Stanford research team finds why corporate AI projects succeed or fail

The development of artificial intelligence tools within companies has seen significant investment, with AI and machine learning job postings increasing 70-80% in early 2024 compared to 2023. A Stanford research team spent four years studying AI project implementation at a multinational fashion company to understand why some initiatives succeed while others fail. Full paper “Eliciting Domain Expertise in the Absence of Formal Authority:
The Case of AI Developers and Domain Experts in a Large Firm.

Research methodology and context: Stanford researchers conducted an extensive study between October 2019 and December 2023, closely observing software developers’ interactions and project outcomes at a major fashion company.

• The research team followed multiple AI-based projects, ultimately focusing on two contrasting cases that highlighted success and failure factors
• The study revealed three critical variables that significantly influence project outcomes: jurisdictional clarity, task centrality, and task enactment
• Both projects were executed by the same team, eliminating team composition or learning curve as potential variables

Key success factors: The research identified clear patterns distinguishing successful AI implementations from failed ones.

• Successful projects had well-defined groups of stakeholders with clear reporting structures and decision-making authority
• Projects targeting core job responsibilities received better cooperation from domain experts
• Uniform task execution across different locations or teams increased the likelihood of successful AI implementation

Case study comparison: Two specific projects illustrated the stark contrast in outcomes based on these key variables.

• A successful supply chain distribution AI tool benefited from having a small, cohesive group of allocation specialists who shared common goals and processes
• A failed retail productivity optimization project struggled due to diverse stakeholder groups, varying store conditions, and differing management priorities across 200 locations
• The failed project revealed the challenges of developing standardized AI solutions for heterogeneous business environments

Management implications: The research yielded specific recommendations for improving AI project success rates.

• Executive leadership must formally mandate AI initiatives and establish clear guidelines for cooperation between developers and domain experts
• Companies should consider creating liaison positions to facilitate communication between AI developers and business units
• Project scope should be carefully managed, with struggling initiatives being scaled down or refocused on specific problem areas

Developer considerations: The study revealed important insights for AI developers approaching new projects.

• Developers should assess jurisdictional clarity and task centrality before committing to project scope
• Success requires understanding both technical requirements and organizational sociology
• Early evaluation of task homogeneity across different locations or teams can help predict project viability

Looking ahead: While AI development teams often receive substantial resources and autonomy, the research suggests that successful implementation depends more on organizational structure and stakeholder alignment than technical capabilities alone. Future corporate AI initiatives may benefit from more targeted, smaller-scale approaches that consider these organizational dynamics from the outset.