In recent months, reports on the state of AI have been piling up, each heralding a major disruption in the way businesses work, make decisions, and create value. Two publications by major strategy consulting firms have just been released in quick succession. McKinsey sees fast but still hesitant adoption, while BCG sees a polarization between a minority that captures almost all the gains and a majority that is going around in circles. Taken separately, these reports shed light on part of the subject, but read together, they say something deeper about how organizations approach transformation: with a lot of technology, a lot of stated ambition, and very little work on what really matters.
Indeed, behind the graphs and percentages, there is one reality that stands out: AI does not lack capabilities but it is organizations that lack the structure to absorb them. Pilot projects are multiplying but not changing anything, gains remain marginal, agents appear more in speeches than they transform value chains, and performance gaps are widening for reasons that, ultimately, have little to do with the models themselves.
This post offers not only a summary but also a critical reading of what these reports show, what they avoid addressing, and above all what they reveal, despite themselves, about the real state of businesses in the face of AI. Because understanding the numbers is not enough; we also need to understand the structures that produce them.
In short:
- AI is widely adopted in businesses, but it generates little value due to a lack of structural transformation: its uses are multiplying, but the economic benefits remain marginal and concentrated among a minority.
- Reports from McKinsey and BCG agree on the conditions for success: value creation requires a profound redesign of workflows and ambitious governance, not isolated experiments.
- Most businesses do not have the organizational foundations to take advantage of AI, unlike “future-built” businesses, an atypical and difficult-to-replicate category that captures most of the gains.
- AI agents are seen as an emerging value driver, but their adoption remains limited, and their organizational implications (reduction of tasks, reconfiguration of roles, increased complexity) are rarely analyzed.
- The reports ignore key factors such as cognitive overload, internal friction, and fragmentation of practices that hinder the integration of AI into daily work and limit its real impact.
- Two different perspectives
- Points of convergence
- Notable differences
- AI is widely adopted but marginally productive
- Workflow redesign: yes, but no real consideration of political and social conditions
- Future-built companies: an overrated category that is difficult to replicate
- The role of agents is overestimated in communication and underestimated in terms of implications.
- Where are the attention economy, organizational friction, and cognitive costs?
- Bottom Line
- To answer your questions…
Two different perspectives
McKinsey – State of AI 2025: Agents, innovation, and transformation
This is a global survey of around 2,000 organizations on the actual use of AI, with a focus on adoption, diffusion by function, measured impacts (costs, revenues, innovation), and the emergence of agents (McKinsey – State of AI 2025: Agents, innovation, and transformation).
It is a descriptive barometer that seeks to answer questions such as “Where does AI stand in the business?” and “Why does large-scale value remain the exception?”
Key figures
- 88% of companies use AI in at least one function (up), but only 32% are in the scaling phase and 7% are fully scaled.
- Agentic AI: 23% of businesses are scaling agents, but in only 1 or 2 functions.
- 39% see an impact on EBIT (often less than 5%).
- The impact is mainly qualitative: innovation (64%), employee satisfaction, competitive differentiation.
- Reasons for the low overall impact: workflows have not been redesigned, difficulty in moving from pilot to scale.
- The “high performers”, who represent 6% of organizations, show stronger ambitions, growth and innovation objectives, and workflow redesign (three times more likely than elsewhere).
Here we have a detailed picture of reality: widespread use, limited value, agents gaining power but in too scattered a manner.
BCG – The Widening AI Value Gap (2025)
This is a strategic study of 1,250 businesses, focusing on value creation and the growing gap between “future-built” businesses (top 5%) and the rest (The Widening AI Value Gap – Build for the Future 2025).
It is deliberately more prescriptive in its approach, identifying winning capabilities and explaining the failures of the vast majority.
Key figures:
- 5% of businesses derive value on a large scale, 60% derive none, and 35% are partially scaling.
- Future-built companies show:
- 1.7× more growth
- 1.6× more EBIT margin
- 3.6× more TSR (Total Shareholder Return) over 3 years
- 70% of AI value is concentrated in core functions (sales, marketing, supply chain, manufacturing).
- Agents: they already represent 17% of AI OA value in 2025, expected to reach 29% in 2028.
- Critical accelerators: multi-year vision sponsored by the CEO, AI-first operating model, workflow redesign, massive upskilling, modular architecture.
Here we have a perspective focused on strategy and performance gap, prescriptive and oriented towards senior executives.
Points of convergence
Unsurprisingly, the two reports converge on many points, even if their figures sometimes differ significantly or complement each other due to complementary angles.
Widespread adoption but still limited value
According to McKinsey, 88% use AI in at least one function, but only 39% see an impact on EBIT, most often of less than 5%.
According to BCG, 60% generate no material value despite investments, and only 5% create substantial value at scale.
