Companies often invest heavily in artificial intelligence hoping for tangible business value, yet many initiatives stall before delivering results. According to recent surveys, 42% of companies abandoned the majority of their AI initiatives in 2025, up from 17% in 2024, and on average, 46% of proof-of-concepts were scrapped before reaching production. This disconnect frequently stems from a misalignment between ambitious innovation goals and a company’s underlying value chains, operating models, and technology stacks rather than limitations in the technology itself. S&P Global Market Intelligence found that only one-third of organizations achieve significant ROI from their AI investments, even though 73% spend more than $1 million annually on the technology.
Two Key Dimensions for Success
Research indicates that AI success depends on where a company falls along two key dimensions: value-chain control and technological breadth. Value-chain control refers to the degree of influence a firm has over the journey from idea to market. Companies with high control can test, iterate, and scale innovations quickly because they own or strongly influence product design, manufacturing, distribution, and customer engagement.
Samsung, for example, can roll out AI-powered display or camera improvements across its entire product portfolio because it controls everything from chip fabrication to global retail outlets. At the other end of the spectrum are companies with low control, such as tier-two suppliers in the automotive sector or brand licensors, which must rely on others to validate or distribute innovations.
The second dimension, technological breadth, refers to the range and interdependence of the technologies a company must integrate to compete. High-breadth sectors, such as semiconductors, autonomous vehicles, and life sciences, require AI to be woven into a fast-moving web of other technologies like sensors, robotics, materials science, and cloud architecture.
Low-breadth industries, such as food processing, building materials, and basic logistics, tend to operate with more stable technology stacks, where AI is used to refine existing processes rather than redefine the environment. These dimensions are dynamic forces that evolve across functions, geographies, and time, meaning a company may have high technological breadth in R&D but low breadth in customer engagement, or exert strong value-chain control in one region while depending heavily on intermediaries in another.
It is common for organizations to assume that a single, overarching AI strategy will apply uniformly across all functions. However, the reality is often more fragmented, with different parts of a company operating in different quadrants of this framework. A global consumer goods company might apply focused differentiation strategies in its supply chain while simultaneously engaging in platform leadership for its digital products. This complexity requires leaders to recognize that strategy may begin in one quadrant, but success is built through a system that adapts to the specific constraints and opportunities of each area.
Four Strategies for Realizing AI Potential
Four Strategic Approaches for Companies
Four distinct approaches emerge from this framework, each suited to a specific organizational position. The first, focused differentiation, applies to companies with limited value-chain control and low technological breadth. These firms operate in mature industries and use AI to fine-tune and optimize products or processes within a defined domain rather than redesigning the system. The global spice manufacturer McCormick & Company narrowed its focus to flavor development.
In 2019, the company partnered with IBM to build SAGE, an AI system trained on decades of sensory data, recipes, and consumer insights. The tool has since become central to McCormick’s product development process, helping the company accelerate innovation. The chief risk for companies in this quadrant is an excess of ambition; Zillow’s home-flipping initiative, which relied on its AI-derived pricing model, spectacularly failed to scale, resulting in a $304 million inventory write-down and the cancellation of the entire Zillow Offers business.
The second strategy, vertical integration, suits companies with strong value-chain control but relatively limited technological breadth. These organizations embed AI into the processes they already own, linking insights across internal systems to reveal synergies and efficiencies. JD.com, the Chinese e-commerce giant, embedded AI across its logistics network, using real-time data to optimize warehouse inventory, delivery routing, and labor scheduling. During the pandemic lockdowns, JD.com’s intelligent system rerouted deliveries based on containment zones and dynamically reassigned inventory to match regional surges in demand, maintaining uninterrupted service while competitors struggled.
In the energy sector, ExxonMobil used AI to interpret seismic data and optimize drilling paths in Guyana, cutting average well-drilling time by 15% and saving millions per site. GE, however, sought to become the Microsoft of industrial AI with its Predix platform but scaled back its ambitions after spending more than $4 billion due to siloed data, internal resistance, and shifting leadership.
Thriving in Complex Ecosystems
The third approach, collaborative ecosystem, applies to companies operating in technologically complex ecosystems but lacking control over how solutions reach the market. Success here comes from partnering strategically rather than going it alone. Novartis and Microsoft created an AI innovation lab aimed at accelerating drug discovery, with tools that helped identify new biomarker combinations for oncology trials, cutting trial design time by more than 30%. BMW Group’s alliance with Intel and Mobileye contributed distinct capabilities—processing power, computer vision, and vehicle integration—to develop autonomous driving solutions.
Pfizer’s collaboration with BioNTech during the Covid-19 pandemic saw BioNTech’s AI models screen more than 10,000 mRNA candidates in days, selecting the formulation that became the vaccine, while Pfizer’s global regulatory and manufacturing capabilities accelerated production. IBM’s high-profile collaboration with the cancer center MD Anderson aimed to transform cancer care with its Watson-powered Oncology Expert Advisor, but the project struggled with organizational and integration challenges and never moved beyond the pilot phase.
The final strategy, platform leadership, applies to companies at the apex of both dimensions—high technological breadth and broad value-chain control. These organizations create infrastructure and ecosystems as well as build products, focusing on setting standards and opening APIs. Bloomberg’s launch of BloombergGPT, a finance-specific large language model trained on more than 700 billion tokens, was a strategic move to define the next generation of financial AI. Siemens Healthineers has achieved a similar position in medical imaging with its AI-Rad Companion suite, which integrates directly with hospital systems to analyze X-rays, CT scans, and MRIs.
Microsoft’s platform approach combines infrastructure, tooling, and ecosystem orchestration, with GitHub Copilot contributing up to 40% of code written in supported languages and Azure OpenAI Service serving as the enterprise backbone for generative AI. Google’s foray into healthcare AI through its DeepMind Health, however, faced setbacks when the team accessed millions of NHS records without proper consent, leading to public backlash and the absorption of the initiative into Google Health.
The Human Factor in AI Adoption
The Human Factor in AI Adoption
A survey of 1,600 enterprise leaders and employees by the AI firm Writer found that 31% of employees admitted to actively pushing back on their company’s AI initiatives—often because they feared being replaced. One in 10 went even further, saying they had tampered with performance metrics or intentionally generated low-quality outputs to undermine adoption efforts. When Rent a Mac, an Apple device rental company, launched an AI-driven inventory management system, it triggered anxiety across its workforce, leading to a seven-week delay in implementation and a loss of about $85,000 in expected efficiency savings.
However, by appointing AI champions to demonstrate real use cases, the company saw engagement levels triple from 31% to 89% in just a few months. Colgate-Palmolive recognized the importance of employee engagement when it launched its internal AI Hub that empowered employees to develop their own assistants, thousands of them, without coding experience, resulting in better workflows and buy-in.
The role of the manager is shifting in AI-powered organizations. Beyond coordinating people, managers must help teams learn to collaborate with algorithms, interpreting machine insights, redesigning workflows, and translating technical progress into human progress. This often requires a cultural shift: creating space to experiment, to fail fast, and to learn in real time. The most successful organizations treat AI not as an answer but as a question: How can we work smarter, together? This approach ensures that AI is used to enhance human capabilities rather than replace them.
