kylebehrend.com. Try clicking the dots.

I took the majority of AI Impact Hub’s foundational courses — prompting, custom GPTs, AI policy templates — and made them completely free for anyone working in the nonprofit sector.

I’d love to tell you it was a strategic masterstroke. The truth? It was driven by something closer to desperation.

Not personal desperation — global desperation.

The numbers that broke me

Seventy percent of nonprofits don’t have a budget for AI training. More than half cite cost as the primary barrier to adoption. And only 14% have even published an AI policy — which is really just the bare minimum signal that leadership has acknowledged this technology exists.

I kept staring at those numbers and thinking: these are the organisations working on the world’s biggest problems. Injustice. Inequality. Poverty. Climate change. Factory farming. Modern slavery. Mental health crises growing faster than the systems built to address them.

The single most transformative technology most of these organisations will encounter in their lifetimes is sitting right there — and cost is locking them out of even understanding what it can do.

So I removed the cost.

The uncomfortable truth I didn’t want to admit

Here’s the part I wrestled with for a long time.

If you’d offered me these same free resources during my worst years of burnout in the sector — the years that nearly broke me — I probably wouldn’t have used them.

I was too busy being busy. Too exhausted to learn something new. Too deep in the cycle of urgent tasks to step back and think about what might actually make things better.

And that’s the trap. The people who need this most are often the least able to engage with it. They’re buried under soul-crushing manual tasks, drowning in admin, running on fumes — and the idea of adding one more thing, even a free thing, even a potentially transformative thing, feels impossible.

Making training free removes a real barrier. But it doesn’t remove the burnout. It doesn’t remove the culture that normalises overwork. And it doesn’t remove the leadership gap that leaves individual staff to figure this out alone.

Where leadership is failing

Fourteen percent. That number haunts me.

An AI policy isn’t magic. But it’s the bare minimum signal that someone at the top has sat down and said: this matters, we need to think about it. And 86% of organisations haven’t even done that.

The real barriers I keep seeing aren’t about technology or cost. They’re about visibility. Staff normalise inefficient tasks because they don’t know alternatives exist. They don’t know that the thing eating three hours of their week is a problem that’s already been solved. No one has shown them what’s possible — and no one at the top is creating the conditions for them to find out.

Then there’s the fear. Some staff worry AI will replace them, which keeps them from engaging with the very tools that could make their work more meaningful and less grinding.

I’ve worked with leadership teams where we met every single week — training them, upskilling them, developing strategy, building an implementation timeline with tool guidance. That top-down investment changes everything. But it’s the exception, not the norm.

And here’s where I have to be honest: by making the training free for individuals, there’s a risk I’m letting leadership off the hook. Reinforcing the pattern of leaving it up to staff to figure it out on their own time, with their own energy, on top of everything else they’re already carrying.

This needs to be both. Top-down and bottom-up. Leadership needs strategy, investment, and guided support. Individuals need accessible resources and permission to experiment. Neither alone is enough.

So where’s the leverage point?

If cost isn’t the barrier, and leadership buy-in is slow, and burnout makes learning feel impossible — where does the actual shift happen?

I keep coming back to the same answer, and it’s almost embarrassingly simple.

Five minutes.

Just open the tool. Pick one task — one thing you’re already doing that feels tedious or repetitive. Open ChatGPT or Claude. Create a project. Drop some relevant files in. Start a conversation. And go back to that same project tomorrow.

That’s it. That’s the intervention.

Slow and steady consistency compounds and wins every time versus trying to do everything all at once. I’ve seen this pattern over and over. The people who break through aren’t the ones who take a weekend course or attend a conference or read every article about AI strategy. They’re the ones who build a small daily habit and let it compound.

The flywheel effect with AI is real. I see it in my own work building agentic systems — agents with skills, memory, and schedules. You can’t do everything at once. But five minutes today becomes ten tomorrow, becomes a fundamentally different way of working within weeks.

The leverage point isn’t access to AI training. It’s the daily habit of actually using the tool.

What an aha moment actually looks like

At a recent conference, someone came up to me buzzing with excitement. She’d discovered Google’s NotebookLM and was recording every presentation she attended, uploading them to a notebook, and using it to continue her learning afterwards — creating infographics, pulling out key insights, turning conference notes into something she could actually integrate into her work.

She hadn’t taken a course. She hadn’t been mandated by leadership. She’d found a tool that solved her specific problem, and suddenly the whole landscape opened up.

That’s what the aha moment looks like. It’s not abstract. It’s deeply personal. It’s: I had no idea this was possible.

You can have more than one. You can have them with different tools, different use cases. But the first one requires you to show up and try — and that’s the five minutes I’m talking about.

What’s at stake if nothing changes

I’ll be direct: by 2028, nonprofits that aren’t prioritising AI are in for a rude awakening.

