Who should own AI in your business?

On the line, doing the same job twice has a name. It's called rework, and it's the most expensive word in the business. We're about to do it a third time.

I had a conversation this week with a CMO at a heavy mining company about who shouldn't own AI in their business.

We both landed on IT. Not because IT can't execute, but because AI isn't an infrastructure problem. Park it with the team that keeps the servers running and it gets treated like a tool rollout instead of a change in how the business works.

He made the case for marketing. And in his business that case stands up, because he has a genuine mandate that runs across functions. That's the point. AI ownership works wherever the authority is real, and someone with cross-business backing can run it from marketing as well as anywhere. What stalls it isn't where it sits. It's the absence of that backing. Without a mandate from the top, AI doesn't get owned at all. It lands in whoever's lap is closest, and then it stalls.

If you've run an operation, you know rework when you see it. We've reworked this exact mistake twice already.

Digital, then analytics

Digital transformation came first. McKinsey, BCG and Bain have all put the failure rate high, somewhere in the 70 percent range, with Bain's analysis of business transformations landing higher still. The exact number gets argued over, and fairly, because everyone defines failure differently. But the cause they name is consistent, and it is not the technology. McKinsey's own transformation research puts it plainly: efforts to improve performance fail about 70 percent of the time, and the contributing factors are insufficient ambition, weak engagement across the organisation, and too little investment in building the capability to sustain the change. Companies rolled out the tools, migrated to the cloud, hired a Chief Digital Officer, and called it a transformation. What actually happened was they digitised their dysfunction instead of redesigning how they worked.

Then big data and analytics. New C-suite title, a central data science team or a Centre of Excellence, a "data-driven culture" programme, and a long tail of models that never made it into production. The promise was democratisation. The reality, in a lot of organisations, was expensive activity and very little operational change.

I lived both of these directly, inside a data and analytics software business and across web and digital marketing. The hub-and-spoke Centre of Excellence model burned time, money and effort trying to get "everyone on the journey" while leadership stopped short of actually leading it.

And now AI, with the receipts already in

Same loop, third time around. A capability arrives, a title and a Centre of Excellence get created, an upskilling push brings everyone along, pilots multiply, and most never scale because decision rights and the operating model never changed.

This is not a software-industry quirk. McKinsey's 2025 State of AI survey, looking across industrial sectors including manufacturing, refining and construction, found nearly nine in ten companies now use AI somewhere, yet two-thirds remain stuck in experimentation or pilots, and only 39 percent report any earnings impact at all. The survey authors are blunt about the cause: real value comes from redesigning how work gets done and from executive leadership that goes beyond cheerleading, not from the technology itself.

And this time the correction has come fast, and it is not one firm's finding. Through 2025 and into 2026 a consistent picture formed across independent analysts. Challenger, Gray and Christmas tracked roughly 55,000 job cuts directly attributed to AI in 2025. Forrester found that a majority of employers who made AI-driven cuts came to regret them, and forecast that half of AI-attributed layoffs would be quietly reversed. Gartner projected that by 2027, half of the companies that cut customer service headcount citing AI would rehire for the same functions, often under new titles. Robert Half found that close to a third of companies that eliminated roles to AI had already reopened the exact same positions. The workforce firm Careerminds, surveying 600 HR leaders who had made AI-related layoffs, found more than a third had rehired over half the roles they cut, and that one in three employers spent more on restaffing than they saved. Different methods, same conclusion: the work did not disappear. It shifted, got messier, and needed human judgment nobody had accounted for.

Klarna is the case everyone cites, so it is worth telling accurately. Its headcount fell from about 5,500 to around 3,400 between 2022 and 2024, mostly through a hiring freeze and natural attrition rather than mass redundancy, while an OpenAI-powered chatbot took on the work of roughly 700 customer service agents. The CEO, Sebastian Siemiatkowski, publicly declared AI could already do the jobs humans do. Then he reversed. He told Bloomberg the AI-first approach to support had produced "lower quality" work, said plainly that the company had gone too far, and began rehiring people, though through a flexible freelance model rather than simply restoring what was cut.

None of that is a technology failure. It's an ownership failure. Nobody held the gap between what the tool could do and what the business actually needed.

So why is this time different?

In one way that matters. Analytics needed data scientists. AI doesn't. It's already in every employee's hands, sanctioned or not. So "everyone on the journey" isn't a choice you make anymore. It's the default whether you chose it or not, which makes a clear, accountable owner more necessary, not less.

Here is the trap leaders fall into. Boston Consulting Group, with Columbia Business School, surveyed around 1,400 employees and leaders in 2025. Seventy-six percent of executives believed their people were enthusiastic about adopting AI. Among individual contributors, just 31 percent actually were. Leaders were more than twice as far off the mark as they thought. That gap is the whole problem in one statistic. You cannot own a change you have misread, and "everyone's on board" is exactly the assumption that lets ownership quietly evaporate.

Bringing people along isn't the problem. Literacy matters more now than it did in the analytics era, and most organisations are not there yet. One 2025 survey found only 44 percent of employees had received any AI training, and just 16 percent received it regularly, even as the large majority were already using AI at work. The problem is when "everyone on the journey" becomes the strategy instead of the consequence of one. It works when leadership is visibly leading and has empowered someone to own the result. It fails when it's used as a substitute for that.

How not to relearn this the hard way

The lesson from the last two waves is sitting right there, and it costs nothing to apply. A few things make the difference between owning AI and being owned by it.

Name an owner with a mandate, not a title. The question isn't who's enthusiastic or who's technical. It's who has the authority to change how work happens across functions, and the explicit backing of the top to do it. A Centre of Excellence with no decision rights is just a cost centre with good intentions.

Start from the operating model, not the tool. Every failed wave began with a purchase and worked backwards looking for value. The ones that worked started with a decision about how the business should run differently, then found the tools to get there.

Decide what stays human before you automate. The rehiring crisis happened because companies cut first and discovered the judgment, escalation and relationship work afterwards. Map that work before you touch it, not after.

Measure outcomes, not adoption. "AI adoption rate" tells you nothing. Time saved, quality improved, decisions made better: those tell you whether anything actually changed.

The businesses that got value out of digital and analytics weren't the ones with the biggest budgets or the earliest start. They were the ones whose leaders led the change instead of delegating it to a programme. AI is the same test. A title and a training course don't create authority. Only the top of the house does.