How AI transformation might play out in charities… if we’re not careful

AI will reshape the work charities do faster than charity strategy cycles can respond.

Since ChatGPT was launched in late 2022, AI use has ballooned, and this is already being felt by charities. Over time, every stakeholder relationship a charity relies on will be reshaped by other people’s adoption of AI tools.

These are not the effects of a charity adopting AI. They are the effects of the world a charity operates in adopting AI. A charity that set a five year strategy in 2023 would find that by 2028 the gaps their charity fills – and the relationships they have with the people they serve – are being redrawn by AI. All while the charity strategy barely mentions that AI exists. Following normal strategy cycles means the transformation so sorely needed to respond to AI will kick off after the transformation of the operating environment has already happened.

Charities will price the risk of action on AI – but won’t price the risk of inaction highly enough.

Even in the absence of organisational strategies that consider the impact AI will have on how charities work, charities will kick off limited action on AI. Often the first thing on the to-do list is to ask someone to create an AI policy. This goes first because risk mitigation is a space of comfort for charities, and AI is perceived as a risk.

But when the risks of AI are being assessed, the risk of indecision and inaction on AI won’t be assessed. Ultimately, passive inaction in the face of AI is the most expensive choice a charity can make, because you lose control over the way AI is implemented. Even principled decisions to reject AI are better than a lack of decisions that leads to the same outcome (AI not being used): they are deliberate, articulated and they give the charity agency in defining its own relationship with AI.

Charities will frame AI as a skills challenge while staff experience it as a combined capability and identity shift.

Once the need to respond to AI becomes undeniable, one of the first actions a charity takes (alongside a policy) will be to commission training. Training is relatively easy to commission and is clearly important: for staff to use AI they need to be able to brief AI well, judge its output, and know which kind of AI assistance fits which task.

But AI changes the shape of the work. Staff who used to write may spend more time validating what AI has written. Staff who used to research may increasingly be commissioning AI to do it, and checking summaries. This is different work and as a result the deepest question staff may face with AI won’t be can I use the tools? It could be who am I professionally, when AI does part of what I did? This links to scary questions around job security or replacement. Identity questions like these aren’t answered in a one-off training programme. In fact, train people without engaging them around an identity shift and the training just creates fear. Conversely, reassure people around their future without investing in their capabilities to understand AI and the reassurance will ring hollow.

Charities that treat AI only as a skills issue to be solved by training will also find themselves surprised by a dispute over values they may not have bargained for. Many teams will have both AI refusers and AI users: AI refusers may see AI users as compromising their morals; AI users may see AI refusers as slowing work down. A team that already contained different working methods now contains explicit moral disagreement about how the work should be done.

“Who am I professionally, when AI does part of what I did?”

Ultimately, only when charities build holistic and ongoing programmes that respond to AI’s impact on people – and boast visible senior sponsorship and protected time for ongoing learning – will the actions seem credible to staff. AI will affect identity and capability at all levels (including leaders) and may require changes to structure, roles, identities and skills at once. A single training course won’t cut it.

There is an opportunity here though. Charities have needed to have conversations about the shape of work for a long time. Moving from implicit norms about how work gets done to explicit agreements about who does what, what is acceptable, what is expected and what good looks like will help humans work better, not just AI.

Charities will use AI to maintain (and speed up) existing processes.

Previous waves of technology have often seen new processes bolted onto existing ways of working, rather than prompting a rethink. In the last wave, we started creating digital teams to manage new digital channels, and squashed these digital teams into existing structures – without ever thinking about redesigning the work and thinking around the speed, breadth and scale of digital and its impact on our work.

If we try to do this with AI, we will fail. You can’t just sprinkle some AI into existing structures and ways of working and hope for the best. With AI every piece of work will potentially pass through multiple new steps and handoffs – humans using AI, AI prompting humans, AI talking to AI.

When ways of working just evolve to add AI in, it may look the same as before, but the people inside it could slowly be demoted to checking AI’s output – and the number of points of failure will multiply as the steps in workflows balloon. It will be incredibly inefficient.

The path of least resistance is to use AI to do what you already do, a little faster – and to leave the true prize on the table.

We’re already seeing this lack of coordination, to some extent, in the way we experiment with AI. Comms, fundraising, services and ops are running their own AI pilots in parallel. But they aren’t sharing what they are learning, so every team is reinventing – and the charity is losing sight of what’s being tried, and the lessons that might come from it.

The biggest process gains come from removing steps entirely with AI, or expanding the breadth, depth and scale of what a team can do – but the path of least resistance is to use AI to do what you already do, a little faster – and to leave the true prize of AI working on the table.

Charities will undervalue the unglamorous technology investments that will turn out to matter most.