The exponential curve is steepening. Funder expectations are shifting — they’ll start asking how organisations are leveraging AI, and “we haven’t started yet” will stop being an acceptable answer. Staff will demand access to tools. NVIDIA is now proposing providing engineers with half their salary’s worth in AI tokens each year. Meanwhile, most nonprofits won’t even give their people a $20/month paid licence.

Completely different spectrums.

I believe we’ll start seeing organisations actually shutting down because of this gap. And something will flip that defies the traditional logic of the sector: smaller, more agile organisations will make a bigger impact than larger ones — with fewer resources — because they found the leverage point faster.

There’s an irony I can’t shake. Some nonprofits celebrate being around for fifty years. But shouldn’t the goal be to solve the problem as fast as possible? Shouldn’t success actually mean putting yourself out of business? AI doesn’t guarantee that — but it’s the most powerful accelerant these organisations have ever had access to. And right now, it feels like a profound lost opportunity.

The pain of staying the same has to eventually outweigh the pain of changing. I just hope that awakening doesn’t come too late.

What I’m asking

The free courses are there. Prompting fundamentals. Custom GPTs. AI policy templates. The building blocks. If you work in the nonprofit sector, I genuinely hope you’ll use them and share them with someone who could benefit.

But I also want to hear from you.

If cost isn’t the barrier — what is? What are the constraints you’re actually facing? What would it take for you to open an AI tool for five minutes tomorrow?

Because I’ve removed one obstacle. I want to know what the others are, so I can work on those too.

Kyle Behrend is an AI, Automations & Systems Specialist with 15 years in the nonprofit sector, based on the Central Coast of NSW, Australia. He is the founder of AI Impact Hub and writes The Leverage Point on Substack — exploring how small shifts create transformative change for mission-driven organisations.

P.S. If you want to start with those five minutes today: open ChatGPT or Claude, create a new project, name it after something you’re working on right now, and ask it one question about that work. Don’t optimise. Don’t strategise. Just start. Then come back tomorrow and do it again.

Not because you’re doing anything bad — but because of a design choice you probably never noticed. That chat bar inside of ChatGPT? It looks exactly like a Google search bar. And after twenty-plus years of Googling, your brain does what it’s been trained to do: type a question, get an answer, move on.

But AI isn’t a search engine. And using it like one means you’re leaving about 95% of its value on the table.

The average of the internet isn’t you

Here’s the thing most people miss. Large language models have been trained on the entire internet — books, videos, articles, everything. So when you ask for an output without giving it context, you get the average of all of that. A generic, one-size-fits-all response.

Which is fine if you want a recipe for banana bread.

But if you’re trying to draft a grant application, write a communications strategy, or figure out how to restructure your giving program — generic isn’t going to cut it.

Think about it this way. If you searched “how to deal with anxiety” online, you’d get a wall of generic advice. But if you sat down with a psychologist, they’d ask you questions — about your life, your triggers, your history — and tailor their approach to you.

AI can work the same way. But only if you let it ask the questions.

The technique: get AI to interview you

This is what’s often called using AI as a strategic thought partner. And the whole method boils down to one shift:

Instead of asking AI to do a task, ask it to understand you first.

Here’s what that looks like in practice. Say you want to draft a LinkedIn post about something you learned last week. Instead of typing “write me a LinkedIn post about X,” you’d say:

“I’d like to create a LinkedIn post about something I learned last week. Please interview me, asking one question at a time, until you have sufficient context to help me draft it.”

What happens next is where the magic is. The AI starts asking you questions — and each one layers on the last, just like a great interviewer would. It draws out your specific experiences, your unique perspective, the details that make the content yours.

The output you get at the end isn’t the average of the internet. It’s something built from your context, your knowledge, your voice.

You don’t need to be at your desk

I know what you’re thinking. “I barely have time to type a prompt, let alone have a whole conversation with AI.”

Fair. But here’s the thing — you don’t even need to be sitting at your computer for this.

Open ChatGPT’s voice mode on the mobile app. Put your headphones in. And have the conversation while you’re doing something you’re already doing — driving to work, walking around the block, cooking dinner.

Even better: create a Project in ChatGPT and have your voice conversations happen within it. The project holds more context and stores memories across sessions, which means each conversation gets richer over time.

I took this idea to an extreme. I built an automation that calls me every day at 3.30pm and asks me five reflective questions — like a voice-powered daily journal. I’d wanted to journal for years but could never stick with it. Now it just happens, because I attached it to something I was already doing: answering the phone.

That’s a leverage point — a small change in the system that creates a disproportionate result. I wasn’t journaling. I never could stick with it. But instead of trying harder, I changed the system: I removed the manual effort entirely and let AI come to me. Now it just happens.