We can therefore agree that there is widespread use, but that this translates into rare and polarized economic value.
Workflow redesign as a determining factor
McKinsey tells us that high performers are three times more likely to redesign workflows in depth.
BCG echoes this sentiment: value comes from end-to-end workflow transformation, not isolated pilots. “Future-built” companies reinvent processes rather than superficially automating them.
Both agree that without organizational transformation, technology alone generates nothing.
Value is concentrated in a few core functions
McKinsey tells us that profits are mainly found in marketing & sales, strategy/finance, product development, IT, and manufacturing.
The BCG makes the same observation: 70% of AI value comes from sales, marketing, supply chain, manufacturing, and pricing functions.
Both firms tell us that value is functional, not cross-functional, and depends on well-defined operational flows.
Leaders adopt an ambitious stance in terms of innovation and growth
According to McKinsey, high performers rely on innovation and growth, not just efficiency, and they aim for organizational transformation (3.6 times more often than others).
BCG, meanwhile, tells us that “future-built” companies have a multi-year vision driven by the CEO with ambitious numerical targets, an AI-first operating model, and more investment (+26% in IT, +64% in AI budget).
Both reports therefore lead us to believe that strategic posture matters more than technology itself.
Notable differences
However, the two reports cannot be compared directly, as not only do they take different approaches, but their interpretation of certain phenomena also differs.
Purpose of the report: descriptive vs. prescriptive
McKinsey provides a factual, photographic overview based on adoption and real impact, while BCG conducts a prescriptive strategic analysis on how to create value and avoid falling behind.
One measures, the other prescribes, so the two reports are best treated jointly if we want to get something out of them.
Focus on “usage” vs. “value”
McKinsey focuses on usage, dissemination, and operational maturity, while BCG focuses on the value generated, the gap between leaders and followers, and winning capabilities.
There are therefore two different levels: one operational (McKinsey) and the other strategic/economic (BCG).
Relative weight of agents
McKinsey is exercising the utmost caution, emphasizing the currently marginal nature of agents (less than 10% in scaling by function).
BCG is more convinced, as agents are presented as a major value accelerator: already 17% of value in 2025, 29% in 2028.
Surprisingly, McKinsey sees pilots, while BCG sees a strategic accelerator lever, although the two can complement each other. It is also a question of horizons.
Organizational analysis
McKinsey highlights specific elements: workflow redesign, ambitions, AI objectives, and the size of the business, while BCG emphasizes an AI-first operating model, business-IT co-responsibility, the role of the chief AI officer, and modular architectures.
BCG therefore goes much further in its organizational recommendations and target architecture.
Reading the overall landscape
McKinsey shows us broad but slow progress and difficulties in scaling, while BCG emphasizes a growing gap, a “winners-take-most” dynamic, and a risk of stagnation for 60% of businesses.
McKinsey therefore describes an evolving landscape, while BCG describes a polarized and unequal landscape.
Let’s now take a critical look at this.
AI is widely adopted but marginally productive
Whether at McKinsey (88% usage, 32% only in the scaling phase, 39% with low EBIT impact) or at BCG (60%generate no value, only 5% capture significant value), we are told the same story, but it is a story that is nothing new, and we may even be disappointed at how long it took to recognize such predictable evidence.
This dynamic is similar to all major technological waves, with widespread use and significant value, but only captured by a small number of businesses. There is a very visible threshold effect here, with most businesses adopting AI as a tool rather than a lever for transformation.
The reports therefore show above all that AI amplifies existing structures. If the organization is fragmented, AI adds to the fragmentation, but if it is integrated, AI becomes a multiplier (If your business isn’t designed for AI, it will end up being designed by AI).
But in fact, this hides the emptiness of a discourse on AI that remains too superficial. It tells us how businesses use AI, but not why they are failing to do so. And the real reasons are rarely technological.
In general, it is also regrettable that organizations think in terms of “tools” rather than “value production systems”.
Workflow redesign: yes, but no real consideration of political and social conditions
Both reports hammer home the point that workflow redesign is the decisive factor. According to McKinsey, high performers are three times more likely to redesign workflows, and according to BCG, value comes from end-to-end reinvention, not pilot projects.
Technically speaking, they are right, because without an overhaul of business chains, AI remains tied to obsolete processes, but they overlook the operational reality: a workflow is not just a diagram, it is also a political territory, with actors, rules, routines, and power relations.
What is omitted is that redesigning workflows involves renegotiating scopes, roles, and hierarchical boundaries. It also calls into question middle managers, decision cycles, and historical control models.
Ultimately, and this is nothing new, many organizations fail not because of a lack of technology, but because of an inability to reallocate authority.
Reports talk about ambition, governance, and vision, but never about the internal power games that block 80% of transformations.
