Everyone has AI. Now, how do you actually make it work?

The real AI challenge is organisational, not technological.

Before artificial intelligence arrived, enterprise technology leaders spent much of the past decade modernising infrastructure, migrating workloads to the cloud, improving data governance, and digitising processes. Now, AI has exposed an uncomfortable truth: many of those foundations were never as modern as organisations believed. 

The prevailing narrative suggests that enterprise AI has reached a curious impasse. Almost every large organisation has acquired the tools. Most have launched pilots. Many can point to isolated examples of improved productivity. Yet few executives would claim they have fundamentally changed how their organisations operate; the gap between experimentation and enterprise impact remains stubbornly wide.

At a recent C-Suite Exchange, in partnership with Nearform, CIOs, CTOs, and senior engineering leaders quickly departed from discussions of language models, copilots, and agents. Instead, they settled on a position far more revealing: the obstacle is no longer access to AI, but the enterprise itself.

AI provides organisational ‘outcome clarity’

For the better part of three years, the assumption has been that organisations simply needed better models, more capable tools, or greater confidence in the technology. Yet the executives on our call described a different reality. Their organisations are not struggling because AI is immature. They are struggling because AI is exposing weaknesses that have existed for years—with or without their knowledge.

Perhaps the most illuminating phrase of the discussion came from one participant, who argued that organisations first need "outcome clarity". Before selecting a model, building an agent, or commissioning a proof of concept, leaders should be able to articulate, with precision, the business outcome they are trying to achieve.

It sounds almost disappointingly simple. Yet its absence explains why so many AI initiatives drift into demonstrations of technical capability rather than meaningful business transformation.

Too often, organisations begin with the technology and then search for somewhere to apply it. The more disciplined approach reverses that logic entirely. Begin with the commercial objective, reduce customer churn, improve engineering productivity, shorten product development cycles, then increase asset utilisation. Only then should leaders ask whether AI is the appropriate mechanism for achieving that goal.

One executive observed that many organisations continue to "AI" existing processes rather than questioning whether those processes deserve to exist in their current form at all. Another offered an equally memorable variation: a poor process, accelerated by AI, remains a poor process.

AI layer cake 

History suggests this pattern is familiar. When digital transformation emerged, most incumbent organisations digitised existing operating models. Think: forms became online forms, and paper workflows became electronic workflows. Existing businesses became marginally faster versions of themselves.

Meanwhile, a generation of digital-native companies approached the same challenge differently. They were not applying digital to existing organisations; they were designing organisations around digital from the outset. Greater efficiencies were gained, yes, but entirely different business models were created, too.

The same distinction may now be emerging with AI.

Many established enterprises are understandably layering AI onto decades of accumulated architecture, governance, and process. They are attempting to retrofit a proxy for intelligence into organisations designed for an altogether different technological era. By contrast, AI-native businesses begin with very different assumptions. Their operating models, development practices, and decision-making processes are conceived around AI from the beginning.

It would be naïve to suggest that a two-hundred-year-old financial institution can simply behave like an AI start-up. Nor should it. Regulation, customer obligations, and operational resilience impose constraints that younger companies do not face.

Nevertheless, the comparison raises an uncomfortable question. Are today's incumbents repeating precisely the mistake they made during the first wave of digital transformation? Are they merely applying AI to existing businesses while tomorrow's competitors are quietly redesigning the business itself?

Businesses need to crack governance

If there was one recurring theme that united almost every contribution, it was governance.

Interestingly, governance itself was rarely portrayed as the villain. Most participants recognise its necessity, particularly in highly regulated industries handling sensitive customer data. The frustration stemmed from something else: governance functions are being asked to evaluate technologies that are evolving faster than governance itself can mature.

The consequence is predictable. Governance becomes perceived as the department that slows everything down, when in reality it is often struggling simply to keep pace.

Several attendees described moving towards governance models that are embedded much earlier in the delivery process. Rather than approving projects at the point of deployment, compliance, legal, risk, and data specialists are becoming part of the incubation process itself. The ambition is to reduce unconstructive friction. Governance shifts, so the theory goes, from gatekeeper to enabler.
That distinction may prove one of the defining characteristics of successful AI adoption over the next few years.

Finally, a surprising discussion of the afternoon concerned neither governance nor technology. It concerned economics—or token-economics.

Cloud computing eventually gave birth to FinOps, as organisations realised that seemingly limitless infrastructure could also generate seemingly limitless bills. AI appears destined to follow the same trajectory.

One executive described power users whose monthly AI costs had risen from tens of pounds to well over a thousand through token consumption alone. Another questioned whether finance functions are equipped for a world in which software costs are no longer largely deterministic, but fluctuate according to prompts, model selection and inference demand.

Most organisations currently discuss AI as a capability challenge. It may soon become a financial management challenge.

That introduces uncomfortable strategic questions. What happens if model providers substantially alter pricing? How portable are AI workloads between competing platforms? At what point does today's helpful assistant become tomorrow's vendor lock-in?

These are not hypothetical concerns. They are emerging operational realities.

Closing thoughts 

Technology leaders often describe resistance to AI as a skills problem. Several participants suggested something more fundamental. AI requires experienced professionals to rethink processes they have spent decades refining. In some cases, it asks them to automate aspects of work that have defined their own expertise. Unsurprisingly, transformation becomes as much psychological as technological.

Perhaps that explains why imagination as a re-emerging skillset surfaced repeatedly throughout the conversation.

The challenge is no longer convincing organisations that AI matters. It is helping them imagine organisations designed around fundamentally different assumptions. That requires leaders to think beyond incremental productivity gains towards operating models that simply were not feasible before.

The irony is that the engineering challenges themselves are becoming more manageable. Writing code is becoming easier; building prototypes is becoming faster; individual productivity is improving at remarkable speed. The harder work has shifted elsewhere.
In fact, if everyone has access to AI it will cease to be a competitive advantage. 

The organisations that pull ahead over the coming decade are unlikely to be those with exclusive access to better models. They will be those that redesign themselves around the opportunities those models create. That is why, for these leaders, the execution gap is not really an AI problem at all, but an organisational one.

In partnership with