Data Architecture, Machine Learning and Firm Productivity

As enterprise IT systems increasingly incorporate machine learning technology, it is crucial to understand complementary organizational practices that allow firms to unleash productivity benefits from adoption. This study examines the relationship between machine learning adoption and data architecture capabilities using detailed corporation-level data. We utilize a survey to measure two clusters of data architecture practices--data fabric and ML infrastructure, for 82 large corporations. We use data on legacy systems and software varieties to produce out-of-sample predictions of data architecture capabilities for an additional 143 large corporations on the Fortune 500 list. We find that corporations with higher data fabric capability adopt machine learning more intensively. A one-standard-deviation increase in machine learning adoption is associated with a 1.5% increase in productivity among corporations with above-median data fabric capability. However, machine learning adoption can negatively affect firm productivity without a sufficiently developed data fabric. These results suggest that data fabric capability is complementary to machine learning adoption. Data fabric facilitates more coordinated and less fragmented adoption of ML software products across the organization because it allows organizations to integrate large data streams across diverse sources and locations.