Determinants of Machine Learning Project Success: A Structural Equation Modeling Study of the Machine Learning Canvas

Machine learning (ML) project outcomes emerge from the interplay of organizational and technical subsystems. A recent study of 65 experienced data scientists reported that more than 80% of AI projects failed, roughly twice the failure rate of traditional IT projects, even as AI coding assistants can accelerate implementation work. This study develops and empirically evaluates the Machine Learning Canvas (MLC), a project-level sociotechnical framework that integrates business strategy, software engineering, and data science in four interdependent dimensions: Strategy, Process, Ecosystem, and Support. Its novelty lies in connecting these organizational and technical dimensions in one measurement and structural model rather than treating business alignment, workflow, and infrastructure as separate concerns. Seven theory-derived hypotheses were tested with covariance-based structural equation modeling using survey data from 150 respondents who reported daily use of AI coding assistants. The results support a sequential association from Support through Strategy and Process to Ecosystem; Strategy and Ecosystem also show positive direct associations with perceived project success, whereas the Process–Success coefficient is negative after the other dimensions are controlled. The overall model fit is good. The findings position ML project success as a property of the sociotechnical system rather than coding productivity alone and present the MLC as a diagnostic and planning instrument. Because the outcome is perceived success and the sample is narrow, the findings do not establish causal effects or universal applicability.