Ask any senior executive whether their organisation is using artificial intelligence, and the answer in 2026 is almost certainly yes. Ask whether that AI is generating sustainable, scalable value — and the picture changes dramatically. Research published by MIT in 2025 found that the gap between ambition and delivery remains stubbornly wide: organisations are running AI experiments everywhere, but transformational systems in production are still the exception, not the rule.
This is the central paradox of the AI era. The technology has never been more accessible. With the arrival of generative AI, any organisation — regardless of size, sector or existing data assets — can begin experimenting in days. Yet the path from first pilot to scaled value creation remains one of the most underestimated journeys in business. Understanding why requires looking beyond the technology itself.
The Pilot Trap
Most AI initiatives stall at the pilot stage for a predictable reason: organisations invest heavily in data science and technology, but underinvest in the organisational decisions that determine whether a working prototype ever reaches production. Choosing the right use cases, structuring the right team, measuring economic impact and securing the right funding are not secondary concerns — they are the primary drivers of whether AI creates value or remains a demonstration exercise.
The most dangerous misconception is that AI transformation is fundamentally a technology problem. It is not. It is an organisational problem that requires technology. The organisations that consistently move from pilots to production treat their AI journey as a series of deliberate decisions — not a single technology deployment.
Generative AI Has Lowered the Barrier to Entry — Not to Scale
The rise of generative AI has genuinely changed the starting conditions for any organisation. For the first time, it is possible to begin extracting value from AI without owning proprietary data. Large language models arrive pre-trained on vast amounts of information and are ready to use from day one. A small manufacturing firm can deploy a customer service chatbot or an internal knowledge assistant this week, without a data science team.
But this democratisation comes with an important caveat. Using a generic LLM is not the same as building a competitive AI capability. The real value — the kind that translates into measurable business outcomes and is hard for competitors to replicate — comes from adapting those pre-trained models using your own organisational data and documents. Fine-tuning and retrieval-augmented generation, which grounds model outputs in your own knowledge base, dramatically increase accuracy and reduce hallucinations. The barrier to entry has fallen. The barrier to sustainable advantage has not.
Who Owns AI in Your Organisation?
One of the most consequential decisions any organisation takes — often without realising it is a decision — is where to place the Chief Data or Chief AI Officer on the organisational chart. Research shows that executives in these roles have an average tenure of just two to three years. That short tenure is not coincidental. It reflects the difficulty of a role that requires both deep technical credibility and significant organisational authority — and that frequently lacks one or the other.
Organisations that place the data and AI function too far from the CEO create a structural obstacle that no amount of talent can overcome. The function needs proximity to strategic decision-making, a clear mandate, and the ability to set priorities that cut across business units. Without that positioning, AI efforts become fragmented, and the centre of excellence — however technically capable — risks becoming a service desk rather than a strategic driver.
Data, AI, and IT: Three Functions That Must Work as One
A second structural decision that determines whether AI scales is the relationship between the data function, the AI team, and IT. These three areas are deeply interdependent, yet in many organisations they operate with misaligned objectives, separate budgets and competing priorities.
For organisations at an early stage of their AI journey, keeping data and AI close to IT is not a sign of immaturity — it is pragmatic. Data platforms are IT systems. Without tight coordination, AI teams wait months for infrastructure that should take days. As organisations mature, the relationship can evolve toward greater autonomy for the AI function, but only once data management and governance processes are robust enough to operate independently. Getting this sequencing wrong is one of the most common causes of AI projects stalling well short of production.
You Cannot Manage What You Do Not Measure
Scaling AI requires knowing where you stand. Yet surprisingly few organisations measure their data and AI maturity in any systematic way. They assume that progress on individual projects equates to organisational readiness — and discover too late that it does not.
Measuring data maturity is a multi-dimensional exercise. It spans the quality and governance of data assets, the sophistication of the technology platform, the organisation of the data and AI function, the skills available, the application of privacy and security standards, and — critically — the business value already being generated. An organisation that scores well on technology but poorly on governance will find that its AI use cases hit a data quality ceiling long before they reach scale. An organisation that scores well on individual use cases but has no structured way to prioritise or replicate them will remain a collection of pilots indefinitely.
The practical value of a rigorous maturity assessment is not that it produces a score. It is that it reveals which specific gaps are blocking progress — and therefore where the next investments should go.
The Journey Is the Strategy
The organisations that consistently create value from AI share a common characteristic: they treat the journey as a strategic discipline. They make explicit decisions about organisational structure, data governance, use case selection, economic measurement and responsible use. They do not leave those decisions to chance or allow them to be made by default.
None of this requires a large budget or a hundred data scientists. The first edition of this framework was written when AI was still a largely corporate concern. In 2026, with generative AI placing powerful tools in the hands of any business, the decisions are the same. What has changed is the cost of getting them wrong. The technology is no longer the bottleneck. Organisational clarity is. The organisations that understand this will move beyond pilots. The rest will keep running them.
