AI is not a software upgrade

This is the first in a two-part series on how organisations can successfully embrace AI. In this piece, we set out why so many AI programmes fail to deliver meaningful value, and why the problem is not the technology itself, but how organisations are set up to use it. 

Artificial intelligence is often discussed as if it were simply the next wave of enterprise technology: another platform to deploy, another set of tools to integrate, another capability to be absorbed into the existing stack. That framing is inadequate. AI is not only changing the technology landscape. It is changing how work is designed, how decisions are made, how expertise is applied and how value is created across the organisation. 

That is why so many AI programmes disappoint. In most large organisations, the issue is not whether the technology is powerful enough – it is. The issue is that AI does not sit neatly inside one function. It cuts across operations, customer engagement, product, risk, compliance, data, technology and leadership itself. The challenge, then, is not simply how to introduce AI tools. It is how to build the operating model, authority and governance required to use AI at scale in a way that is commercially valuable, operationally sustainable and trusted. 

This is now the central dividing line. The question is no longer whether organisations are engaging with AI. By 2025, McKinsey found that 88% of organisations reported regular AI use in at least one business function, and the pace of adoption has only accelerated since then. The more consequential distinction now is not between organisations using AI and those that are not, but between those generating AI activity and those redesigning themselves to capture AI value at scale. 

Technology alone has never been the point 

History suggests this should not surprise us. General-purpose technologies rarely create lasting advantage simply because they are adopted. They create advantage when organisations redesign around what the technology makes possible. 

When manufacturing electrified, some firms simply replaced steam engines with electric motors and saw modest gains. Others redesigned the factory around electricity and transformed productivity. When retail digitised, some incumbents treated the internet as another channel; others rebuilt the business around digital infrastructure and changed the economics of the industry. AI is producing the same split now. One group of organisations is layering copilots and isolated use cases onto existing workflows. Another is rethinking workflows, roles and service models from first principles. The former will gain efficiency. The latter will change their competitive position. 

That distinction matters because AI is not just another tool to bolt onto yesterday’s operating model. It is a capability that forces organisations to ask what should be redesigned, what should be automated, what should remain human-led and how work should flow when intelligence is no longer scarce. The organisations that answer those questions well will move beyond incremental productivity and begin to reshape cost structures, service models and competitive advantage. 

Where most organisations are getting stuck

The current pattern across enterprise AI is not one of inaction. It is one of fragmentation. Most organisations have pilots. Many have promising use cases. Some have whole portfolios of experiments spread across functions. But relatively few have turned that activity into repeatable enterprise value. The result is what might be called PoC purgatory: a state in which the organisation can point to dozens, sometimes hundreds, of proofs of concept and pilot initiatives, yet very little ever makes it into production in a form that changes cost, productivity, control, or customer outcomes. 

That is what makes PoC purgatory so deceptive. It creates the appearance of momentum. There are demos, working sessions, vendor conversations, internal showcases and a steady flow of experimentation. From the outside, it can look like the organisation is moving quickly. In reality, it is often spinning its wheels. Initiatives remain trapped in pilot mode because the harder decisions are never made: which use cases truly matter, which should be scaled, who owns them, how they will be governed, how workflows must change around them, and what trade-offs the organisation is willing to make to bring them into production. 

This is why PoC purgatory is not simply a technology problem. It is a failure of operating model and organisational resolve. The organisation keeps proving that AI can do interesting things, but never builds the conditions required for AI to do consequential things. Proofs of concept multiply because experimentation is easier than prioritisation, easier than redesigning workflows, easier than assigning accountability and easier than forcing adoption. The result is AI theatre rather than AI transformation: visible activity, limited production and very little measurable value. 

That is why AI should not be treated as a software problem, or even as a technology problem with some change-management implications attached. It is an operating-model problem. It touches strategy, decision rights, role design, governance, performance management and leadership behaviour. The real gap is not between technical capability and technical implementation. It is between technical capability and organisational readiness.