The same principle applies to how you use AI every day. You don’t need to become a prompt engineer. You just need to make one small shift in how you interact with it — stop giving AI instructions, and start letting it ask you questions. That’s the leverage point. One sentence added to a prompt changes the entire dynamic, and the outputs you get back will be staggeringly better because of it.

I walked through this technique live in a recent video — from the first prompt all the way through to a finished draft. If you want to see exactly how the questions build on each other, give it a watch:

The five-minute version

If you want to try this right now, here’s the simplest possible starting point.

Next time you’re about to type a prompt into ChatGPT, before you hit enter, just add one line at the end:

“Before you start, please ask me three to five questions one at a time to get more context and information.”

That’s it. One sentence. It shifts the dynamic from you trying to give AI the right instructions, to AI helping you figure out what it actually needs.

Try it once and see what happens. I think you’ll be surprised by how much better the output is — and by what the questions draw out of you that you didn’t even know was there.

Try it this week. Use this technique on something real — a report, a social post, an email you’ve been putting off. Then come back and tell me what you found in the comments. I’d love to hear what surprised you.

Kyle Behrend is an AI, Automations & Systems Specialist with 15 years in the nonprofit sector, based on the Central Coast of NSW, Australia. He is the founder of AI Impact Hub and writes The Leverage Point on Substack — exploring how small shifts create transformative change for mission-driven organisations.

I didn’t know what was wrong. I was in my early twenties, giving everything I had to an organisation I believed in, and I was falling apart. I didn’t have the language for it then. I didn’t know what burnout was. I just knew I wasn’t myself anymore.

I’ve since learned the word. I’ve also watched it happen to dozens of others — brilliant, passionate people who entered the nonprofit sector because they wanted to make the world better, only to be ground down by broken processes, impossible workloads, and systems that were never designed to support them.

Many of them left the sector entirely.

That’s not a staffing problem. That’s a systems failure.

The task no one questioned

A few years ago, I was consulting with a nonprofit during a discovery session about potential AI projects. In passing — almost as an afterthought — a staff member mentioned something about their campaign season workflow.

I asked them to stop and tell me more.

During their busiest period, staff were staying up late every night manually collecting every media mention of their organisation. Copying URLs into spreadsheets. Pasting journalist names, article titles, dates — one by one. All so they could run analytics at the end of the campaign.

They’d been doing this for years. No one had questioned it. It was just “part of the job.”

We automated the entire process. A Google News alert feed handled the bulk of it. For anything missed, staff could drop a URL into a simple input, and AI would extract the journalist’s name, the title, the date — populating the spreadsheet automatically.

The task that had been keeping people up at night during their most demanding season simply... disappeared.

This is what a leverage point looks like. Not a massive organisational overhaul. One small shift in how a task gets done — and suddenly, people get their evenings back.

The concept that changed how I see everything

The systems thinker Donella Meadows wrote a famous paper called Leverage Points: Places to Intervene in a System. Her core insight was that in any complex system, there are specific points where a small shift can produce enormous change.

I named my newsletter after this idea because I’ve lived it.

Over the past few years, I’ve watched small interventions — an AI tool, a prompt, an automation — create disproportionate results. Not incrementally better. Transformatively different.

But here’s the thing Meadows also warned about: the higher the leverage point, the harder the system resists change.

And nowhere is that resistance more visible than in the nonprofit sector.

The real barrier isn’t technology — it’s visibility

The biggest leverage points in nonprofits are hiding in plain sight. They’re buried in the tasks people have normalised. The manual data entry. The meeting notes no one has time to write up properly. The reporting processes held together with spreadsheets and sheer willpower.

Staff don’t flag these as problems because they don’t know there’s an alternative. They’ve never seen what’s possible.

This realisation is what shifted my entire approach. I’d been building AI solutions for nonprofits as a consultant — and getting great results. But I kept coming back to the same thought:

The people inside these organisations will always be better positioned to spot the leverage points than I am.

They know where the friction lives. They feel it every day. They just don’t know it’s solvable.

So I shifted from implementation to education. I built the AI Impact Hub with courses, tutorials, and resources — because the real leverage point isn’t any single automation. It’s giving people the knowledge to see opportunities they didn’t know existed.

That’s a meta-leverage point, if you like. Moving from intervening at the level of individual fixes up toward changing how people see their own work. Meadows would call that shifting the information flows of a system — and it sits much higher on her hierarchy than any single automation ever could.

Beyond efficiency: the possibilities no one is talking about

Most conversations about AI in nonprofits start and end with efficiency. Write emails faster. Summarise documents. Save a few hours a week.

That matters. But it’s not the real story.

I worked with an organisation that had a twelve-part framework for guiding people through a process of change — traditionally delivered as a book. They were curious about whether a chatbot might help.

