After investing hundreds of billions in artificial intelligence, a majority of executives still see no significant impact on revenue or costs. The paradox is only apparent. The problem is no longer the technology; it has become an organizational one.
More than half of the executives at the world’s largest companies say they do not see a tangible return on investment from their artificial intelligence initiatives. At first glance, this figure is surprising. Yet financial markets are valuing AI companies at record highs, data centers are proliferating, technology spending is skyrocketing, and every executive committee now has its own AI roadmap.
How can we explain such a disconnect between promise and results?
The answer may lie in a confusion that has become almost systematic: many companies have adopted artificial intelligence tools without truly transforming the way they work.
AI has made its way into organizations. It has not yet become part of their operational model.
The Trap of Constant Experimentation
Since the emergence of large language models, companies have ramped up their experimentation.
Internal chatbots, document assistants, office copilots, automation of repetitive tasks: use cases now number in the hundreds within large organizations.
Yet, in many companies, these initiatives remain confined to a few volunteer teams.
Experimentation has become a destination rather than a step along the way.
This situation is reminiscent of the early years of digital transformation. Even back then, some companies believed that launching a website was enough to become digital. Others reimagined their entire value chain.
Twenty years later, the winners are clear.
AI today seems to be following exactly the same trajectory.
The question is no longer whether a company has use cases. The question is whether it has succeeded in scaling them up.
Productivity is not a corporate project
Another misunderstanding arises in the way organizations evaluate their investments.
For two years, the main promise of AI has centered on productivity.
Produce faster.
Write faster.
Automate more.
Reducing certain operational costs.
These gains exist. They are real. But they are often insufficient to justify the massive investments made.
Saving a few minutes on an administrative task improves a metric. It does not transform a company.
Organizations that are beginning to see tangible results seem to be following a different approach. They use AI not only to do things better, but to create new sources of value.
The challenge is no longer solely about increasing employee productivity.
The challenge is to develop new services, accelerate innovation, or expand into new business areas.
In other words, the true return on investment does not always come from avoided costs.
It often comes from the additional revenue generated.
Data Before the Model
One of the most striking paradoxes of our time is the obsession with the models themselves.
Each new generation of LLMs dominates media attention.
Yet, in businesses, model performance is generally not the main limiting factor.
Data quality is.
Without robust governance, a reliable document architecture, and well-managed repositories, even the most advanced systems produce unreliable results.
Executives are gradually discovering a reality that is unspectacular but essential: AI primarily rewards companies that have done their groundwork on data.
Others often end up simply automating their existing disarray.
The Hidden Cost of Artificial Intelligence
Another issue is beginning to emerge on corporate boards: the economic management of AI usage.
Demos are often impressive when a few hundred employees use advanced models.
The equation changes when several thousand people interact with multiple LLMs every day.
Token consumption then becomes a strategic budget line item.
Some organizations are discovering that systematically using the most powerful models for simple tasks is like using a long-haul airplane for a short city trip.
The issue is no longer purely technological.
It becomes a financial one.
The most mature companies are now developing multi-model approaches, selecting the most appropriate tool for each task through a constant balancing act between cost, speed, and performance.
This discipline could become one of the key drivers of competitiveness in the coming years.
Learning as a Competitive Advantage
The most interesting lesson from this new phase ultimately has nothing to do with models, infrastructure, or even budgets.
It concerns the ability to learn.
For decades, competitive advantages were based on assets, patents, or economies of scale.
Artificial intelligence is gradually shifting the center of gravity.
Models are becoming accessible to everyone.
Infrastructure is becoming more widely available.
Technological capabilities are eventually spreading.
What remains a differentiator is the speed at which an organization learns to integrate these innovations into its processes.
The companies that will succeed are unlikely to be the ones with the best tools.
They will be the ones that can challenge their work methods more quickly than their competitors.
The real issue: transforming the business, not just adding technology
The most common mistake is still to view AI as just another layer added on top of what already exists.
Yet the first observable results show exactly the opposite.
Organizations that create value don’t simply integrate AI into their processes.
They redesign their processes around AI.
The distinction may seem subtle.
In reality, it is fundamental.
Because artificial intelligence is not merely a technological advancement.
Rather, it acts as an organizational catalyst.
It exposes silos, governance weaknesses, data deficiencies, and cultural resistance.
In that sense, perhaps the most troubling statistic isn’t that 56% of executives still don’t see a return on investment.
The truly concerning figure is that many continue to look to technology for the solution when the problem lies elsewhere.
Tomorrow’s competitive advantage will likely not be held by those who use AI.
It will be up to those who have learned to transform their organizations to fully realize their potential.
Cette publication est également disponible en :