When AI underperforms – and it will, often – the first instinct will be to blame the technology. Sometimes that reading may be fair: the technology is uniquely unreliable. But a more accurate read is often that the underperformance is a visible symptom of a deeper and hidden problem, where a charity has undervalued the importance of infrastructure transformation.

Past experience has shown us how much the least visible investments in technology matter most. Results rely on how clean and well structured the data is. Or how well systems have been connected and integrated. Or who has decided which tools are available to use for whom and why. Or what budget exists for investing in the right tools for the right jobs. This is very likely to be true for AI too.

Where people across the organisation put thought and energy into building the foundations, there will be real opportunity to do things in totally new ways. Where the foundations are messy, the same AI tools will deliver less, more slowly, with more risk. Take one example: your data. What used to be called your data was mostly the structured, machine-readable content held in a CRM or a finance system. But your data now means something far larger: the files, documents, content, knowledge and institutional memory your staff use every day to do the work. When that material lives in inconsistently-tagged folders and half-finished briefing documents, AI produces confident-sounding summaries that are wrong in ways nobody can easily check. Knowledge management – long treated as the unglamorous corner of IT – is best understood as data quality for unstructured content. Without doing it well, charities will struggle to get the results they expect from AI.

There will also be existing challenges that have new dimensions when it comes to AI technology. AI can connect to and build websites, apps and systems in seconds. But with this ability comes security risks. Balancing security standards with the flexibility to create and test new ways of delivering systems isn’t glamorous, but it matters. Charities that under-invest in this balance will find themselves unable to use AI to build the things that could be massively impactful for supporters, staff or beneficiaries – even as the technology to do so becomes cheaper and faster by the month.

There is a welcome consequence to this. AI gives charity leaders the business case for the data and knowledge investments that have always been necessary but rarely funded. Where the argument for cleaner data used to be “it will make reporting easier”, it is now “we will not be able to deliver work without it.”

Charities will try to transform their operating models with AI without first understanding how they actually work.

The three dimensions above – people, processes, technology – are not separate dials. They meet, in every organisation, in one place: the operating model – the real answer to how does this place actually work? Most charities have managed without writing it down. With AI that stops working.

You are wiring a new form of capability – AI that works alongside humans – directly into your work, and if you cannot see how work happens, the wiring is accidental. Different teams will pick different systems, build different AI support tools, curate data differently – and none will be compatible – in fact they will confuse AI and each other. Charities that can see how their operating model works can choose where AI should change it, and do it intentionally. Everyone else gets a steady drip of small fires that all look like AI problems but aren’t.

Readiness for change will be the single biggest predictor of how well a charity adapts to AI.

The Charity Change Collective first started because we were frustrated with not seeing change happen fast enough in the sector, and felt that charities were losing relevance as a result. And now AI speeds up the timelines for change again: ChatGPT reached an estimated hundred million users within two months of launch – the consumer web took years to build that audience.

The muscle of changing deliberately – seeing your organisation clearly, making uncomfortable choices, shifting resources, learning from each cycle – is one some organisations have practised more than others. Those organisations will be best placed to do so again, and we predict they’ll adapt better to AI than others. But even they will need to move faster than before.

Those who create a transformed vision of what their charity could be in the AI age will be best placed to seize the opportunity.

Everything we have talked about so far paints transformation in a reactive way: adapting the organisation you have to the changes AI brings, and naming the traps you’d want to avoid as you do. That reactive transformation work isn’t optional – and avoiding the traps is important while you do it – but it is the floor, not the ceiling, for ambition around AI transformation.

The bigger prize goes to charities that start from the impact they exist to create and work backwards to the operating model AI now makes possible. Ethan Mollick’s pair of questions help:

What valuable thing that you do today is no longer valuable – and what impossible thing can you now do?

To illustrate this, people often talk about factory electrification as a parallel to AI. Most factory owners used electric motors to replace steam engines in the same configuration; the gains only came when someone redesigned the factory around what electricity made possible. Most charities will use AI to do existing things faster. A few will use it to rethink what they are for and how they deliver it.

The charities that engage with AI earliest and deepest will have the greatest chance to influence how AI plays out.

What society does about the ethical and commercial dimensions of AI – energy, water, where training data comes from, job displacement, the political associations of big tech – will be decided over the coming years. The charity sector adapted to social media one charity at a time, and ended up taking platform rules written without it in the room. The same risk exists with AI, and it is more pressing.

The charity sector adapted to social media one charity at a time – and took platform rules written without it in the room.

Right now hundreds of charities are quietly duplicating the same learning behind their own walls. Imagine what would happen if they used their influence to act together – on shared evidence, on procurement expectations, on the terms of engagement with AI vendors. The choice between adapting alone and shaping together is itself a strategic decision. This time, we could choose to make it earlier – and seek to influence the society AI creates as a result. But this will only happen if we proactively choose to engage. Otherwise we will end up disengaged by drift, not by intent.