The Bandwidth Problem

Over the last few weeks, I have spoken with dozens of business leaders about their growth bottlenecks. The picture is remarkably consistent: leaders and their teams are stretched thin, pulled between shipping products, running R&D projects, keeping customers happy, and driving more sales, all at the same time.

The paroxysm of this is when MDs, CEOs, and senior leadership teams are so consumed with delivering their strategic plan and executing every tactical step they carefully built to support it that there is simply no bandwidth left for anything else.

AI adoption sits in that “anything else” category. Not because they don’t see its value, quite the opposite. Every interaction they have with AI leads to the same conclusion: this is a powerful technology that deserves proper attention to be deployed effectively. But “deserves attention” and “gets attention” are two very different things when your diary is already full.

The Default Starting Point

So they start where most companies start. They buy a batch of Copilot licences and hope that productivity goes up and that this somehow flows to the bottom line.

In some exceptional cases, it does. But the majority confess that it simply adds cost. The general consensus is that the quality of deliverables — SOPs, manuals, marketing content, pitch decks, brochures, and application notes is noticeably better and more consistent. But better-quality documents do not necessarily translate into financial gain.

But why?

Because individual productivity improvement does not automatically mean the company banks on that improvement. In fact, quite often, the opposite happens. People become more efficient, which gives them more time, and that time gets absorbed elsewhere: achieving a better work-life balance, tackling long-standing issues they never had the capacity for, or simply building more social connections with colleagues. All perfectly reasonable things to do with newfound time. But none of them shows up on the P&L.

People finish in six hours what used to take eight. But the working day is still eight hours, and the extra two hours simply fill up with other things.

The Data Confirms It

This is not just anecdotal. The research now overwhelmingly supports what business leaders are experiencing on the ground.

A recent study published by the National Bureau of Economic Research surveyed nearly 6,000 CEOs, CFOs, and senior executives across the US, UK, Germany, and Australia. Around 90% reported that AI has had no impact on productivity or employment at their business, despite roughly 70% actively using AI. The CEOs themselves averaged just 1.5 hours of personal AI use per week.

McKinsey’s State of AI 2025 report paints a similar picture. 88% of companies now use AI in at least one business function, up from 78% a year earlier. But only 39% report any impact on EBIT, and for most of those, the impact is less than 5%. The majority of organisations remain stuck in experimentation or pilot stages, with approximately one-third having begun to scale their AI programmes.

BCG’s research is starker still. Their 2025 Build for the Future study of 1,250 organisations found that only 5% are creating substantial value from AI at scale. 60% generate no material value despite active investment.

Economists have given this phenomenon a name. They are calling it a modern version of Solow’s productivity paradox, a callback to Nobel laureate Robert Solow’s observation in the 1980s that computers were obviously transformative but didn’t show up in the productivity statistics. As he put it at the time, you could see the computer age everywhere except in the productivity numbers. Today, you can see AI everywhere except in the financial results.

The J-Curve: It Gets Worse Before It Gets Better

And it gets worse before it gets better. Research from MIT Sloan examining US manufacturing firms found that AI adoption frequently leads to a measurable but temporary decline in performance, followed by stronger growth in output, revenue, and employment. Researchers call this the productivity J-curve.

The numbers are stark. Organisations that adopted AI for business functions saw an initial productivity drop, and when researchers corrected for selection bias (the fact that companies expecting higher returns are more likely to be early adopters), the short-run negative impact was substantial. This is not just about growing pains. The decline points to a fundamental misalignment between new digital tools and existing operational processes.

However, over time, the same research found that organisations adopting AI tended to outperform similar firms that did not use AI in both productivity and market share. The firms seeing the strongest gains were those that were already digitally mature before adopting AI and those that invested in complementary changes, automation technologies, workflow redesign, and strategic reallocation of resources.

The Expendable Booster

So is the Copilot phase pointless? Not exactly. But it is important to be honest about what it is and what it is not.

What I consistently observe is that most companies need to go through this phase. It is how they build familiarity with AI, learn how to prompt, try, fail, and learn again. It is the foundation, not the destination.

Think of it a bit like the first 50 years of space exploration, where we accepted as perfectly normal the fact that we would build enormously expensive Stage 1 rocket boosters, use them once, and let them burn up or crash into the ocean. Nobody questioned it. It was just how things were done until SpaceX proved in 2015 that you could land them and reuse them. The waste was real, but it was also, for most, an unavoidable part of the learning curve that eventually led to a fundamentally better way of doing things.

The Copilot phase is the expendable booster. It gets you off the ground. It teaches you how the technology works. But the real value only comes when you stop accepting the waste and start redesigning the mission, which is when AI agents and workflow transformation discussions begin.

What Separates the Companies That Win

Here is what the research says separates companies that get there from those that stay stuck.

Focus beats scatter. BCG found that struggling companies pursue an average of 6.1 AI use cases simultaneously, spreading their efforts too thin. Leading organisations focus on just 3.5, yet anticipate generating 2.1 times greater ROI from their more concentrated approach. Less than one-third of companies have even upskilled a quarter of their workforce to use AI effectively, and most don’t track financial KPIs for their AI initiatives at all, making it impossible to know whether their investments are paying off.

Transformation beats incrementalism. Organisations achieving significant enterprise-level impact (representing just 6% of McKinsey’s survey respondents) are 3.6 times more likely to pursue transformative change rather than incremental improvements. They are 2.8 times more likely to fundamentally redesign workflows, which the analysis identifies as one of the strongest predictors of success. These companies are not the ones that patiently matured through Copilot to agents. They are the ones who redesigned how work gets done from the outset.

Leadership beats technology. McKinsey found that AI high performers are three times more likely to report strong senior leader ownership of and commitment to their AI initiatives, including actively role-modelling the use of AI themselves. The technology journey matters, but the leadership journey matters more.

Intent beats patience. There is also a risk that the Copilot phase, without clear intention, creates bad habits rather than a useful foundation. MIT research found that older, established firms actually saw declines in structured management practices after adopting AI, and that alone accounted for nearly one-third of their productivity losses. Microsoft’s own 2025 Future of Work Report identifies the rise of what it calls “workslop”: AI-generated content that appears useful but lacks substance or accuracy, forcing others to spend time correcting it. Better tools do not automatically produce better outcomes if the intent and discipline are not there.

The Uncomfortable Conclusion

The uncomfortable answer, based on what the research now confirms and what dozens of conversations with business leaders reveal, is that the companies that will win are not the most patient ones. They are the most intentional ones. The ones who treat AI adoption not as a technology rollout but as an organisational transformation. The ones who stop hoping that Copilot licences will somehow compound into strategic advantage and start making deliberate choices about where AI should fundamentally change how they operate.

The booster got you off the ground. Now it is time to land it and reuse it.

Sources:

• National Bureau of Economic Research (NBER), Business Outlook Survey, 2026
• McKinsey & Company, “The State of AI in 2025: Agents, Innovation, and Transformation,” November 2025
• Boston Consulting Group, “The Widening AI Value Gap: Build for the Future 2025,” September 2025
• MIT Sloan / MIT Initiative on the Digital Economy, “The Rise of Industrial AI in America: Microfoundations of the Productivity J-Curve(s)”
• Microsoft Research, “New Future of Work Report 2025”
• Federal Reserve Bank of San Francisco, “The AI Moment? Possibilities, Productivity, and Policy,” February 2026