For several years, executives have been asking whether AI could create meaningful business value. That question is beginning to be answered.

New BCG research involving 152 CEOs of companies with annual revenues above $500 million found that nearly nine in ten already see cost or revenue benefits from AI in targeted areas. The experimentation phase is producing results. Yet most organisations remain unable to scale those results into measurable enterprise-wide impact. More than half of the CEOs surveyed still struggle to connect AI initiatives directly to the P&L. Only 14% have clearly defined the expected P&L impact for every AI initiative.

This is an important shift. The primary AI challenge is no longer proving that the technology can work. It is building the execution system required to turn local successes into repeatable business performance.

The first stage of enterprise AI was naturally dominated by exploration. Organisations deployed copilots, created innovation teams, tested generative AI tools and developed portfolios of potential use cases.

Many of these experiments worked.

Marketing teams accelerated content production. Sales teams prepared proposals faster. Customer-service functions automated routine interactions. Finance teams shortened reporting cycles. Employees saved time researching, analysing and documenting their work.

These benefits are real. But they can also create a false sense of progress.

A company may report dozens, or even hundreds, of successful AI initiatives while its underlying economics remain largely unchanged. Productivity improves in individual activities, but revenue growth, conversion, margin, customer retention, launch velocity or decision quality barely move.

The organisation has adopted AI. The business has not yet been redesigned around it.

BCG’s findings capture this divide. Nearly two-thirds of surveyed CEOs say their companies pursue AI pilots, but only 26% have embedded AI within a broader business transformation. High-performing companies are approximately seven times more likely to redesign workflows and reshape the business end to end, rather than simply automate existing work.

Most AI initiatives begin with a task.

Can AI write this document? Can it analyse these data? Can it generate this campaign? Can it answer this customer request? Can it automate this report?

These are useful starting questions, but they rarely expose the full opportunity.

The greater value usually appears when leaders ask a different question:

How should this complete business outcome be delivered now that AI can perform, support or coordinate part of the work?

That distinction matters.

Automating a task preserves the existing workflow and makes one component faster. Redesigning the workflow changes how work moves across functions, how decisions are made, where human judgement is required and who remains accountable for the outcome.

Consider a product launch. AI can accelerate competitive research, generate positioning alternatives, produce content and summarise customer feedback. Each use case may save time. But the larger opportunity is to redesign the complete launch system.

Market signals could be captured continuously. Customer expectations could be compared with the value proposition earlier. Claims could be tested before final product decisions. Sales and marketing materials could be generated from a shared source of truth. Retailer feedback could immediately influence activation. Post-launch signals could trigger corrective action before underperformance becomes entrenched.

The value does not come from adding AI to separate tasks. It comes from connecting those capabilities into a faster and more responsive execution loop.

It is tempting to explain weak AI returns through technology limitations, data quality or insufficient employee adoption. These barriers matter, but BCG’s research points towards a deeper management problem.

More than half of the CEOs surveyed identify linking AI to the P&L as a major obstacle, yet only 14% define the financial impact of every AI initiative. People redesign is identified as a barrier by 55% of CEOs, but HR is involved in AI governance at only 30% of companies, compared with 82% involvement for technology functions. Weak value tracking and insufficient funding for people and change management remain significant obstacles.

Companies recognise what is preventing scale, but their governance and investment choices do not yet reflect that recognition.

This is the AI execution gap.

AI programmes are frequently launched without establishing:

Without these elements, organisations measure deployment rather than impact. They track licences, users, prompts, agents, pilots and hours saved. These indicators may show activity, but they do not prove that the business is becoming more competitive.

A sales copilot used by 500 employees is not necessarily valuable. It becomes valuable when it measurably improves account coverage, proposal quality, conversion, sales-cycle duration or revenue per salesperson.

A marketing content engine is not valuable because it generates more assets. It becomes valuable when it improves relevance, speed to market, campaign performance or the team’s capacity to address priority audiences.

A customer-service agent is not valuable because it handles conversations autonomously. It becomes valuable when it reduces service costs while protecting resolution quality, satisfaction and retention.

The unit of AI transformation should therefore not be the tool or the use case. It should be the business outcome and the workflow that produces it.

Broad experimentation was rational when capabilities, risks and potential applications were poorly understood. Organisations needed to learn.

But the economics of the next phase are different.

Every additional pilot consumes management attention, technical resources, employee time, governance capacity and change effort. When experiments are allowed to accumulate without clear prioritisation, the AI portfolio becomes fragmented.

