I am deeply suspicious of transformation programmes. Most are bought, not built. Most sell the same comforting illusion: buy the right framework, operating model or, the current favourite, AI, and you’ll emerge transformed. You won’t. In the age of AI, that illusion is dangerous.
AI will not rescue your organisation from itself. Neither will the latest framework, a rebranded generic operating model or a consultancy’s twelve-step methodology. What determines whether you thrive is the conditions inside your organisation: whether people can tell the truth, whether small multidisciplinary teams are trusted to do the work and learn from it, and whether your funding model lets you act on what you’ve learnt, fast. The litmus test: when a team learns something, does it just fix the service, or call a meeting about it?
AI is an amplifier. If your organisation already works in small, empowered teams, close to users, learning fast, AI will make you extraordinary. If it’s a mess of silos, handoffs and theatrical governance, AI will make you an extraordinary mess, faster. Air Canada’s chatbot didn’t hallucinate a bereavement policy because the model was faulty. It hallucinated one because the organisation around it was.
Every UK boardroom should sit with this example. In early 2025, Octopus Energy overtook British Gas, the world’s first public utility, chartered in 1812, to become the UK’s largest energy supplier, barely a decade old. The lazy explanation is Kraken, Octopus’s much-admired technology platform. It is genuinely good: cloud-based, continuously deployed, real-time data flowing everywhere.
But here’s the tell. Kraken is now an independent business, licensed to anyone, including Octopus’s competitors. If the platform were the secret, the advantage would have evaporated. It hasn’t. The edge, Octopus’s own leaders admit, isn’t the technology. It isn’t even AI. It’s how the company is set up: small groups having real conversations, trusted to take action, learning fast from feedback. Octopus’s AI tools for frontline staff, summarising interactions and drafting responses, amplified a culture that already worked. The transformation wasn’t outsourced to AI; people were trusted to experiment.
It works in tightly regulated industries, too. In 2015, Dutch bank ING decided to move at the pace of its customers, borrowing from the “Spotify model” of squads and tribes. It didn’t copy it; it interpreted it, adapting it to banking’s compliance realities, piloting in less critical domains, evolving as teams learnt. A decade on, ING ranks among the world’s best banks. When generative AI arrived, the same instincts kicked in: start small, bring risk people in from day one, build bank-specific guardrails, test with a handful of real customers. The polar opposite of Air Canada. Principles provide alignment; frameworks provide compliance.
Cynics will say: fine for energy scale-ups and Dutch banks, impossible in government. Wrong. In 2025-26, the Government of Alberta needed to replace two creaking legacy systems tracking 4,000 government-owned buildings. The quoted procurement route: $54m and four years for just one of them. Instead, a small in-house team, amplified by AI, did both in ten months for $2.64m. AI vision models turned 50 hours of user screen recordings into build specifications in minutes; 300 throwaway prototypes in a fortnight tested ideas with real users.
The kicker, from Alberta’s minister Nate Glubish: “It might look like an overnight success story. But it took 7 years.” Seven years of guardrails, data governance, AI training, and a culture, in risk-averse government, where experimentation was rewarded, not punished. The conditions came first. The AI miracle followed.
So what should leaders do? Not what most do. McKinsey’s own research puts transformation success rates below 30%. Each failure leaves wreckage: eroded trust, departed talent, cynical stakeholders. The alternative is slower, harder and more useful. Start with principles, not methodology. Build trust deliberately: an operational imperative, not an HR nicety. Find a pilot team and fiercely protect it. Spend its first 90 days attacking friction. Then scale the conditions, not the process, learning from resistance, not steamrollering it.
None of this is new. A stubborn crowd of us have been banging on about it for twenty years. What’s new is the urgency. The gap between organisations with the right conditions and those without is about to be amplified beyond recovery. Whitehall leaders commissioning “AI transformation programmes” from the usual suspects should read the Alberta story twice, then quietly cancel the procurement.
You don’t need a transformation programme. You need transformation capability. Not a blueprint. A muscle. And the only way to build a muscle is to use it.
Public Digital has recently published The Intelligence Era Organisation: Creating the Conditions to Thrive in the Age of AI, a guide for leaders navigating organisational change in the age of AI.
Tom Loosemore is co-founder of GDS and founding partner at Public Digital.