I encouraged them to experiment with a tool called Lovable. They vibe-coded a working MVP of an interactive app — no developers, no grant funding for a tech build, no six-month procurement process. Just experimentation and a willingness to try.

They plan to take that prototype to funders.

A year or two ago, that would have required a significant grant just to reach the proposal stage. Now, a small team with the right tools can build a working proof of concept in days.

This is the part of the AI conversation that’s being missed entirely. It’s not just about doing existing work faster. It’s about doing things that were never possible before — apps, interactive tools, personalised experiences — with the resources nonprofits actually have.

The time from ideation to execution has dramatically shrunk. And I’m just not hearing enough about nonprofits seizing that opportunity.

The survival trap

So why aren’t more nonprofits acting on this?

Meadows wrote that the goals of a system are among the highest leverage points. In nonprofits, the stated goal is always mission impact. But the actual operating goal — the one that drives daily decisions — is often survival. Keep the lights on. Meet the grant deliverables. Retain the donors. Don’t burn out the staff (or at least, replace them when they do).

When you’re in survival mode, you can’t lift your head long enough to see transformative possibilities. Every hour is accounted for. Every task feels urgent. The idea of taking time to experiment feels irresponsible when there are fires to put out.

And the resistance runs deeper than just time. The nonprofit sector attracts a special type of person — people driven by passion, by wanting to make the world better. That’s what makes them extraordinary. But it also means the sector hasn’t traditionally attracted or developed deep technical talent. Leadership often doesn’t know what’s possible with AI, and what you don’t know exists, you can’t prioritise.

Layer on top of that a natural fear of change, the very real challenges of change management, and the fact that even when organisations want the outcome, they resist the process of getting there — because the cost is immediate but the benefits are delayed — and you have a system that naturally favours the status quo.

The result is a self-reinforcing loop. Too busy to learn AI, so you stay busy, so you never learn AI. The only way to break the cycle is a deliberate leadership decision to prioritise it.

Breaking the loop

The good news is that the flywheel effect on the other side is massive.

Start with something simple — meeting transcription tools so staff don’t have to manually take notes, draft summaries, and compile action items. That saves time. Use that time for deeper exploration. Which saves more time. Which opens up new possibilities.

One practical starting point: run a time audit. Ask staff to track how they spend their time for a week. Where are the hours going? Which tasks are repetitive, manual, and draining? That’s your map of leverage points.

I recently saw an online publication take an entire week off normal operations just for AI experimentation — every staff member, dedicated time, no other expectations. Imagine what a nonprofit could discover with even a fraction of that commitment.

But none of this happens without leadership making the call. Whether it’s top-down — leaders prioritising AI literacy as a strategic imperative — or bottom-up — staff requesting access to paid AI tools, training, and support — someone has to decide that this matters more than the next fire to put out.

Because every task you choose to do is a task you’re prioritising over one you don’t. The question is whether AI experimentation makes that list.

The 14% problem

Late last year, Infoxchange released a study showing that only 14% of the 800-plus nonprofits surveyed had an AI policy.

Not an AI strategy. Not training programmes. Not implementation plans. Just a policy — the bare minimum signal that an organisation is thinking about this at all.

That means 86% of nonprofits haven’t even started the conversation.

An AI policy might sound like a small thing. But creating one means leadership has acknowledged AI as a priority. It means they’re talking to staff about when and how to use these tools. It means the door is open.

Without it, the door isn’t just closed — most organisations don’t even know there’s a door.

The cost of staying the same

I think a lot about the idea that for change to happen, the pain of staying the same must outweigh the pain of changing.

That tipping point is coming — fast.

If you look at what’s happened with AI over the last three years and project forward, the organisations that haven’t invested in AI literacy, training, and experimentation are going to face serious consequences. Not hypothetically. Practically.

The nonprofits that are experimenting now — building prototypes, automating workflows, upskilling their teams — will pull away from the pack. They’ll write better grants. They’ll deliver more innovative programmes. They’ll attract more funding and support.

And the ones that don’t? Some of them will get wiped out.

That’s not fear-mongering. That’s the reality of what happens when a sector built on passion and manual effort meets the most transformative technology of our time and chooses not to engage with it.

Meadows warned that systems which cannot self-evolve are doomed on a highly variable planet. The nonprofit landscape is shifting rapidly. The organisations that refuse to experiment aren’t just missing opportunities — they’re becoming fragile.

What this is really about

At the end of the day, this isn’t a technology conversation. It’s a conversation about dignity.

The people who work in nonprofits are sacrificing a lot. Less pay. Fewer resources. Limited career progression. They do it because they believe in the mission. They believe the work matters.

The least we can do is not make the work harder than it needs to be.