Different teams procure overlapping tools. Similar use cases are developed independently. Data and security decisions are repeated. Employees receive contradictory guidance. Pilots remain active because nobody wants to declare them unsuccessful. Leadership sees significant movement but lacks a reliable view of value.

The result is not only wasted investment. It is organisational complexity.

BCG recommends that companies concentrate people, capital and capability on a limited number of high-value areas capable of changing the business, rather than continuing to experiment broadly. It also argues that the expected P&L impact and value logic should be defined before an initiative begins, with finance validating results from the outset.

This demands a more selective approach.

An AI initiative should earn the right to scale by demonstrating a credible connection between capability, workflow change and measurable impact. Those that cannot should be redesigned, merged or stopped.

AI should not become a portfolio personally managed by the CEO. Nor should it remain primarily owned by the technology function.

The CEO’s role is to establish the conditions under which the organisation can repeatedly convert AI opportunity into business value.

That requires orchestration across strategy, operations, technology, finance, HR and the commercial functions.

BCG describes the CEO as the orchestrator while placing delivery accountability with CXOs and P&L owners. This distinction is essential. Technology leaders can ensure that systems are secure, reliable and scalable. They cannot independently own improvements in revenue, margin, customer experience or operating performance.

Business leaders must be accountable for business results.

The CEO agenda should therefore move beyond questions such as:

The more consequential questions are:

The objective should not be generic productivity. Leaders should identify where AI could materially influence revenue, margin, conversion, working capital, launch speed, service economics or customer lifetime value.

The focus should move from individual tasks to the complete chain of activities producing the desired outcome.

Every significant AI initiative needs an accountable business owner. Technology may enable the solution, but a P&L or functional leader must own the impact.

The baseline, expected outcome, value logic and measurement method should be defined before implementation. Finance should be able to validate the result.

AI transformation alters roles, skills, decision rights and collaboration. Training employees to use a tool is not the same as redesigning how work gets done.

Concentration requires choices. Initiatives that remain disconnected from strategic priorities or measurable outcomes should not continue indefinitely.

The disciplines behind successful transformation are not new. Strategic focus, executive accountability, workflow redesign, capability building and value tracking have always mattered.

AI raises the stakes.

Its potential value is larger because it can influence knowledge work, customer interactions, decision-making and operational processes simultaneously. But its speed and accessibility also make fragmentation easier. Teams can launch tools and experiments faster than the organisation can govern, integrate or evaluate them.

Weak execution can therefore multiply rapidly.

A conventional transformation might suffer from slow implementation. A poorly governed AI transformation can generate an expanding ecosystem of disconnected agents, tools, workflows and data flows, each creating activity without necessarily improving the enterprise.

Technology is accelerating the rate at which organisations can act. It is not automatically improving the quality of what they choose to do.

This is why AI maturity should not be defined only by technical sophistication. A mature AI organisation is one that can consistently:

That capability is execution intelligence.

Access to advanced AI capabilities will continue to broaden. Models will improve. Costs will decline. Tools will become easier to deploy. Competitors will increasingly have access to similar technologies.

Technology alone will therefore provide a narrowing window of differentiation.

The more durable advantage will come from the organisation’s ability to translate new capabilities into changed decisions, workflows and economics faster than its competitors.

Some companies will continue accumulating AI use cases. Others will develop an operating system for AI-enabled transformation.

The first group will demonstrate activity. The second will compound value.

Do you know how many AI initiatives are currently active across your organisation?

More importantly, do you know:

The first question measures activity.

The others reveal whether AI has become an execution capability.

Is your organisation scaling AI, or merely scaling the number of AI initiatives?

Boston Consulting Group, CEOs Are Starting to See Value from AI. Now Comes Execution, July 2026. The underlying CEO research behind the findings discussed in this article.

Boston Consulting Group, Nearly Nine in Ten CEOs See Some Cost or Revenue Benefits from AI in Targeted Areas, But Most Are Struggling to Scale It, July 2026. A concise summary of the execution, governance and P&L gaps identified across 152 CEOs.

Boston Consulting Group, From Potential to Profit: Closing the AI Impact Gap, 2025. Earlier BCG research finding that only a small minority of companies had reached the maturity required to generate transformative bottom-line impact at scale.

The Strategic Brief, The Execution Gap Is Now Measurable. How external evidence, operating signals and structured assessment can expose where strategy is being lost in execution.

The Strategic Brief, Why AI Projects Fail Before the Technology Does. Why unclear ownership, fragmented workflows and weak value logic undermine AI initiatives before technical limitations become decisive.

The Strategic Brief, The AI Capability Gap Is Becoming a Business Risk. Why access to AI tools does not automatically create the organisational capability to use them effectively.