For years, I have helped organisations transform in so many different ways: new business models, new operating models, new products, new teams. Now, though, there seems to be only one single, giant project on everyone's agenda: What are we doing about AI?
Every leadership team I talk to is somewhere between excitement and quiet panic. And the temptation is stronger than ever to treat AI as something completely new. A separate track. A special team. A shiny project.
I always try to push back on that.
AI might be the most powerful general-purpose technology of our lifetime. It will reshape jobs, industries, and the rhythm of daily work. But after spending the last months studying the research and working with companies deploying it, I am more convinced than ever: The hard part of transformation has not changed.
Why do I think that?
A recent Stanford study (source below) looked at over 50 enterprise AI deployments that actually delivered measurable business value. The authors expected to find a story about technology. Instead, they found something else:
77% of the hardest challenges were invisible: change management, data quality, process redesign. Not the model. Not the technology stack.
61% of successful projects had a prior failed attempt. The difference between the failure and the success was almost never the technology. It was the organisation.
Same use case, same AI models, vastly different timelines. One company took weeks. Another took years.
Read that again. The difference was never the model. It was sponsorship. Process maturity. Data access. Willingness to change. Permission to fail. The same things that have always separated the 12% of transformations that work from the 88% that don't.
That is actually good news. It means we already have most of the tools we need to navigate this. The rules have not changed. If anything, they matter more than ever. Because AI doesn't just accelerate what you do well. It amplifies everything, including the misalignments, the unclear ownership, the processes nobody has questioned in years.
“AI accelerates what works - and exposes what doesn’t.”
So before we talk about models or use cases – here are the five things that will actually determine whether your AI transformation works.
1. Fix the process before you add the technology
Every organisation has a pile of old workflows that have accumulated over years. Layer AI on top of a broken process, and you get a broken process that runs faster and costs more. Again, the Stanford researchers found that companies who succeeded with AI almost always had to rebuild the underlying workflow first. The ones who tried to shortcut this step ended up in the 88% that failed.
This is not glamorous work. Nobody puts "documented the invoicing process" on their LinkedIn. But this is where transformation actually happens – or doesn't.
2. The invisible work is the work
Process documentation. Data architecture. Change management. Conversations with the compliance team. Coffee with the concerned middle manager who will decide, quietly, whether this project lives or dies.
Everyone underestimates how much of this is needed. Every time. I have stopped trying to convince leadership teams that the 20% they see – the pilot, the dashboard, the demo – is only a small part of the work. I built it into my transformation programs and make sure it gets done in the background. By the end, the 80% they never saw is the part they can't stop pointing to.
3. Resistance is not the enemy. Surprise is.
The biggest source of resistance to AI is not the teams who actually (need to) use it. In most cases it’s in the staff functions: Legal, HR, Risk, Compliance. (Btw: One more reason why AI transformation does not sit right in the IT department.)
This is not because these people are obstructionists. It is because they carry risk for the organisation and have been trained to slow things down when they see uncertainty. Bring them in at the end, and they will block you. Bring them in at the start, give them a role in shaping the guardrails, and they often become your strongest advocates.
The same is true for middle managers. They are not afraid of AI. They are afraid of being held accountable for outcomes they don't understand, with tools they haven't mastered, on timelines they didn't set. Resistance in my experience is almost always a rational response to a legitimate concern that was never addressed.
4. Leadership is a verb, not a title
Sponsorship is the single biggest predictor of whether an AI project succeeds or stalls. But "sponsorship" is not approving a budget and checking a quarterly slide. The sponsors who actually move things forward do four things consistently:
They clear blockers before teams escalate.
They follow the process.
They tie AI adoption to real incentives. (OKR, compensation logics)
They protect teams and define space for experiments and failure.
That last one matters more than the other three combined. In every successful project I have seen and every one the Stanford researchers studied, if the same executive stayed through the failure there was a high likelihood this person led the successful second attempt.
5. Keep it lean. Especially now.
Transformation has never needed more bodies. It has always needed the right ones. In the AI era, this is even more true. A small team equipped with the right tools, clear priorities, and a sponsor who protects them will outperform a large programme with governance committees and workstreams every single time.
This also puts more pressure on the individual talents. The ones that will thrive in this new era, are the ones that are able to consolidate various skills within a single role. Thinking more holistically and entrepreneurial is becoming key (T-shaped 2.0).
What I am not saying
I am not saying the technology is easy. Building reliable agentic systems, managing model choice, protecting sensitive data – these are all genuinely hard engineering problems and they require real expertise.
What I am saying is this: the technology is the part that you can define a budget for. You can buy the model. You can hire the engineers. You can subscribe to the platform. What you cannot buy is the willingness of your organisation to let go of old ways, the clarity of purpose behind your initiative, the trust of your teams, and the patience to do the invisible work.
Those things are still built the old way.
One honest conversation at a time.
Where this leaves us
The companies that will win the next decade will not be the ones with the best models. Instead, it will be the ones that built the muscle to keep evolving with the technology instead of against it.
That is the work. That has always been the work.