Last month at The Economist's AI conference in London, I watched a room full of CFOs independently arrive at the same conclusion, one after another, as if reading from a script nobody had distributed. The topic was AI. The setting was a technology conference. And yet every single panel, from DHL and Thermo Fisher to Swissport, Meta, AWS, and the Danish National Bank, ended up in the same place: the problem isn't the technology. It's the people. You could practically see the event organizers wondering whether they should have booked a psychologist instead of a keynote.

Mark from DHL (not his real name, to protect the guilty) put it most bluntly. His team had built a chatbot to handle internal support tickets. The technology worked beautifully. Usage was essentially zero: people just kept writing emails, presumably out of a deep emotional attachment to their inbox. His conclusion: "AI is not a tech challenge. It's a change challenge. Eighty percent is change." Over at Meta, the finance director framed it differently but landed in the same spot: "We're not transforming finance with AI. We're transforming our people." When I hear Meta talk about transforming people, the phrase "what could possibly go wrong" does spring to mind, but in fairness, the sentiment was right even if the messenger invites a certain existential dread. And Rafael, speaking on a panel about whether AI is delivering enough value, ticked through a list of what enterprises actually need (executive sponsorship, tolerance for failure, cross-functional coordination, usability at scale) and then observed, almost as an aside, that everything he'd just described was organizational, not technological. He seemed mildly surprised by his own conclusion, which tells you something about how deeply the technology narrative has taken hold.

None of this should surprise anyone. And yet it apparently does, every single quarter, as a fresh wave of AI investments encounters the same immovable object: an organization full of human beings who did not get the memo. The enterprise AI conversation remains stubbornly fixated on models, tools, and platforms, as if choosing the right large language model will somehow override the fact that Sally in accounts receivable hasn't opened the training email. The data tells us otherwise, and it tells us with remarkable (one might say tedious) consistency.

The Numbers Don't Lie, But Leadership Might

BCG's research across 1,800 C-suite executives reveals that roughly 70% of challenges in AI projects stem from people and process issues, not technical ones. If you find yourself thinking "we've known this for decades about every technology adoption cycle," congratulations: you have been paying attention, which puts you ahead of most boardrooms.

McKinsey's latest State of AI report sharpens the point: the biggest barrier to scaling AI isn't employee pushback. It's leadership inertia. The executives who say their companies have created real value from AI are three times more likely than their peers to report that senior leaders demonstrate ownership and actively model the use of AI. Where leadership is absent, adoption stalls regardless of what tools have been deployed. The fish, it turns out, rots from the head.

But here's where it gets properly interesting. BCG and Columbia Business School surveyed both executives and frontline workers, and the perception gap is the kind of thing you'd find funny if it weren't so expensive. Eighty percent of executive leaders believe their employees are well-informed about the company's AI strategy. Among individual contributors, that number is 29%. Leaders think 76% of their workforce is enthusiastic about AI. Workers put that figure at 31%. This isn't a communication gap. It's an alternate reality. And it matters because employee centricity, meaning how well an organization actually listens to and equips its people, explained 36% of the variance in AI maturity. More than industry. More than department. More than company size.

So the single most powerful predictor of whether your AI investment will pay off is whether your people feel heard and prepared. And the people making the investment are wildly, confidently wrong about whether that's happening. There is a term for spending large amounts of money based on a fundamentally incorrect understanding of your own organization, but in polite company we just call it "strategy."

The Structural Trap

Leadership inertia creates the conditions for failure. But the failure itself plays out structurally, in three overlapping ways that will be immediately recognizable to anyone who has worked in a large corporation, which is to say, they are entirely predictable and apparently unavoidable.

The first is silo fragmentation. A 2025 survey found that 42% of C-suite executives believe AI adoption is actively creating organizational rifts, with 71% reporting AI applications being built in silos and 68% noting AI-induced tension between IT and the rest of the business. Each function builds its own tool, optimizes for its own metrics, and wonders why nothing connects. If you wanted to design an operating model specifically to prevent AI from delivering value at scale, you could hardly do better than the average enterprise org chart.

The second is what Deloitte calls "culture debt," a term I think deserves wider circulation. Their 2026 Global Human Capital Trends report found that 65% of organizations believe their culture needs to change significantly because of AI, and 34% say culture is currently blocking their AI goals. Meanwhile, 60% of executives are already using AI in decision-making, but only 5% say they're managing it well. Read that again: sixty percent doing it, five percent doing it well. Organizations are scaling AI faster than they're building the accountability structures, governance norms, and trust frameworks to support it. They're writing checks their culture can't cash, and the overdraft fees are measured in failed pilots and executive turnover.

The third is workflow rigidity. Only 37% of survey respondents reported meaningful investment in change management, training, or incentives to help people integrate AI into how they actually work. You can build the most sophisticated AI tool ever conceived, but if nobody redesigns the workflow it's supposed to improve, you get what BCG's research shows: an average of 4.3 pilots per organization, with only 21% reaching production scale. 4.3 pilots is the corporate equivalent of owning four gym memberships and still not being fit.

The Last Mile: Where Initiatives Go to Die

Even when leadership is engaged and structures are supportive, there's a final set of obstacles at the frontline that consistently kill momentum. This is where the irony gets thick enough to cut.

