Can large language models assess enterprise digital transformation?

Enterprise digital transformation reshapes business models and operations through digital technologies, making scientific measurement essential for empirical research. This study proposes a large language model (LLM)-based multi-indicator and multi-level text measurement framework for identifying and classifying enterprise digital technology applications from two dimensions: digitalization and effectiveness. We further adopt a three-dimensional experimental design covering output modes, prompting strategies, and fine-tuning to evaluate model performance. The results show that LLMs generally outperform traditional machine learning (ML) models, even under Zero-shot settings. Synchronous multi-indicator output generally performs better than one-by-one output, whereas the effects of chain-of-thought (CoT) prompting and fine-tuning vary across models and tasks. Model performance varies across specific models and task settings, with both recent general-purpose models and reasoning-oriented models showing competitive results in different scenarios. Further analysis indicates that LLMs perform better on explicit digital technology indicators but show weaker performance on digital effectiveness indicators requiring contextual inference. Additional validation with non-Chinese multilingual LLMs and significance analysis supports the robustness of the main findings. This study provides a methodological tool and empirical evidence for measuring enterprise digital transformation using corporate disclosure texts.