How to unlock the full potential of AI?

The latest global survey of over 1,200 executives reveals that only 5% of organizations are currently extracting significant value from their AI investments, while 60% struggle to translate their spending into measurable results. These early adopters are already enjoying 1,7x higher revenue growth and 1,6x higher EBIT margins than laggards, illustrating the widening gap between future-ready companies and those that follow.

The phenomenon is cumulative: the gains generated are systematically reinvested in new capabilities, further accelerating competitive advantage. The most advanced companies plan to increase their IT budgets by 26% this year and to devote nearly two-thirds of these additional budgets to AI. At the same time, companies in the emerging or stagnating phase report modest profits at best; some even admit to having obtained no concrete returns despite dozens of pilot use cases.

Functional analysis shows that 70% of the potential value is concentrated in the core business: sales, marketing, supply chain, production, R&D. Projects affecting customer relations and industrial planning generate the most visible benefits, far ahead of support functions. However, IT is doing well: its share of the value created jumped to 13% in one year, stimulated by the demand for modular data architectures and accelerated development tools.

The new wave of autonomous agents, combining predictive and generative capabilities, is changing the game. Already credited with 17% of the value generated in 2025, they could exceed 29% by 2028. These “digital workers” observe, reason, plan, and act within entire workflows, under the supervision of human teams that are moving from an execution role to one of orchestration and ethical guarantee.

Sectors are progressing at very different rates. Software, telecoms, and fintech dominate, driven by strong technological maturity and near-universal access to generative AI tools. Conversely, chemicals, fashion, and construction still lack structured data and show limited adoption rates. Geographically, Asia-Pacific already devotes more than 5% of its IT budgets to AI, ahead of North America and Europe, and is counting on faster cost reductions thanks to automation.

The key to success lies less in the multiplication of experiments than in a multi-year strategic ambition, steered at the highest level. Leading companies translate business challenges into funded roadmaps, sequenced, and tracked by value indicators (incremental revenue, cost reduction, time-to-market). They practice co-ownership governance: business and IT share responsibility for decisions, avoiding the "silo" effect and the accumulation of orphan prototypes.

Operationally, the priority is no longer incremental automation, but the radical overhaul of a few high-ROI journeys. Restructuring, for example, the quoting process or the end-to-end management of a production line unlocks gains greater than the sum of one-off optimizations. A cosmetics brand launched a virtual beauty assistant deployed in twenty markets; the hyper-personalized recommendations doubled the return on investment of its e-commerce journeys and generated more than $100 million in additional sales.

This reinvention requires a flexible technical architecture: a central agent orchestration platform, shared models and prompts, and standardized connectors to existing systems. Organizations that have internalized an "AI platform factory" cut their production time in half by capitalizing on reusable components. Conversely, heterogeneous environments, the result of opportunistic purchases of vertical solutions, see maintenance costs and security risks explode.

Responsible governance becomes critical as agents gain autonomy. Leaders define clear safeguards: ethical charters, bias checks, robustness tests, audit logs. This rigor does not hinder the speed of execution; on the contrary, it removes regulatory obstacles and reassures businesses about the risks of deviation or error. Today, nearly three-quarters of companies still recognize unresolved security vulnerabilities; those that correct them early accelerate deployment without compromising trust.

The human aspect is just as decisive. Mature organizations are aiming to upskill more than half of their workforce by next year, are setting aside time for learning, and are involving employees in co-designing solutions. In these hybrid environments, the value of teams is shifting toward contextual expertise, oversight, and creativity. Managers are becoming conductors of agent portfolios, capable of quickly composing new workflows based on business objectives.

Budgeting is also evolving. Instead of projecting returns limited to labor savings, finance departments are now incorporating the potential for innovation, new business models stemming from AI-native products or services, and data monetization. One industrial conglomerate, for example, valued its future EBIT at over $300 million thanks to a library of agents tasked with planning production and identifying defects.

The external ecosystem is becoming an essential lever for quickly capturing missing technological building blocks. Partnerships with hyperscalers, low-code platform publishers, and specialized startups lower entry costs and provide access to best practices. Companies that orchestrate these alliances are three times more likely to already have agents in production and to reuse models and prompts from one use case to another.

Faced with this urgency, a simple rule must be applied: 70% of efforts must focus on people and processes, 20% on technological architecture, and only 10% on models and algorithms. Most of the obstacles identified relate to culture, change management, and clarifying the value proposition. Without clear strategic alignment, the accumulation of POCs results in a "GenAI" debt that is difficult to resolve.

For organizations lagging behind, a three-step acceleration plan is emerging. First, engage senior management in a value-driven transformation mandate, supported by publicly available financial and non-financial metrics. Second, select a handful of critical operational flows to reinvent end-to-end, integrating governance and impact measurement from the outset. Third, industrialize a shared agent platform, leveraging maximum reuse and open interoperability standards to avoid vendor lock-in.

Every month saved counts: exponential advances in models make catching up increasingly costly. Companies that delay risk finding themselves stuck in a vicious circle—limited budgets, fleeing talent, outdated architecture—while leaders will have already locked in their positions. Conversely, those that adopt a structured approach now, focused on value creation and redesigning operating methods, can still close the gap and, tomorrow, join the select club of true value generators.

To go further: explore why AI culture is becoming a driver of competitiveness, Why The adoption of AI depends first and foremost on humans. and how create momentum for AI adoption.