Good systems empower people to do great work. They lift people up. They enrich the experience of working at an organisation. They free people from the soul-crushing manual tasks that lead to late nights, burnout, and eventually — too often — leaving the sector altogether.

I know this because I’ve lived both sides. I’ve been the person calling my dad in tears, not understanding why I was falling apart. And I’ve been the person watching an automation give a team their evenings back during campaign season. The distance between those two experiences is often just one small shift — one leverage point that someone finally decided to look for.

Every nonprofit leader should be asking themselves one question: Are our systems setting our people up for success, or are we burning through their goodwill and hoping the next hire will survive what the last one couldn’t?

The tools exist now. The leverage points are there. They’re hiding in the tasks your team has stopped questioning.

All it takes is someone willing to look.

Kyle Behrend is an AI, Automations & Systems Specialist with 15 years in the nonprofit sector, based on the Central Coast of NSW, Australia. He is the founder of AI Impact Hub and writes The Leverage Point on Substack — exploring how small shifts create transformative change for mission-driven organisations.

P.S. If you’re a nonprofit leader wondering where to start, try this: ask your team to name the one task they dread most each week. That’s your first leverage point. You might be surprised how solvable it is.

Last week I was invited to the launch of OpenAI Australia. It was an incredible evening filled with demonstrations, conversations, and presentations — but one line stuck with me more than anything else.

Canva CEO Melanie Perkins explaining her vision of how AI is moving us from an age of information to an age of imagination (watch a video snippet below).

Over the last 24 months I’ve encouraged multiple non-profit organisations to keep a wish list. Everything they hope AI can help them do. The campaigns they could only dream of because of time and budget constraints. The operational improvements. The workflow automations that would free up hours each week.

And each day, as the technology improves and tools expand their capabilities, we’re closer than ever to those dreams becoming reality.

Just last month we saw incredible model improvements. But Nano Banana Pro has been stealing the show. The ability to create jaw-dropping infographics around any topic, article, or concept in minutes. Complete with design, spacing, and fully legible text. It’s left a lot of people speechless.

Think about what this means practically.

A year ago, creating professional visual campaigns required a designer or an entire team. Now? AI image generation tools can create incredible visuals for social media, email, and websites. You can even mock them up on billboards or in public spaces to show funders and donors what your campaign could look like — before you’ve spent a cent on media buying.

The possibilities are genuinely exciting. But here’s the thing.

While non-profits have these opportunities available to them, the real challenge isn’t access to the technology. It’s turning that access into something actionable.

Without the right scaffolding, all of this potential just sits there. Unused. Under-utilised. Another shiny tool that never quite delivers.

That’s where an AI Strategy comes in.

What is an AI Strategy?

An AI Strategy is a collection of individual components that, when brought together, empower your organisation — and the staff and volunteers within it — to actually move into this new age.

It’s not a single document. It’s not a one-off training session. It’s a framework that creates the conditions for AI to genuinely make a difference.

Here are the six components that make up a comprehensive AI Strategy.

1. An AI Policy (That Actually Lives)

This isn’t like any other policy gathering dust in a shared drive.

An AI Policy needs to be a living document. One that’s revisited regularly as the technology evolves. It should create an opportunity for discussion — allowing everyone to express their views, concerns, and ideas.

More importantly, it creates a safe space. Staff know what’s allowed and what isn’t. They understand when they can use AI and when they shouldn’t. They have clarity on privacy, security, and data handling.

It should allow for exploration within clear boundaries. Not a document that shuts things down, but one that opens things up — safely.

Enforcing privacy settings on AI tools is also vital. For example, disabling options like “Improve the model for everyone” in ChatGPT or “Help improve Claude” in Claude helps protect your organisation’s data. For higher security needs, consider enterprise-grade AI services with stronger encryption.

2. A Budget for Paid Tools

Ethan Mollick said it best: the free versions are demos, paid versions are tools.

You really can’t go wrong with ChatGPT, Gemini, or Claude — all offering different non-profit discounts and plans which I’ll explore in future articles. The important thing is to pay for access and get everyone on a shared platform.

This removes the chaos of people using their own personal subscriptions. It creates opportunities for sharing prompts, templates, and learnings. It brings AI usage into the organisation rather than leaving it scattered across individual accounts.

Here’s a simple ROI calculation: someone earning $45,000 per year only needs to save 12 minutes per week to justify a $20/month subscription. Most people save far more than that once they know what they’re doing.

3. A Maturity Model

Based on your policy and your tools, what does progression actually look like?

What skills and tasks should staff be able to do at each level? What’s level 0 versus level 3? Where do you see most people striving towards, and why?

A maturity model gives people a roadmap. It shows them where they are, where they could be, and what it takes to get there. Without it, AI adoption feels aimless. With it, there’s a clear path forward.

