AI is exposing the gap between strategy and execution in organisations

A few years ago, we wrote about why digitisation so often failed – not because technology was lacking, but because organisations tried to innovate without changing how decisions were made, how teams were structured, and how work actually flowed. 

At the time, digitisation was the dominant narrative. Today, AI has taken its place. What is striking this time is not how much has changed, but how familiar the patterns are.

AI is now firmly on the board agenda. Pilots are launched at speed. Generative tools are rolled out across functions. Investment continues to rise. And yet, for most organisations, meaningful impact remains elusive.

The data is remarkably consistent.

• A 2025 MIT Sloan study found that 95% of enterprise generative AI pilots produced little or no measurable business impact, with only a small minority scaling successfully beyond experimentation.
• Boston Consulting Group’s 2025 global AI survey reported that only around 5% of companies are realising significant value from AI, while close to 60% report little to no benefit despite active initiatives.
• McKinsey has observed that around 40% of digital and AI transformations stall at the scaling phase, never embedding into day-to-day operations.
• Deloitte’s AI Trends research repeatedly highlights the same blockers: unclear ownership, legacy processes, skills gaps, and weak alignment between strategy and execution.
• Harvard Business Review and MIT Sloan Management Review consistently point to organisational factors – culture, incentives, decision rights, and structure – as the decisive variables in AI success or failure.

This is not a temporary lag in AI adoption. It is a structural issue. As with earlier waves of digitisation, the conclusion is the same across methodologies and perspectives: AI does not fail at the technical level. It fails at the organisational one.

AI is not the solution. It is the stress test.

Many leaders approach AI – implicitly or explicitly – as a shortcut: to productivity, to better decisions, or around organisational complexity.

In practice, AI behaves very differently. It acts as a stress test, amplifying whatever already exists inside the organisation. In practice, this shows up in predictable ways: fragmented adoption, pilots that fade after early success, and AI outputs that fail to change how decisions are made.

This is not a failure of ambition, but a failure of readiness. At the risk of repeating ourselves, we have seen this before. During earlier waves of digitisation, organisations adopted new tools while leaving underlying structures untouched – how decisions were made, how teams were organised, and how work actually flowed. The result was progress on the surface, and friction underneath.

AI compresses that gap. It exposes structural misalignment faster and more visibly than previous technologies ever did. As McKinsey and others have increasingly emphasised, the challenge is not deploying new tools but rewiring the organisation itself – its talent models, workflows, and decision-making structures – so that AI can be absorbed into daily operations. Without that work, the outcome remains the same: impressive pilots, limited value.

The real constraint: organisational legibility

All of these failure modes point to the same underlying constraint: organisations struggle to clearly see and understand how they actually operate.

In our work at Up Strategy Lab, this shows up with striking consistency. In these organisations, things look coherent on paper. But operationally, a different reality emerges, and shows up in familiar ways:

• Work flows cut across teams in informal, undocumented ways
• Decisions are delayed by invisible dependencies
• Capabilities are assumed rather than understood
• Skills are treated as static attributes instead of dynamic resources
• Teams are assembled by role and availability, not by capability fit

This lack of legibility was inconvenient in the past. With AI, it becomes a hard constraint. AI systems require clarity: clear inputs, clear ownership, clear integration into real work, and clear accountability for outcomes. Without this, AI remains trapped in experimentation – no matter how sophisticated the model.

Why pilots work – and then stall

One of the most widely observed patterns in AI adoption is what happens after the pilot. Pilots tend to work. They demonstrate technical feasibility, surface interesting insights, and generate early enthusiasm. And then, quietly, they stall.

The reasons are rarely technical. What tends to stop progress is that nothing around the pilot changes. Ownership becomes ambiguous once the project phase ends. Outputs are not absorbed into existing workflows because those workflows were never redesigned to take them in. Decision rights remain where they were before, leaving teams unsure who can act on new information. Incentives continue to reward existing behaviours, not AI-enabled ones. And without a clear skills strategy, early learning never compounds beyond a small group of individuals.

In effect, the organisation stays the same while the technology moves on. AI cannot compensate for that imbalance. This is why firms like BCG increasingly emphasise that value is not created by introducing AI into existing ways of working, but by redesigning work end to end. The real gains appear when organisations stop treating AI as an overlay and start rethinking how work flows across functions, not just within them.

Skills: the invisible infrastructure problem

This brings us to one of the least discussed – and most consequential – dimensions of AI readiness: skills clarity (which is also our favourite subject).

Skills matter here not because AI requires more training, but because skills are how organisations understand what work they can realistically change. Without a clear, shared view of capabilities, leaders cannot decide where AI should redesign workflows, shift decision rights, or redistribute responsibility. In that sense, skills are not an HR concern; they are execution infrastructure.

