Measuring Data-Driven Innovation in Firms: A Multi-Source Approach Using Large Language Models

This study presents a comprehensive framework for identifying and measuring Data-Driven Innovation (DDI) at the firm level. We develop a conceptual definition of DDI grounded in the chain-linked model of innovation, where data infrastructure and advanced analytics augment each stage of the innovation process. Because DDI is intangible, embedded, complex, and heterogeneous, traditional innovation metrics fail to capture it. We therefore propose a multi-source measurement strategy exploiting Large Language Models (LLMs) applied to four distinct textual data sources: corporate communications, patents, company websites, and startup descriptions. The identification strategy relies on detecting the co-presence of data-driven means (i.e., technologies, infrastructure, algorithms) and innovative purpose (creating, improving, or optimizing products, processes, or services) within the same textual unit. Applying this framework to datasets covering listed firms in Italy (102 firms) and the US (3,714 firms), 3.4 million USPTO patent filings, 134,208 Italian SME websites, and 9,705 Crunchbase startup profiles, we document the diffusion, intensity, and heterogeneity of DDI across firm size, age, industries, and countries.