Job displacement anxiety is real but more nuanced than headlines suggest. Deloitte found that 70% of workers are open to offloading work to AI to free up time, but 28% worry about their jobs. It probably doesn't help that the CEOs of both Microsoft and Ford have publicly suggested that large percentages of their workforces will be eliminated through AI (presumably motivational speeches intended to accelerate adoption). Workers who express concern about this are sometimes dismissed as modern-day Ned Ludds, smashing the looms out of ignorance. But Ludd's followers, it's worth remembering, weren't wrong that the machines would eliminate their jobs. They were wrong that smashing the machines would help. The modern equivalent isn't sabotage: it's the quieter act of simply not logging in. The academic literature adds an important wrinkle: workers resist top-down mandated AI products that prioritize efficiency over quality and creativity, because those mandates implicitly say "you are a cost to be optimized" rather than "you are a professional to be augmented." Generational segmentation matters here too. Many employees, particularly Gen X, won't adopt tools that force them to conform to a standardized way of working. They've survived three previous technology revolutions by adapting on their own terms, and they're not about to stop now.

Then there's the skills gap, which is arguably the most damning indictment of the whole enterprise. In BCG's global survey, 62% of C-suite executives cited talent and AI skills as their biggest challenge to achieving AI value. Yet only 6% said they've begun upskilling their workforce in a meaningful way. Sixty-two percent identifying the problem. Six percent doing anything about it. If a doctor told you that 62% of their patients had the same treatable condition and they were treating 6% of them, you might question the quality of the healthcare. But in enterprise AI, we call this "the adoption journey."

Constantine, a startup CFO speaking at The Economist conference, shared survey data from 1,500 finance and procurement professionals that captured the whole farce perfectly: close to 100% see the benefits of AI, but fewer than half have actually automated any processes with it. Universal enthusiasm, sub-50% action. If AI adoption were a diet, we'd all be standing in front of the fridge at midnight explaining that we fully intend to start on Monday.

Why This Hits Customer AI Particularly Hard

Everything I've described so far applies across the enterprise. But if you're in the business of Customer AI (predictive analytics, customer intelligence, AI-driven experience management) you face a compounded version of these obstacles, and the reasons are specific to how the CX discipline was built.

Customer experience, as a corporate function, was constructed as a measurement discipline. For two decades, the operating model has been: survey customers, compute a score, report the score, try to improve the score. Entire careers, organizational structures, and vendor ecosystems were built around this feedback loop. NPS, CSAT, voice-of-customer programs: these aren't just tools, they're identity. They define what CX teams do and how they see their own value. Asking them to abandon this for Customer AI is, to put it mildly, asking turkeys to be enthusiastic about Christmas.

Customer AI asks those same teams to become something fundamentally different: a predictive discipline. Instead of asking customers how they feel after the fact, you're predicting what they'll do before they do it. Instead of relying on survey responses (which capture maybe 10-30% of customers at best), you're modeling behavior across the entire customer base using operational, financial, and behavioral data. The silent majority, the 70-90% who never respond to a survey, turns out to be the signal, not the absence of one.

This is not a tool upgrade. It's an identity shift. And identity shifts trigger every cultural barrier in the book, often simultaneously.

The leadership inertia problem is acute because many CX leaders built their reputations on the measurement paradigm. Embracing Customer AI means acknowledging that the approach they championed, the one that gave them their title, their budget, their seat at the table, was fundamentally limited. That's a hard pill, and most people prefer to keep refilling the old prescription.

The trust deficit compounds because predictive analytics creates a kind of transparency that survey-based CX never did. When you can see which customers are at risk months in advance, when the model surfaces operational failures that directly predict churn, you lose the comfortable ambiguity of lag indicators. The quarterly NPS report let everyone nod along and pledge to do better next quarter. A predictive model that says "these 200 accounts are at risk because of these specific operational failures" demands accountability right now, from specific people. It's the difference between a weather forecast and a flood warning: one invites discussion, the other demands action. Many organizations have built rather comfortable cultures around the discussion part.

The silo problem is compounded because Customer AI, by its nature, requires data from across the enterprise: sales, product, support, finance, operations. It cannot live in a single function. Yet most CX teams sit in a silo, often without the political capital or technical infrastructure to access the cross-functional data that makes prediction possible.

And the skills gap is perhaps worst of all. CX professionals were trained to design surveys, interpret sentiment data, and build closed-loop response programs. Customer AI requires an entirely different toolkit, one rooted in data science, operational analytics, and financial modeling. The 6% upskilling figure from BCG's research? In CX, I'd wager it's even lower. And the reason is circular: the leaders who would need to commission the upskilling are the same ones whose expertise it would render obsolete.

What Actually Works

The research points to a number that deserves attention: organizations that invest heavily in culture change alongside technology see 5.3 times higher success rates than technology-only approaches. Five point three times. You would think a multiplier like that would get people's attention. And it does, right up until the moment they have to choose between buying another platform and doing the difficult, unglamorous, deeply human work of changing how their organization actually operates. The platform wins almost every time, because platforms have vendor demos and culture change has difficult conversations.

For CX specifically, the path to Customer AI runs through culture first and technology second. You need leadership that's willing to let go of the measurement-era playbook. You need trust frameworks that make predictive transparency an ally rather than a threat. You need cross-functional data access and governance. And you need to retrain your people, not in the abstract ("here's what AI is"), but in the concrete ("here's how your job changes, and here's why that's an opportunity").

If you've read any of the management literature on cultural change, you'll know that each organization's culture responds to different levers. A control culture (hierarchical, process-driven, plan-oriented) needs a different AI implementation approach than a competence culture, which is competitive and individually rewarded. Imposing the wrong change methodology on the wrong culture is one of the most common and most expensive mistakes in enterprise transformation, and AI is no exception.

The technology works. The economics are compelling. We know, and have known for quite some time, that it's the culture, stupid. The organizations that actually act on this will win. The rest will keep running pilots, keep attending conferences where CFOs confess that it's really about the people, and keep going home to buy more technology.

That's what I think. What do you think?

I'm Richard Owen, co-founder and CEO of OCX Cognition. We build predictive customer analytics for companies who'd prefer to know which customers are at risk before those customers have already decided to leave.