Why AI stalls in organisations

The models are ready. Most organisations are still built for a world without them.

Organisations have never spent more on AI. Gartner expects worldwide AI spending to reach about USD 2.7 trillion in 2026, almost half as much again as in 2025.¹ Nearly nine in ten organisations now use AI somewhere. Yet in McKinsey’s survey of the state of AI, only 37% report any effect on EBIT at all.² Gartner also expects more than 40% of agentic AI projects to be cancelled by the end of 2027, for reasons of cost, unclear value and weak risk controls.³

The money is real and the technology works. So where does the value go?

One use case at a time

Most organisations bring AI in the way they brought in every earlier technology: one use case at a time. A chatbot for HR questions. A summariser for contracts. A copilot for one team. Each use case gets its own project, its own integration and its own training.

This feels safe. In practice it is slow and expensive, and it rarely scales. Each use case needs its own data, its own approvals and its own champion. Employees end up with five assistants that know nothing about each other or about their actual work. After a year, there is a portfolio of pilots and very little change in how the organisation runs.

Organisations are built around human limits

The deeper reason is in the design of the organisation itself. Hierarchies, processes, handovers, status meetings and approval chains exist because people can only do, coordinate and remember so much. A manager can follow a limited number of reports. A team can keep only so many threads in mind. Every handover loses context, so we add a meeting to recover it.

These structures were sensible answers to human limits. AI moves those limits. It can take on much of the execution, the coordination and the memory. Purpose, judgement, relationships and accountability stay with people.

When AI is added to a structure built for the old limits, it can only make small parts of that structure a little faster. The meetings still happen, the handovers still lose context and the approvals still wait in someone’s inbox. Real value comes from redesigning the organisation.

Three shifts

The shifts we build for

AI as a toolAI as a workforceAI does part of the work, with a role, a mandate and a record

Many appsOne backboneWork, context and decisions live in one system

Rented intelligenceOwned intelligenceWhat the organisation learns stays inside its boundary

Three shifts of the AI-native organisation

We see three shifts in the organisations that get this right.

From AI as a tool to AI as a workforce. AI stops being something an employee opens and starts doing part of the work, with a clear role, mandate and record.
From many apps to one backbone. Work, context and decisions live in one system, so AI and people see the same picture.
From rented intelligence to owned intelligence. What the organisation learns from its own work stays with the organisation, inside its own boundary.

What this means for leaders

The first question is how work should be organised now that AI can do part of it. The question of which tool to buy comes after that. Three questions are worth asking at your next leadership meeting:

Which decisions in our organisation wait longest, and what are they waiting for?
How much of our people’s week goes on moving information between systems and meetings?
If an AI did half of a role’s routine work tomorrow, what would we want that person to do instead?

The answers point to where redesign pays first. In the next article, we describe what an organisation looks like once that redesign has happened.

People and AI, working as one. Talk to us

Sources