4. A Survey to Understand Where People Are

Before you can move people forward, you need to know where they’re starting from.

A simple survey can help you understand current experience levels, existing use cases, comfort with different tools, and goals for progression.

This isn’t about judging anyone. It’s about meeting people where they are and designing support that actually helps.

5. A Plan for Training and Upskilling

With all of the above in place, you need a plan — and a budget — for training.

If you’re building training internally, make sure it’s delivered by people who understand the practical side of AI. People who can make it human through their teaching. AI is difficult to teach because it’s so general, and complexity scares people. The best trainers make it feel accessible and relevant.

If internal expertise isn’t available, invest in external training from people who’ve done this work before.

6. Making AI Part of the Conversation

This might be the most important component of all.

Create a Slack channel or Teams group where people can share what worked and what didn’t. Wins and cool tools. Failed experiments and lessons learned.

Make 10 minutes at the end of team meetings to chat about AI. Start with just 5 minutes if that feels more manageable, then progress to dedicated 1-hour AI discussions as confidence grows.

Carve out time. Give staff at least 1 hour per week specifically for learning and experimentation. Put a sticky note on every desk or screen that asks: “How can AI help with this?”

AI needs to be visible. It needs to be explored, experimented with, and used consistently for it to actually make a difference.

The Scaffolding That Makes It Work

Those six components together create your strategy.

And here’s the truth that leadership needs to hear: no matter how amazing and powerful this technology becomes, non-profits will never move into the age of imagination without this scaffolding.

An AI Strategy isn’t a nice-to-have. It’s becoming a requirement.

If you want your non-profit to be as effective as possible — if you want to actually capture the value that AI can deliver — then building this foundation isn’t optional. Without it, the technology won’t work. People won’t adopt it. The potential will remain just that: potential.

With it? Your team will be empowered to step into the Age of Imagination. One where AI helps make what was once impossible, possible. Opening up opportunities that will help non-profits create a bigger impact than ever before.

The tools are ready. The question is: is your organisation?

But here's the thing: while creating complex AI workflows can be valuable, it's not the best starting point for most people exploring AI. Instead, I live by a simple mantra: start small, but start today.

I understand the paradox of choice that AI presents. With countless possibilities, personas, and approaches available, where do you begin?

Let me share my daily go-to use case: email refinement. I tend to write long-winded emails where one sentence morphs into a complex tangle of ideas. As the communicator, it's my job to ensure clarity, even when the message makes perfect sense in my head.

My solution? I copy the text into ChatGPT and simply ask it to "clean up" or "refine" the paragraph. No fancy prompts needed. Within seconds, it restructures the ideas, simplifies the sentence structure, and helps clarify my thoughts. The result is cleaner, clearer communication.

We all send emails. We all need to communicate ideas effectively. Why not let AI help us do it better?

After all, the journey of a thousand miles can begin with a single email.

So what's the point?

We consume information in a range of ways. From the good ol' book, to websites, videos, podcasts and the like. But AI tools like ChatGPT have transformed how we can engage with it. It shifts a passive experience into an active one.

Subscribe now

We could manually summarise a video or meeting, but why bother when AI can do it in a fraction of the time? Plus, we can now do a whole lot more.

We can ask questions, extract pull quotes, get talking points, summaries, action items, key takeaways. And we can personalise it based on exactly what we want and who we are all with the simple tweak of a prompt.

Information from long-form podcasts and videos is not often found online. Sometimes it takes a long discussion to dig into the depths of the topics, to uncover those golden nuggets hiding beneath.

I recommend everyone try this new go-to strategy. After watching a video or listening to a podcast, download the transcript and use AI to explore it further. This method allows us to extract more value from the content we consume, uncovering insights we might have missed during the initial listen or watch. It's like having a personal research assistant that can sift through hours of dialogue in seconds, highlighting the most relevant information for my needs.

Reflecting on this process, I'm struck by how AI is not just changing how we generate information but also how we consume it. It's as if we're developing a new kind of literacy - one that combines traditional comprehension with AI-augmented analysis.

In a world where information is abundant, the key to wisdom will be how we distill it.

Every visit to Bunnings presents the same challenge. Finding a staff member for advice feels like you're a participant in The Amazing Race, and when you do, there's typically a line of customers waiting.

Yesterday was no exception. After finding the right staffer, I was told there were still two customers before me. So, I instinctively opened my phone to scroll LinkedIn and check emails but decided to open ChatGPT. I uploaded a photo of the cracked foam padding next to the concrete on my driveway and asked for repair solutions. We had a quick back-and-forth and ChatGPT provided the pros and cons of different products.

This was quicker. I received more information in a fraction of the time. But Bunnings could do one better. They could augment their staff, not replace them, with AI, leading to a redefined shopping experience.

