Why most AI transformations will fail, and what to do instead
Curia AI | AI Advisory for Charities | Introduction to the Responsible AI Accelerator
Over the last few months I have found myself asking a slightly different question in conversations about AI.
Not whether organisations have an AI strategy, but whether they have an AI operating model.
By operating model, I mean the structures, roles, processes and rhythms through which an organisation continuously identifies, governs, deploys and learns from AI.
Strategies are beginning to emerge across the sector. Working groups are forming. Policies are being drafted. Pilots are underway.
What is much harder to find is an agreed approach for how organisations build capability, develop governance, capture learning and prepare people to oversee AI as it becomes more embedded in services, fundraising and operations.
That gap, between recognising the opportunity and building the organisational capability to pursue it responsibly, continuously and at pace, may prove to be one of the defining challenges facing charities over the next decade.
Importantly, this does not mean charities should pause experimentation until they have designed the perfect operating model.
Quite the opposite.
For many organisations, the right place to start is with a small number of practical pilots that solve real problems and generate evidence of value.
The question is whether those early experiments remain isolated initiatives, or whether they become the foundations of a broader organisational capability.
And that is where the dominant transformation model may be poorly suited to AI.
Equally, the answer is not to replace one heavyweight methodology with another.
Most charities do not need a comprehensive AI strategy before they begin experimenting, nor do they need sophisticated governance structures before they have evidence that AI can create meaningful value.
They need a pragmatic approach. Enough governance to move safely, enough structure to learn from experience, and enough momentum to keep moving.
In practice, that often means starting with a small number of carefully chosen pilots, building confidence alongside capability, and allowing governance, literacy and operating practices to mature as adoption grows.
The model that no longer fits
For two decades, technology transformation has often followed a familiar playbook. Define a target operating model. Describe the future state. Map the gap between where you are and where you need to be. Run a programme to close it. Hand over to business as usual.
That approach made sense when organisations could reasonably articulate the destination they were designing towards. Enterprise resource planning systems, CRM programmes and many digital transformation initiatives were complex undertakings, but they generally assumed a relatively stable target state, even if the route to get there was challenging.
AI appears different.
AI is not simply a technology that is implemented and then operated. It is a capability that must be continuously developed because the technology itself, the models, tools, risks, regulatory expectations and potential applications, is evolving faster than many organisations can realistically track.
The challenge is not that charities are moving too slowly. It is that many of the methods organisations have traditionally used to manage change assume a stable destination. AI asks something different of us. It asks organisations to become better at adapting.
Research increasingly points in this direction.
McKinsey’s 2025 global survey found that AI adoption is becoming widespread, but meaningful organisational impact remains difficult to achieve. The organisations generating the greatest value tend to be those investing not only in technology, but in operating models, skills, governance and organisational change.
The nonprofit sector appears to be following a similar pattern. The 2026 Nonprofit AI Adoption Report from Virtuous and Fundraising.AI found that 92% of nonprofits are using AI in some capacity, yet only 7% report major improvements in organisational capability. Much of that use remains individual and informal, with relatively few organisations having established common workflows, governance mechanisms or shared approaches to learning.
Individual staff members are becoming more productive.
That does not necessarily mean organisations themselves are becoming more capable.
The problem is not the technology.
The problem is the container.
Building capability over time
Traditional transformation models assume organisations can define a destination and then work systematically towards it.
AI requires something slightly different.
It requires organisations to build the capacity to continue moving responsibly, even as the destination itself shifts.
That means developing three things in parallel rather than in sequence.
First, principles.
Not a policy document that sits in a folder, but a living set of commitments about how AI will and will not be used, who is accountable for what, how risks are assessed proportionately and what it means to maintain trust with beneficiaries, donors, regulators and communities.
Principles that are owned by trustees and understood by frontline teams, not simply drafted by consultants and filed away.
Second, capability.
The ability to identify worthwhile opportunities, test them safely, learn from experience and embed what works so that each governed deployment makes the next one easier, safer and less expensive.
Not a programme with an end date.
An operating rhythm that compounds over time.
And third, people.
Because one of the questions many charities have not yet had to answer, but may soon need to, is who will govern the increasingly autonomous systems that are beginning to emerge.
