本研究以2019年至2024年台灣上市公司為樣本，透過年報文字探勘建構 TF-IDF 加權之AI導入程度指標，實證檢驗企業導入AI對營運效率及市場價值 之影響，並進一步探討先進製造業與金融業之產業調節效果。 實證結果顯示：第一，AI導入程度對營運效率具有顯著正向影響，表明AI 技術之應用能有效提升企業之投入產出效率。第二，AI導入程度對市場價值亦 呈顯著正向關係，顯示資本市場對企業之AI投入給予正面之估值溢價。第三， AI 導入對營運效率之正向影響在不同產業間並無顯著差異，暗示AI對營運效 率之提升具有跨產業之普適性。第四，AI導入對市場價值之正向影響在先進製 造業與金融業中顯著更強，反映資本市場對技術密集型產業之AI投入賦予更高 之估值溢價。 本研究另以ChatGPT問世為分界進行結構性變革測試，發現生成式AI之發 布顯著強化了資本市場對企業AI投入之正面評價。此外，透過二階段最小平方 法之內生性檢定、DEA模型替換、AI導入程度替代衡量、AI 導入企業子樣本 穩健性檢定以及排除金融業樣本等六項敏感性分析，整體結果與主測試方向一 致，確認研究結論之穩健性。本研究之發現為台灣企業之AI策略制定、投資人 之價值評估以及政府之AI產業政策提供實證參考依據。;This study uses Taiwan-listed companies from 2019 to 2024 as the sample to construct a TF-IDF weighted Artificial Intelligence (AI) adoption index through text mining of annual reports. It empirically examines the impact of corporate AI adoption on operational efficiency and market value， while further exploring the moderating effects within the advanced manufacturing and financial industries. The empirical results indicate: First， the degree of AI adoption has a significant positive impact on operational efficiency， demonstrating that the application of AI technology effectively enhances corporate input-output efficiency. Second， AI adoption exhibits a significant positive relationship with market value， suggesting that the capital market assigns a valuation premium to corporate AI investments. Third， the positive impact of AI on operational efficiency does not vary significantly across different industries， implying a cross-industry universality in AI-driven efficiency gains. Fourth， the positive effect of AI on market value is notably stronger in the advanced manufacturing and financial sectors， reflecting the market’s higher valuation of AI investments in technology-intensive industries. Furthermore， this study conducts a structural break test using the launch of ChatGPT as a demarcation point， finding that the release of generative AI significantly strengthened the capital market′s positive evaluation of corporate AI adoption. In addition， six sensitivity analyses—including Two-Stage Least Squares (2SLS) endogeneity tests， alternative DEA models， proxy measures for AI adoption， the exclusion of financial sector samples， and an AI-adopting firm subsample analysis—yielded results consistent with the main tests， confirming the robustness of the findings. These results provide empirical references for AI strategy formulation， investor valuation， and government industrial policies in Taiwan.
