Insights
What we know, what we
see, what we believe.
Download
Get the full report
The PDF, formatted for sharing with your team or presenting to the board.
No spam. One email with your download link.
Thanks. Check your inbox for the download link.
Contact The Five
Enablers
What it actually takes to accelerate your AI programme.
Most organisations aren't behind on AI because of the technology.
The tools exist. The budget is there. The board is asking questions. But somewhere between intent and impact, things stall. Pilots that worked quietly die. Confidence stays uneven. Usage spikes and drops. The gap between "we're doing AI" and "AI is actually changing how we work" stays wide.
After building AI transformation programmes for Arla and TeamViewer, we've identified five things that determine how fast any organisation moves. They're not technology questions. They're organisational ones. When all five are working, progress accelerates. When even one is missing, it stalls.
These are the five enablers.
01
Vision
A large retailer we worked with had an AI vision that survived three budget cycles.
Not because it was ambitious. Because it was specific. The vision was simple: make everyone more valuable. It was anchored to a real constraint, growth targets that couldn't be met through headcount alone, and it connected directly to what the CFO was already tracking. Every AI investment could be evaluated against it. It didn't require translation.
That's what a working vision looks like: simple enough that anyone in the organisation can explain it, tied to a business priority the board cares about, with KPIs that prove whether it's working.
McKinsey's research found that 92% of organisations plan to increase AI investment in the next three years. Only 1% have reached what they define as AI maturity. The gap isn't ambition. It's the absence of a vision clear enough to drive decisions and survive a budget cycle.
The organisations that stall are usually the ones trying to "AI everything." The ones that accelerate pick something specific and strategically important, build the metrics around it, and embed it into every initiative. The vision doesn't have to be grand. It has to be useful.
02
People
When we survey teams at the start of an AI programme, the same distribution appears almost every time.
Around 15% of people are already running ahead. They've found their own tools, built their own workflows, and are generating real output. Enthusiasts, and genuinely valuable. But there's a danger at both ends of that curve. Overfocus the programme on them and you build something that works for 15% of the team. Leave them unsupported, and well-intentioned individuals start experimenting without structure, building things that aren't consistent, governed, or scalable.
The work is in the middle. Moving the early mainstream and the lagging mainstream. Building confidence. Demonstrating practical value. Creating enough visible momentum that the sceptics follow.
What leaders model matters more than what they mandate. The two metrics we track to know whether something is actually changing: time spent with AI tools, and confidence. Not adoption headlines. Not pilot numbers. Those two.
03
Process
Every organisation has a graveyard.
Pilots that ran. Experiments that showed results. Then nothing. You ask what happened to the best one that never went anywhere and you almost always find the same answer: a governance gap, an ownership question, a sequencing mistake. Not a technology failure.
McKinsey's State of AI research found that nearly 70% of organisations are piloting AI. Fewer than 20% have scaled anything enterprise-wide. The graveyard is full everywhere.
What most organisations don't see going in: the first job in Process isn't finding the right pilot. It's asking whether your processes are documented at all. Most organisations discover at this stage that their best workflows live in the heads of experienced people. Implicit knowledge walks out the door at 5pm. You can't give a system context it can't access. Codification before automation. That's the order of operations.
There's something else worth knowing about killed pilots. In our experience, an experiment that looked like a failure often becomes the foundation for something much bigger twelve months later. The thinking doesn't disappear. A structured approach just makes sure you extract the learning before the people involved move on.
The test of this enabler isn't how many things you're trying. It's how many are embedded in daily workflows, used by the whole team, and generating learnings whether they succeed or not.
04
Technology
The official technology picture almost always surprises the senior team.
There's what IT has sanctioned. There's what people are using on their own. And then, reliably, there's something someone built over a weekend that six colleagues have quietly adopted because it works better than anything on the approved list.
BCG, presenting at the WFA Global Marketer Week in Stockholm in April 2026, reported that 70% of the effort needed to close the AI gap is change management, not tools. Most organisations are underinvesting in one and overfocusing on the other.
The Technology enabler works across three layers. First, enterprise tools. Microsoft, ChatGPT Enterprise, the platforms already paid for and approved. The productivity gains available here are consistently underestimated. This is where scale starts. Second, custom tools, built for specific teams or workflows, often grown from a successful pilot. These need to be managed like products, with a named owner, an engaged user community, and usage metrics tracked week to week. Third, shadow IT.
The reflex is to treat shadow IT as a risk. The smarter response is to treat it as a signal. People using personal tools instead of approved ones are usually doing so because the approved options don't meet their needs. The colleagues experimenting on their own time are often the first to tell you what to ask for next. Encourage it as learning. Draw the line only where company data is going into unmanaged systems.
05
Governance
Governance is usually the silent killer. Not through malice. Through slow reactivity.
When a grey-area question takes three weeks to get an answer, people stop asking. They stop using, or they use and don't tell anyone. Neither is a good outcome for an organisation trying to move with speed.
McKinsey's diagnosis across thousands of organisations is consistent: the primary barrier to AI impact is "not a technological problem, but one of governance."
The difference between governance that accelerates and governance that creates friction usually comes down to where it lives. Low-maturity programmes govern by memo from legal or brand: "thou shalt not." It creates friction without safety. People work around it. High-maturity programmes govern at team level, with short, memorable principles that people can actually act on.
A working example: Arla's in-house agency, the Barn, runs on one AI governance principle for creative content: no fake cows, the food is always real. Everyone on the team remembers it. It answers a real question about AI-generated imagery. It doesn't require a 40-page policy.
One distinction that proves useful in almost every governance conversation: what is playbookable, and what needs a human? Some questions have a definitive answer that can be built into a system. Others involve judgement, context, and brand risk that no rule can fully capture. Helping a team draw that line is some of the most practically useful work in an AI programme.
Governance isn't written once. It's updated as new speed bumps appear and new things become possible.
What the five enablers are for
They're not a readiness checklist. They're not a maturity model you file away.
They're the five things that, working together, allow an organisation to accelerate. To move from "we're doing AI" to "AI is doing what we actually need it to do."
Most organisations have some of them. Few have all five. The ones that stall usually know where the gap is. They just haven't said it out loud in the same room.
The Deft AI Diagnostic maps where you stand across all five. 10 days. Fixed price. Board-ready outputs. You leave with a clear picture of what's working, where the friction is, and the three changes that would move you fastest.