Picture this: you download the Bunnings mobile app and are greeted by a friendly AI bot asking if you need advice. You can upload photos, speak or type your query. The bot will use RAG to retrieve product information and AI to formulate a response with different product options, pros and cons, and guide you to the correct aisle and shelf.

This will reduce staff pressure to rush between customers, improve user experience, and provide a welcome option for introverts like my wife who hate asking staff for help.

This solution doesn't replace staff, it augments and supports them. It's a win-win. Until Bunnings develops their AI bot FixIt Fiona, Nail It Ned, or AIva the Builder, try ChatGPT.

As I left Bunnings, I wondered if the future of retail isn't just about products on shelves—it's about seamlessly integrating AI into the shopping experience.

Bunnings, you've got the hammer. AI is the nail. It's time to build something revolutionary.

Subscribe now

Pretty deep for an AI.

This marks the step into the next wave of AI advancements. Current AI LLMs work similarly, outputting the most likely next word based on what came before it.

It’s a bit more complicated than that and can be controlled with variables like weights and fine-tuning, but OpenAI’s latest models, o1-preview and o1-mini-preview, have been programmed to ‘think’ before providing their output.

This process finally solves the viral problem showing LLM models could not count how many r’s are in the word strawberry.

These new models can.

Counting r’s in “strawberry” may not sound impressive, but it’s the start of AI being able to reason and really think through a task before simply outputting an answer. It’s better at maths, data analysis, and processes that require steps.

But the big question is: can we leverage this ‘thinking’ technique to other AI LLMs?

The answer is yes. Kind of.

A prompt shared with a new model Reflection70B asked the AI model to ‘think’ through the process, output an answer, and then ‘reflect’ on its thinking and output to identify any mistakes.

This is a technique called Chain of Thought prompting, which has variations such as ‘think through this process step by step before answering’. Techniques like this consistently produce better outputs, so it’s exciting to see AI companies embedding these thought processes into their models natively. This will give new users a better, albeit slower, experience using AI.

Thinking takes time.

Experiment and have fun with these new models and the Reflection prompt below.

It’s a quick win.

Reflection prompt:

You are a world-class AI system, capable of complex reasoning and reflection. Reason through the query inside <thinking> tags, and then provide your final response inside <output> tags. If you detect that you made a mistake in your reasoning at any point, correct yourself inside <reflection> tags.

So, she turned to AI. She asked ChatGPT to help her craft a kind breakup message, requested it in simple, teen-friendly language, and then edited it to sound just like her.

No one taught her how to do this. She just gets it.

Watching her use AI so naturally reminded me of how quickly kids adapted to smartphones. They didn’t need instructions; they just knew. Today’s teens are the same with AI—they’re growing up with it, integrating it into their lives in ways that feel effortless.

And here’s the thing: these AI natives aren’t waiting for us to catch up. They’re moving forward, fast.

If you haven’t started your AI journey yet, now’s the time. Because if you don’t, you might find yourself one day asking your grandkids how to use ChatGPT 9.5o Turbo Max Extreme.

As I watched my daughter navigate this experience with the help of AI, I couldn't help but reflect on the broader implications. Are we witnessing the birth of a new era in human-AI interaction? How will this shape the way future generations approach communication, problem-solving, and even emotional intelligence?

These are questions we'll need to grapple with as AI becomes increasingly woven into the fabric of our daily lives. But one thing is clear: the future isn't just coming—it's already here, in our homes, our classrooms, and our children's hands.

Subscribe now

I help organisations struggling to adopt AI. The struggle is not that they don't care - quite the opposite. They're overwhelmed by the possibilities, paralysed by the choices of where to begin.

But what if we could overcome that paralysis with just two words?

"How can..."

This phrase is more than a conversation starter. It's a mindset shift. It moves us from passive observers of AI's potential to active explorers of its applications. It transforms overwhelming possibilities into focused opportunities.

Think about it. "How can AI improve our donor engagement?" sparks creativity. "How can we use AI to streamline our volunteer coordination?" prompts problem-solving. These questions invite curiosity, experimentation, and innovation - the very qualities nonprofits need to thrive in our rapidly changing world.

But here's the kicker: you don't need to be an AI expert to ask these questions. In fact, AI itself can help you answer them. Imagine this:

You're a fundraising coordinator intrigued by AI but unsure where to start. You open ChatGPT and type:

"Hi, I'm a fundraising coordinator for a local wildlife sanctuary. How can I leverage AI to improve our fundraising efforts?"

Within seconds, you're presented with a list of ideas tailored to your specific role and organisation type. Maybe it suggests using AI to analyse donor data for personalised appeals or an AI-powered chatbot to answer common donor questions.

