AI: Why 94% of Companies Remain Skeptical

AI in 2025: Massive Use but Little Perceived Value

According to an in-depth study conducted by McKinsey, titled The State of AI in 2025: Agents, Innovation, and Transformation, an overwhelming majority of companies—nearly 90%—will have integrated artificial intelligence into at least one of their business functions by the end of 2025. However, a surprising finding emerges from this survey: 94% of these organizations have not yet observed significant value stemming from their AI investments. This gap, explored in detail in the article Where AI Will Create Value and Where It Won't published in the McKinsey Quarterly in April 2026, does not lie in the adoption of AI itself, but rather in how this technology is framed and utilized. Most companies are merely using AI to accelerate their existing processes, while true returns on investment require a radically different approach to work.

A concrete example illustrates this issue: a team managed to reduce its discovery cycle from six weeks to just ten days thanks to AI. While this time gain is impressive and the output is real, it has not led to any significant new discoveries. The questions asked and the answers obtained remained unchanged, simply acquired more quickly. This highlights a use of AI that, while efficient in terms of speed, does not add value in terms of content or innovation.

Through the teams and founders I work with, I have observed that strategic questions are becoming increasingly smaller. We are moving from fundamental questions like "what is worth building?" to more operational questions such as "how can we test this faster?" or "can we synthesize these interviews more efficiently?". This trend shows that AI is often used to optimize existing processes rather than to explore new opportunities.

Joe Smiley described a similar dynamic in his article for UX Collective, titled The Most Popular Experience Design Trends of 2026. He explains that AI "reduces the time between idea and artifact, which seems like progress. But when everything can be generated instantly, teams miss out on the fundamental parts of the design process: framing, research, and exploration." Thus, the strategic questions that were once at the heart of discovery have gradually receded, or even disappeared in some cases.

Productivity: A Defensive Reflex

According to McKinsey, the productivity gains achieved through AI are primarily defensive in nature. Indeed, competition tends to erode these gains, and it is often the customers who benefit the most. Productivity redefines the minimum threshold of industrial performance but does not contribute to raising the maximum level. This means that if a company's AI strategy is limited to productivity gains, it risks stagnating on a treadmill where everyone moves at the same speed.

A Revisited Historical Pattern

The history of electricity adoption in factories illustrates a similar pattern. This story is recounted by Ajay Agrawal, Joshua Gans, and Avi Goldfarb in their book Power and Prediction. Initially, productivity did not increase much because manufacturers simply replaced steam engines with electric motors without rethinking the layout of factories. It was only when factories were reorganized around the possibilities offered by electricity that the true revolution occurred. McKinsey applies this parable to AI by identifying three waves of value: productivity, differentiation, and reduction of transaction costs. These waves overlap and increase in complexity, with productivity being the most accessible, while differentiation and reduction of transaction costs require bolder strategic decisions.

The waves are ranked in order of difficulty. The first wave, that of productivity, is the easiest to achieve because it does not require making new decisions. The second wave, that of differentiation, demands determining what is worth offering. Finally, the third wave, that of reducing transaction costs, involves betting on how an industry will evolve. As one progresses through these waves, the question shifts from "how to do the work faster" to "what the work should be."

The Pitfalls of Productivist Thinking

Teams and founders are particularly vulnerable to the temptation of productivity gains, as discovery seems to benefit directly. For example, interviews can be automatically transcribed and analyzed, and hypotheses mapped from notes. However, there is a crucial difference between accelerating discovery and transforming it. It is in this transformation that the most complex questions reside.

José Torre, in his article for UX Collective, emphasizes that AI allows for such rapid advancement that one can miss moments where attention and judgment are essential. Thus, compressing a six-week cycle into ten days is only valuable if the questions asked evolve, if new opportunities are explored, or if novel hypotheses are tested.

Rethinking the Framing of Questions

Most teams wonder how to use AI to make their discovery process more productive. While this question is natural, it may not be the most relevant. A more prudent approach would be to ask what AI could change about what is worth building. This framing question assumes that the offerings, customers, and business models worth exploring may have radically changed with AI.

This is the difference between using AI to find better answers to existing questions and using AI to ask better questions in the first place. The former is a productivity practice, while the latter is a strategic practice. They yield very different long-term returns.

The True Beneficiaries of Productivity Gains

A study conducted by Erik Brynjolfsson, Danielle Li, and Lindsey Raymond on over 5,000 customer support agents using an AI conversational assistant revealed an average productivity gain of 14%. However, this figure masks a more nuanced reality: novice workers saw their performance improve by 34%, while experienced workers hardly progressed. Thus, AI has primarily raised the performance floor rather than the ceiling.

In this context, AI has not elevated the ceiling for already high-performing workers. It has raised the floor for those who were not. This aligns with McKinsey's argument on a smaller scale: productivity gains often go to those who are closer to the floor than to the ceiling.

Sustainable returns come from a different type of work. In my previous article for UX Collective, The Anatomy of Product Discovery Judgment, I described this as framing judgment: the work of asking, before an artifact is created, what problem is worth solving, which customer is worth serving, and which hypothesis should be tested first.

These decisions are upstream of every artifact a team produces. They are also where AI's productivity gains help the least, and where human judgment accumulates the most.

A Competitive Readjustment

McKinsey concludes that AI does not constitute a revolution in productivity but rather a competitive readjustment. The companies that will succeed will not be those that adopt AI the fastest, but those that think most deeply about the direction value is taking. For teams and founders, this is primarily a discovery problem. The teams that will stand out will be those that have asked the right questions about their offerings, business models, and customer relationships.

Discovery has always been about what is worth building. AI raises the stakes by lowering the cost of error in this regard.

The Urgency of Asking the Right Questions

The allure of productivity is understandable, as it is easy to measure and celebrate. However, the true competitive advantage lies in the ability to ask the right questions. AI does not answer these questions, but it makes them more urgent. The teams that will make the most of AI will be those that take the time to make informed choices.

This has always been the discovery question. The window for asking it has narrowed.

Where to Start

To assess the impact of AI, it is helpful to list the last three important discovery decisions made by a team. If the role of AI in these decisions would have been the same five years ago, it indicates that AI is being used to accelerate old work rather than transform the discovery process. Two questions to ask before the next discovery cycle: what problem are we seeking to understand more deeply? And what would need to be true for our current offering to be the wrong one? These questions help refine the framing and create space for surprising answers.