When skills are treated as static role attributes rather than as a dynamic view of organisational capacity, AI has nowhere to land. Leaders cannot confidently determine which teams can absorb AI-enabled ways of working, where gaps will stall progress, or how capabilities need to evolve. The result is predictable: AI becomes another layer added to an already opaque system, rather than a lever for redesigning how work is actually done.

When speed hides inertia

This pattern is often reinforced by another illusion: speed.

Progress appears rapid because tools are deployed quickly, pilots launch within weeks, and dashboards materialise overnight. But this surface-level velocity often masks a deeper inertia. Beneath it, decision-making remains slow, ownership fragments, and escalations persist exactly as before.

Again, what changes is the technology. What does not change is how the organisation moves. AI is layered onto existing processes without redesigning how decisions are made, who has authority to act, or how work flows across teams. As McKinsey has repeatedly argued, applying AI to existing processes without redesigning operating models tends to produce incremental efficiency gains rather than structural improvement.

AI accelerates execution only when the organisation itself is designed to move. Without clarity on decision rights, accountability, and workflow, faster tools simply make organisational friction more visible, not less.

This is not an AI problem

It is tempting to frame all of this as an AI maturity issue. That would be convenient – and incorrect.

The obstacles now surfacing did not arrive with AI. Siloed structures, slow governance, role-based staffing, limited skills visibility, and weak links between strategy and execution have shaped organisational performance for years. AI simply removes the buffer. It compresses the gap between intent and reality, making structural weaknesses visible faster and harder to ignore.

This is why efforts to treat AI as a technical rollout consistently disappoint. As experts have argued before: AI functions as a cultural transformation. Value emerges only when adoption, trust, and behavioural change are addressed alongside technology. Leaders who recognise this focus less on deploying tools and more on reshaping how work is coordinated, how decisions are made, and how accountability is defined.

The implication is straightforward. Organisations do not get a free pass on fundamental change management simply because the technology is new or exciting. AI does not bypass organisational constraints – it exposes them.

Organisational clarity comes first

Across research and practice, organisations that succeed with AI share a common trait: They do not start with tools but with making the organisation legible to itself by by rewiring how work, decisions, and capabilities are organised.

Before attempting to scale AI, they invest in understanding how work actually flows across teams, where decisions are truly made, and how responsibility is distributed in practice. They develop a clear view of their capabilities – not as static roles or titles, but as the real skills and combinations of skills that determine what the organisation can execute. Teams are designed around outcomes rather than org charts, and incentives are adjusted so that new ways of working are reinforced rather than quietly resisted.

This is not visible work. It does not generate headlines or quick wins. But it is the work that determines whether AI creates sustained value or persistent friction.

As Sylvain Duranton, Global Leader of BCG X, said: “Companies cannot simply roll out GenAI tools and expect transformation. The real returns come when businesses invest in upskilling their people, redesign how work gets done, and align leadership around AI strategy.”

Where strategy breaks – or gets executed

This is the space where Up Strategy Lab operates: where organisational structure determines whether strategy can actually be executed.

Our work does not sit at the level of tools, nor at the level of abstract frameworks, but at the intersection of strategy, organisation, and execution. In many engagements, progress begins with a deceptively simple question: is the organisation clear enough about itself to change?

That question often exposes blind spots that technology alone cannot resolve. Decisions that appear clear on paper turn out to be fragmented in practice. Capabilities assumed to exist are unevenly distributed. Work flows in ways no one has explicitly designed or taken ownership of.

In some cases, creating that clarity requires changes to operating models. In others, it requires a more precise understanding of how skills are organised, combined, and deployed across the business. It was this recurring challenge – encountered repeatedly in consulting work – that led to the creation of our skills management platform MuchSkills: not as a product in search of a problem, but as infrastructure designed to support organisational clarity where traditional tools fell short.

The insight behind it is simple. No tool or technology can compensate for organisational opacity. But the right infrastructure can help organisations see it – and address it – once they choose to.

The question leaders should be asking now

As AI initiatives accelerate, the most important question is not: What AI should we deploy?

It is: Are we structurally ready to use it?

That readiness is not defined by vendors, models, or pilots but by far more fundamental factors: clarity about capabilities, the speed and coherence of decision-making, visibility into skills, and the degree to which the organisation itself is aligned around how work gets done.

AI rewards organisations that understand themselves – and exposes those that do not.

A final thought

Every major technological shift repeats the same mistake. We overestimate what technology can fix and underestimate what organisations must change. AI is no different – except that it is faster, more visible, and less forgiving.

The challenges facing AI transformation in 2025–2026 are less about algorithms and more about organisations. This is not a coincidence. It is a repeat of the same digitisation playbook organisations have been struggling with for over a decade.

The organisations that succeed in the AI era will not be those with the most advanced models, but those with the clearest understanding of how their people, skills, and structures actually work.

If answering that question feels difficult internally, that difficulty itself is a signal worth paying attention to.