We have been here before.
In 2015, Big Data was the thing nobody could afford to ignore. Organisations everywhere started building Data Lakes and Data Warehouses. The language became standard. Consultants were hired. Strategies were written. And then, quietly, most of those Data Lakes became data swamps — technically present, practically useless — because the organisations that built them hadn't yet developed the culture, the architecture, or the leadership proximity to make good decisions with what they'd created.
The organisations that got genuine, lasting value from Big Data were not the ones who moved fastest. They were the ones who already had strong digital foundations, a tolerance for experimentation, and leaders close enough to their technical reality to ask the right questions. Sound familiar?
In April, PwC published its 2026 AI Performance Study — a survey of over 1,200 senior executives across 25 sectors — and the headline figure is striking: 74% of AI's economic value is being captured by just 20% of organisations. The remaining 80% are investing in AI and seeing little to nothing back.
I'm not surprised. I've watched the same pattern play out before.
The evangelical phase is over. What comes next matters more.
Between 2024 and 2025, I watched many business leaders become almost evangelical about AI; sometimes to what felt like an irrational degree. AI became the frame for everything. Every conversation, every strategy document, every board agenda found a way to include it.
That phase is passing. And I think that's healthy.
The conversations I'm having now are more honest. Leaders are beginning to ask more practical questions. They're acknowledging when applying AI is counterproductive. They're less interested in being seen to use AI and more interested in whether it's actually working. The hype around AI 'slop', lazy, low-quality AI output that makes it into the world unchecked, has sharpened people's instincts in useful ways.
But here's the problem I'm still seeing: even in this more pragmatic moment, most leaders are reaching for AI as a necessity rather than as an answer to a specific problem. They know they can't ignore it. The press is relentless. Their peers are talking about it. So they adopt it, but at the level of least resistance. Customer-facing chatbots. Knowledge base tools. Individual productivity applications. Useful, sometimes, but nowhere near the transformational value the headlines promise.
PwC's data captures exactly this. The 80% stuck in pilot mode aren't failing because they chose the wrong tools. They're failing because they're pointing AI at the wrong questions.
The real dividing line isn't the technology
PwC's research concludes that what separates AI leaders from laggards is primarily strategic, not technological. The leading organisations aren't deploying more AI, they're deploying it differently. They're using it to pursue growth, to reinvent their business models, to do things they couldn't do before. The rest are using it to do the same things slightly faster.
I'd push that further. In my experience, the real dividing line isn't even strategic intent — it's organisational readiness.
Here's what I mean. When a leader tells me their technical team "isn't really engaging with AI," my first question is whether that's actually a capability problem or a culture problem. Because in most cases, the technical people do understand the systems. They understand the data, the architecture, the constraints. They have views on where AI could genuinely help.
What they often lack is the environment to act on that understanding. There's no safe space to experiment. There's no fail-fast culture that lets teams try something, learn from it, and move on without it becoming a political event. The transparency and collaborative trust between technical teams and business leadership that genuine AI adoption requires, that's where the gap actually lives.
So leaders, frustrated that nothing seems to be happening internally, turn outward. They bring in external parties. They buy platforms. They follow the press. And the cycle continues, because the underlying conditions haven't changed.
Why scale is the hard problem
There's an important distinction I keep coming back to: AI as a personal tool versus AI at organisational scale.
Embedding AI into individual workflows can be quick, genuinely rewarding, and relatively straightforward. The barriers are low, the feedback is fast, and the wins are real. Many people I know, across every kind of organisation, have found ways to use AI that have meaningfully changed how they work day to day. This is real and shouldn't be dismissed.
But deploying AI at scale is a fundamentally different problem. It requires confronting the same age-old challenges that Big Data did: data architecture, data quality, data portability. The technical knowledge is still required. The governance questions don't disappear because the interface looks friendlier. And the organisations best equipped to handle those challenges are the ones that already invested in getting their foundations right — which, again, is exactly what PwC's data shows.
The most exciting AI stories — the ones circulating in the press, the ones leaders read and feel the pressure of — come from two places: early-adopting technology companies for whom experimentation is simply how they operate, and large enterprises that had already built mature digital foundations before AI arrived. They had already embedded the culture, the architecture, and the leadership behaviours that make genuine adoption possible. AI didn't transform them. It amplified what was already there.
What this means if you're leading an SME or a charity right now
The uncomfortable truth is that if your organisation hasn't yet developed strong digital foundations — clear data ownership, transparent technical leadership, a culture that can tolerate intelligent failure — then most AI investment is going to disappoint you. Not because the technology doesn't work, but because the conditions for it to work don't yet exist.
That's not a reason to do nothing. It's a reason to be honest about what to do first.
Before asking "which AI tools should we use?", ask: do our teams have the trust and transparency with leadership to experiment openly? Do we understand our own data well enough to know what AI could actually do with it? Are our leaders close enough to the technical reality to make good decisions, or are they operating on borrowed language?
The organisations that will be in the 20% in three years are probably not the ones doing the most with AI right now. They're the ones building the conditions that make sustained AI value possible. The same was true in 2015 with Big Data.
We've been here before. The lesson was available then. It's available now.
Tim Cawood is the founder of Cawood.io, working with SME and charity leaders on digital strategy, architecture, and organisational change.
Read the full PwC 2026 AI Performance Study.