Future-built companies: an overrated category that is difficult to replicate
BCG focuses on an elite group comprising less than 5% of organizations.
These businesses already have a high level of digital maturity, stable leadership, advanced technological architectures, and the financial margins to invest continuously, but statistically they are in a very atypical zone: large global businesses, often tech-native or already transformed.
This argument is open to criticism because it wrongly generalizes what is in reality only possible for a minority, underestimates the significant cost, time required, technical culture, and organizational quality of these businesses, and above all fails to mention that most businesses do not have a future-built structure and never will.
In short: the gap is widening not because some are moving too fast, but because the majority do not have the structural foundations to keep up.
The role of agents is overestimated in communication and underestimated in terms of implications.
For BCG, agents already account for 17% of AI value, which is a strong, even optimistic narrative. But the expectation of 29% by 2028 assumes fast adoption despite significant constraints (data, security, architectures).
However, the report fails to mention that most businesses do not have the data quality, governance, or internal skills required for reliable agents.
More generally, it promotes a “platform-centric” vision that is far removed from the average reality.
But McKinsey is not immune to criticism on this subject either. Agents are described lucidly (limited use, exploratory, not very scalable), but the discussion of the organizational impact (skills, distribution of roles, partial automation of decisions, etc.) is only touched upon.
Both reports quickly gloss over a key point: agents reduce the granularity of work by merging tasks that were previously distributed across several roles. This is an organizational time bomb, but nothing is said about it.
Where are the attention economy, organizational friction, and cognitive costs?
Surprisingly, certain key topics are not addressed.
The widespread introduction of new tools, including AI, further muddies an already crowded digital landscape. The proliferation of interfaces, notifications, and workflows directly impacts the attention span of teams, who must manage increasing cognitive overload even as they are promised greater simplicity. As everyone adopts their own uses, their own shortcuts, and sometimes their own tools, practices become fragmented and operational consistency breaks down. AI then no longer evolves in a stable environment, but in an assembly of fragile processes, imperfect data, and heterogeneous routines, which it tends to amplify rather than correct. This gap is further reinforced by the diversity of individual uses: some employees make intensive use of AI capabilities, while others bypass or ignore it, creating performance gaps that are difficult to manage and complicating any attempt at standardization.
These reports remain focused on the formal value chain, never on the informal one, which is nevertheless critical to adoption.
In reality, there is a risk that AI will exacerbate existing frictions as much as it will create gains.
Bottom Line
Both reports show that AI is not a technological issue but a problem of organizational design.
Mass adoption without redesign produces little value.
The gap between leaders and followers stems from the ability to redefine work, roles, decision-making systems, and internal architectures, not from the technology itself.
To answer your questions…
Despite widespread adoption, the economic impact remains limited because AI is often integrated as an additional tool rather than as a lever for transformation. Pilot projects are multiplying without changing value chains, which reduces potential gains. Fragmented organizations see their dysfunctions amplified, while more structured ones truly benefit from AI. The main obstacle is organizational: without an appropriate working model, value does not materialize. For decision-makers, the challenge is to treat AI as a system change, not just a technology.
Businesses that redesign their workflows capture much more value because they reorganize activities rather than adding tools to outdated processes. However, changing a workflow means changing roles, responsibilities, and decision-making processes, which creates internal tensions that are often underestimated. The difficulty stems less from technology than from renegotiating authority and boundaries. For leaders, anticipating and managing these adjustments is essential for successful AI integration.
Organizations described as “future-built” represent only a very advanced minority: high digital maturity, stable leadership, solid architectures, and significant investment capabilities. Their model is difficult to replicate because most businesses do not have these foundations in place. The performance gap stems as much from their lead as from the structural limitations of others. For leaders, the priority is to strengthen the organizational foundations before hoping to replicate these models.
Agents are presented both as a source of growing value and as a technology that is still very unscaled. Their real impact lies in transforming tasks and redefining roles, an aspect that is largely overlooked. Many organizations still lack reliable data, governance, and skills to deploy them effectively. For leaders, their adoption must be viewed as a redesign of work, not just another form of automation.
Several key issues are rarely addressed: cognitive overload caused by the proliferation of tools, fragmentation of practices, and the importance of the informal value chain. AI is being introduced into an already unstable environment, which may increase friction rather than reduce it. Differences in usage among employees also make standardization difficult. For leaders, taking these everyday realities into account is crucial to prevent AI from exacerbating organizational disorder rather than improving performance.
Image credit: Image generated by artificial intelligence via ChatGPT (OpenAI)
Related articles:
Generative AI in the workplace: revolution or illusion?
Generative AI: your organization is worth more than your tools
AI won’t kill consulting, but it will force it to grow
How can we measure the productivity gains of AI?
Who is handling your artificial intelligence projects? Probably not the right people.
Enterprise 2.0 does not tolerate halve measures