AI has its challenges, but setbacks can lead to unexpected breakthroughs when approached with curiosity.

Take the case of a nonprofit that tried implementing an AI chatbot on their website. The chatbot sometimes hallucinated - providing invalid URL links. Instead of abandoning the project, they asked, "How can we still use this technology?" They ended up repurposing the chatbot for internal use on their Slack channel, helping staff quickly draft social media posts based on the content it was trained on. It still sometimes hallucinated but this now presented an opportunity to the team to identify gaps in their online resources. An initial "failure" became an innovative solution.

So, I challenge you to embrace the "How can" mindset. Start small. Be curious. Don't be afraid to experiment. Write down three "How can" questions for your organisation. Then, take them to an AI tool like ChatGPT. You might be surprised by the opportunities it creates.

Remember, in the world of AI, we're all learners. There's no need to have all the answers. Sometimes, all we need to do is ask the right questions.

I expected non-profits to jump at this technology. It offers precisely what they've been seeking: vast knowledge, expertise, time-saving capabilities, and affordability. The value these tools provide for a mere $20 a month is unprecedented in the software world.

Yet, adoption remains slow, as evidenced by numerous reports.

I understand the challenges non-profits face. A friend shared a meme that perfectly illustrates this: cavemen pushing a cart with square wheels, too busy to consider the round wheels being offered. This sadly reflects the reality in the non-profit sector.

The beauty of AI adoption lies in its vast utility. It doesn't require a complete organisational overhaul. You can gradually inject AI into your workflow, using it as a tool throughout the day. It's not like replacing your entire CRM or team structure; it's an add-on to enhance your existing processes.

I encourage non-profits to seize this golden opportunity. Early adopters will gain a significant advantage. AI is an exponential technology – the longer you use it, the more impactful it becomes. As you advance, the technology advances, creating a powerful compounding effect. Waiting a year to start means falling far behind.

Start small. Use AI to generate recipe ideas from a photo of your fridge contents. Ask for ways to improve workplace engagement. Create personalised thank-you songs for staff. Have fun with this technology – that's whats missing. People often get bogged down in tools and prompting techniques. But fun can be the ultimate entry point to AI.

By using AI playfully and curiously, we naturally develop our knowledge. The exciting aspect of AI is that no one knows everything about it. We're all learning daily how to achieve the best outcomes. This means everyone has the opportunity to be an explorer and discoverer in this space.

Consider the research showing that asking AI to "take a breath" before responding yields better results. Who would have thought to ask a computer system without lungs to breathe? This exemplifies the importance of experimentation. We may find some approaches don't work, but we'll be amazed by those that do.

The non-profit sector can be leaders of innovation in this space by experimenting, exploring, and sharing ideas and experiences. If many engage, the entire sector will elevate.

My vision for NFPs.AI is to spark curiosity, encourage engagement, and promote the sharing of use cases and experiences. Start by bookmarking ChatGPT, adding it to your startup windows, and downloading the app to your phone's home screen. Use it once or twice daily for tasks like writing bedtime stories for your kids. You might find it opens up worlds you never imagined, helping you fulfil your mission more effectively.

Non-profits are filled with extraordinary, passionate individuals dedicated to making a difference. Now, they have an invaluable ally – a tireless, knowledgeable assistant ready to help in countless ways.

All we need to do is ask.

Me, Myself, Claude Sonnet 3.5 and Lex (dot) Page

The goal was to create a tool that could transform text into plain language. You know the confusion after reading an impressive sentence, needing a thesaurus and dictionary to decipher it. Take a moment to experience this sentence: "We frequently employ ornate and elaborate terminology to delineate concepts, when the utilisation of straightforward, unembellished English would be markedly more suitable and render the discourse more comprehensible and accessible to a broader audience." When really, I could just say, "We often use fancy words when simple ones would work better and be easier for everyone to understand."

The tool did its job, but I instructed it to shorten lengthy paragraphs, and the results were inconsistent.

The culprit? That one word: lengthy. To me, it meant long-winded paragraphs that needed trimming. But to an AI, lengthy is subjective. My definition could differ from yours, ChatGPT, or Claude’s.

By removing that one word from my prompt, the problem almost vanished. It was a lightbulb moment: every word carries weight.

The words we use with AI are as important as the words we choose when talking to people. The parallels are clear. Objective instructions are vital whereas ambiguity can lead to misinterpretation and unexpected results. In our increasingly AI-integrated world, improving our communication skills - both with humans and machines - is more crucial than ever.

So, the next time you're crafting a prompt or explaining a complex idea to a colleague, remember: choose your words carefully. In the dance between human and artificial intelligence, language is our rhythm - and mastering its nuances is the key to a masterful performance.

Me, Myself, Claude 3.5 Sonnet, Lex (dot) page