Finding green start-ups at scale: Automating selection through Natural Language Processing algorithms

Research and policy interest in green entrepreneurship is growing, but identifying green start-ups remains a major challenge because the concept of “green” is inconsistently defined across studies, and existing identification strategies often rely on costly, non-scalable, or discretionary procedures. This paper develops an automated procedure to identify start-ups with an explicit environmental orientation using Natural Language Processing (NLP) with minimal human intervention. To ground the approach conceptually, we derive a green classification framework from Sustainable Development Goal (SDG) targets specifically related to environmental sustainability and use it to map start-up activities into a common set of green topics. We then apply the framework to the websites of more than 10,000 Italian innovative start-ups founded between 2009 and 2023, implementing three complementary NLP pipelines based on an ML dictionary, LDA, and BERTopic approach, to detect firms with environmentally oriented products, services, or processes. The results show that automated text-based identification can capture a broad and policy-relevant landscape of green start-ups while preserving thematic heterogeneity across environmental domains. To illustrate the analytical value of the framework, we compare the resulting distribution of these green start-ups across topics with Italy's National Recovery and Resilience Plan investments and with SDG performance indicators. This application reveals both areas of alignment and mismatch between entrepreneurial emergence, policy priorities, and environmental sustainability challenges. By offering a transparent, scalable, and SDG-grounded procedure for identifying green start-ups, the paper contributes to the measurement of green entrepreneurship and provides a useful tool for technological forecasting and policy planning.