By Jacques Bughin
Artificial intelligence is reshaping corporate strategy beyond tools and timing. Jacques Bughin argues that it is a commitment to redesigning operating models and embedding learning loops that drive long-term performance. The real divide is between firms that experiment and firms that transform through sustained adaptation and continuous improvement.
The corporate conversation around artificial intelligence has entered a more cautious phase. After two years of experimentation, most organizations have deployed copilots, launched pilots, and trained employees. The technological progress is undeniable, yet the financial impact often remains ambiguous. Productivity improves in specific tasks, but firm-level performance rarely changes immediately. Many executives therefore reach a reasonable conclusion: the technology is promising but immature, and waiting until reliability stabilizes may be the prudent course of action.
The technological progress is undeniable, yet the financial impact often remains ambiguous.
This perception is reinforced by empirical evidence. A widely cited analysis conducted with MIT researchers suggests that the overwhelming majority of corporate AI initiatives — often quoted at close to 97 percent — fail to generate measurable operational impact beyond pilot stage. For many leaders, this appears to confirm that artificial intelligence still belongs to experimentation rather than transformation.
However, the interpretation is misleading. It assumes that artificial intelligence behaves like a conventional IT investment whose value should become visible shortly after implementation. In reality, AI operates less as a tool and more as a capability-formation process. Its economic effects unfold gradually as organizations reorganize around it. The disappointing early signal therefore does not indicate technological weakness but a mismatch between evaluation horizon and mechanism.
Traditional software improves efficiency inside a stable operating model. Once properly integrated, its benefits can be measured directly through cost savings or speed. Artificial intelligence differs because it alters how decisions are made, how workflows interact, and how authority is distributed between humans and systems. Before these adjustments occur, performance gains inevitably appear limited. Companies interpret this as failure when in fact they are observing the installation phase of a learning system.
Each AI deployment therefore creates a hidden strategic choice. Firms believe they are selecting models or vendors, yet they are actually determining a trajectory. Every use generates data; data improves predictions; improved predictions reshape processes; and new processes enable additional uses. When this loop stabilizes, performance begins to compound. When projects remain isolated, performance plateaus. Two companies may deploy similar technologies but diverge dramatically because one embeds learning while the other conducts experiments.
Corporate examples already illustrate this distinction clearly. Ferrari has not adopted generative AI primarily to automate engineering tasks. Instead, it has redefined the relationship between manufacturer and driver by building telemetry-driven services, personalized digital experiences, and subscription layers that transform ownership into a continuous interaction. The value arises not from faster design but from a new revenue logic supported by ongoing data feedback.
Adobe followed a similar trajectory by embedding proprietary generative models directly into the creative workflow. Rather than simply accelerating production, the firm reorganized the creation process itself, coordinating designers, software, governance, and distribution into an integrated ecosystem. Over time, usage improves models and improved models increase usage, turning software into an evolving creative infrastructure.
Amazon operates at an even deeper level, orchestrating forecasting, logistics, pricing, and cloud infrastructure through interacting AI systems. Each decision enhances subsequent decisions, producing cumulative operational learning. Competitive advantage stems less from any single algorithm than from the reinforcing interaction of processes.
By contrast, some enterprise software firms possess strong AI capabilities yet retain largely unchanged strategic positioning. They embed models within existing processes and achieve incremental efficiency improvements but not structural performance change. The difference across these cases is therefore not technological sophistication but organizational configuration.
Understanding this distinction also clarifies why waiting can increase risk. With conventional technologies, later adoption allows firms to benefit from maturity. Artificial intelligence derives part of its value from accumulated operational experience. Organizations that begin adapting early redesign workflows and governance alongside technological progress. Late adopters can acquire comparable models but cannot immediately replicate years of embedded knowledge.
Simulations of firm performance dynamics confirm this effect. Organizations that integrate learning loops initially progress slowly but then accelerate as feedback stabilizes. Firms relying on isolated pilots experience modest early gains yet quickly plateau. Over a medium-term horizon, trajectories diverge markedly: a minority of firms capture most productivity growth while others see limited improvement. Rather than convergence as technology diffuses, markets display widening dispersion resembling the emergence of superstar firms in earlier digital transformations.
This explains the paradox highlighted by the high failure rate of AI projects. Early deployments increase supervision, introduce friction, and rarely generate immediate payoff. Companies abandoning initiatives at this stage interpret the outcome as technological disappointment. Companies persisting through the adaptation phase begin to experience accelerating improvement once learning loops operate reliably.
Artificial intelligence should thus be understood less as an investment in software than as a commitment to a future operating model.
The strategic question for leaders is therefore not whether artificial intelligence has matured sufficiently to deploy safely. It is whether the organization is prepared to evolve into a system capable of learning continuously with the technology. By the time uncertainty disappears, competitive positions will largely be established because some firms will have accumulated years of adaptation while others remain at experimentation stage.
Artificial intelligence should thus be understood less as an investment in software than as a commitment to a future operating model. Technologies can always be acquired later. Trajectories cannot.
Over the coming decade, competition will not separate companies that use AI from those that do not. It will separate organizations that chose early to become learning systems from those that waited for certainty.
