WIRED's Katie Drummond asked Dan Swan, senior partner at McKinsey and global leader of the Tech & AI practice, a simple question: what separates companies that are actually making progress with AI from those that aren't. The answer was sobering. Most companies aren't transforming. They buy software, roll it out across the organisation, and call that transformation.
It's a statement that should land uncomfortably for many executives. Not because they're doing nothing with AI, quite the opposite. Most organisations by now have bought a licence, have a pilot running, or have a chatbot in place. But according to Swan, that's precisely the problem. Getting AI transformation right means rewiring how people work, how workflows run, and how leadership makes decisions. Everything else is buying tools and hoping they apply themselves.
As it happens, what Swan draws on is called Rewired, McKinsey's playbook on what makes transformation succeed and what doesn't. The first edition was about digital transformation in general, the latest about AI. The insight connecting both: what used to hold true for successful digital transformation holds equally true for AI. It isn't primarily about technology. It requires people.
Why most AI initiatives stall
McKinsey describes AI maturity as a four-rung ladder. The first rung is asking a language model a question, the way you use ChatGPT or Claude today. The second is embedding a language model into existing software, think automatic summaries of a Teams or Outlook meeting. The third and fourth rungs go a level deeper: fundamentally redesigning an entire process, or building a new way of doing business that couldn't have existed without AI.
Most companies never get past rung two. That's no accident. Rung two is democratised. Every employee with a Microsoft or Google licence benefits from it immediately. It requires no difficult decision from management. No process that needs overhauling, no team that has to work differently, no investment anyone has to defend. That's exactly why it's a trap. It feels like progress, it generates enthusiastic quotes for the intranet, but it never shows up on the P&L.
Three principles that actually make the difference
From Rewired, Swan distils three principles for getting past that trap. The first is about focus. Choose two or three domains, and go deep with a full process, rather than spreading AI across dozens of separate applications. The historical failure pattern is well known: small pilots that never scale, never become visible on the P&L, and ultimately mostly generate frustration among the people who put time into them.
The second principle is going deeper than speeding up: fundamentally changing a process, or building something new, not just making the same process run a bit faster. The third is a matter of ownership: treat it as a genuine business transformation, not a technology transformation. The owner of the business problem sits in the driver's seat, not the IT department.
Those three principles converge in a single question, and it's the most important one from the entire interview. Not: where can I apply AI? But: what is the business problem, and how can AI help solve it? It sounds like a detail. It's the difference between a technology project and a business project.
What this means at the scale of a Flemish scale-up
The examples Swan cites play out at companies with tens of thousands of employees. At Toyota, a VP once asked out loud: if Amazon can track a $10 book, why can't we track a $40,000 car? The answer became a digital twin of the entire supply chain, which today processes a billion decisions in 15 minutes, work that used to take 100 people and 75 spreadsheets. Estimated value: around $800 million.
For an organisation of 50 to 500 knowledge workers, that scale isn't the point. The underlying principle is. Not launching ten AI experiments at once across ten departments, nine of which quietly die off after six months. Choosing one process that genuinely matters, order processing, customer service or planning, for instance, and rethinking that process thoroughly with AI as the starting point, not as a layer on top of what was already there.
Translated to that scale, it comes down to five points of attention.
Choose one process, not ten experiments. Scattered pilots feel safe, because no one loses much if they fail. That's exactly why they rarely produce anything that sticks. An organisation that tries a bit of AI everywhere ends up measuring the difference nowhere.
Reverse the question. Not "where can we deploy AI", but "what costs us the most time, money or frustration today, and how can AI solve that". That reversal moves the conversation from the IT department to the boardroom table. A technology question produces a technology project. A business question produces a business result.
Turn your people into co-builders, not just users. At Toyota, sales, manufacturing and planning experts were at the table from day one. Not as end users who had to test it later, but as co-builders who went on to earn the trust of the rest of the organisation. Whoever only approaches the people on the floor at launch has already built the thing without them.
Steer from above, not beside every step. McKinsey calls this "humans above the loop" rather than "humans in the loop". The human sets direction, safeguards quality and steps in where needed, instead of standing beside every single step checking it. Standing beside every step to check it automates nothing. It just adds an extra layer of control.
Expect growing pains that get smaller. DBS, the bank from Rewired, started with two or three priority domains, and only then expanded to four, five, six. Each following wave moved faster, because the organisation had already built up capabilities: data in order, people who can work with agents, leaders who know how to steer that kind of journey. The first time is the slowest. That's not a counterargument, that's the investment.
The real lesson
Swan is clear about what ultimately determines returns, and they aren't technological factors. Process owners need to take full ownership: the initiative has to come from them, not be something that happens to them from an IT roadmap. On top of that, capability building is needed among the people who'll actually use the tools. Without that, the investment sits on the shelf, no matter how good the technology is. And finally, measuring whether you're achieving the intended impact, and if not, finding out why.
HR plays a bigger role in this than most organisations realise. Not because HR needs to understand the technology, but because mindset and communication, capability building, and how work itself gets organised, are all HR territory. The skills a team needs today already look different a year from now.
What lasts, according to Swan, is durable. Companies that get in now and learn the lessons of this first wave build an advantage that doesn't disappear the moment a competitor buys the same software.
It's precisely that logic, business problem first, ownership with the client, that Rescope has built its own 3R approach on.
The question Swan puts to business leaders is the same question every organisation of 50 to 500 knowledge workers should be asking itself. Not: do we have AI? But: have we actually rewired our people, our processes and our leadership, or have we just bought software?