Some charities are already experimenting with AI-assisted triage, drafting and recommendation systems.
Today these may be relatively contained applications.
Tomorrow they may become systems that influence fundraising communications at scale, service delivery decisions, identify beneficiary needs, prioritise interventions or support increasingly complex operational activities.
The direction of travel appears clear.
As AI becomes more embedded within organisations, the people overseeing those systems will need something more than digital literacy.
They will need judgement.
They will need principles.
They will need practical experience of governing AI in real conditions.
And they will need to begin developing that capability before those systems become commonplace.
That capability gap may become one of the most important organisational challenges of the next decade.
Why we called it Curia
I have spent much of my career working at the intersection of data, technology, privacy and trust.
One observation has remained remarkably consistent.
The organisations that navigate periods of change most successfully are rarely the ones with the largest technology budgets or the biggest compliance teams.
They are the ones that develop genuine governing capacity.
The collective judgement to make good decisions on behalf of the communities they serve.
And the structures that ensure those decisions remain accountable to mission and purpose.
That is why the name Curia is not accidental.
A Curia was one of the governing institutions of civic Rome, a forum through which collective judgement was exercised and public affairs administered on behalf of the wider community.
In the Roman Catholic tradition, the Roman Curia provides continuity of governance, helping sustain the mission and administration of the Church beyond the tenure of any individual office holder.
Across these traditions, the common thread is remarkably consistent. Institutions endure because they develop mechanisms for collective judgement, continuity and accountability.
Good governance is not a constraint on action. It is what makes sustained, trusted action possible.
That is what Curia AI exists to help charities build.
Not a transformation programme.
Not a governance framework bolted on after the fact.
But a governing model for responsible AI adoption, one that develops principles, capability and, above all, the people who will govern AI today and oversee increasingly autonomous systems tomorrow.
What this looks like in practice
The Curia AI Responsible AI Accelerator is built around this argument.
It is designed to help charities move from isolated experimentation towards governed, repeatable and scalable AI adoption, supporting organisations at every stage of the journey, from their first pilots through to becoming Frontier Charities capable of deploying AI responsibly, repeatedly and at scale.
At one end of the journey may be a charity exploring its first three AI pilots.
At the other may be an organisation embedding AI into fundraising, services, operations and decision-making.
The objective is not to prescribe a fixed destination.
It is to help organisations build the principles, capability and people needed to move confidently through each stage.
The Accelerator comprises six phases, from early experimentation through to frontier leadership, but it is not intended to be understood as a rigid maturity model.
Instead, it is an operating approach that seeks to create value and build foundations in parallel, strengthen internal capability at every stage and ensure that knowledge remains within the organisation rather than leaving when external support concludes.
In practice, that may mean identifying three safe pilots, training trustees alongside delivery teams, establishing proportionate governance mechanisms and creating reusable patterns so that every deployment becomes easier than the last.
It begins with proportionate discovery, recognising that some organisations are exploring their first use cases while others are already thinking about AI at scale. The objective is not perfection at the outset, but to build capability progressively, ensuring that governance remains proportionate, something I call minimum viable governance (MVG), to the opportunities being pursued and the risks being managed.
It progresses through pilots and initial deployments, building evidence and confidence together.
It develops the foundations needed to scale what works, from literacy and governance through to data readiness and organisational learning.
And throughout the process it seeks to develop the people who will become an organisation’s AI governors.
Not technical specialists.
But leaders equipped with the judgement, confidence and principles needed to oversee increasingly autonomous systems on behalf of the communities they serve.
The organisations that benefit most from AI over the next decade are unlikely to be those that commissioned the largest transformation programmes.
They are more likely to be the organisations that built the governing capacity to keep learning, keep adapting and continue earning the trust of the people they exist to serve, even as the technology evolves around them.
A frontier charity is not the one with the largest AI budget.
It is the one that built the governing capacity to use AI responsibly, repeatedly and at scale.
In other words, it built its Curia.
Sources: McKinsey & Company, The State of AI in 2025, McKinsey Global Survey, 2025. Virtuous and Fundraising.AI, 2026 Nonprofit AI Adoption Report, March 2026.
