本研究旨在檢驗 AI 技術導入對企業績效之影響，並以台灣高科技電子產業為研究對象。本文依企業公開揭露資訊辨識 AI 技術首次導入年度，建構 AI 技術導入後虛擬變數，並結合台灣經濟新報資料庫之財務資料，建立 2010 年至 2024 年、364 家公司、5，460 筆公司—年度觀察值之平衡長期追蹤資料。研究方法採用雙向固定效果模型，並以導入後第 1、3、5、7 與 9 年之績效表現作為觀察期間，同時進行導入前趨勢檢驗與產業固定效果穩健性檢驗。
實證結果顯示，AI 技術導入對企業績效之影響並非立即且全面正向，而較接近短期成本壓力與中長期部分改善之型態。獲利能力方面，毛利率與營益率皆呈現導入初期受壓、後期回升之趨勢；營運效率方面，營收佔資產總額比例於中期出現較明顯改善，但營收成長率未呈現穩健結果；報酬率方面，ROA 於導入初期顯著下降，ROE 則未呈現一致效果；費用結構方面，用人費用率於導入初期上升較為一致。整體而言，本文發現 AI 導入效果具有時間遞延特性，企業可能需先承擔導入成本與組織調整壓力，後續效益才逐步反映於部分績效指標。;This study examines the impact of artificial intelligence (AI) technology adoption on firm performance， using Taiwan’s high-tech electronics industry as the research setting. AI adoption is identified from firms’ public disclosures， and a post-adoption indicator is constructed based on the first year in which AI technology is observed to enter production， management， research and development， or operational processes. The financial data are obtained from the Taiwan Economic Journal database. The final sample consists of a balanced panel of 364 firms and 5，460 firm-year observations from 2010 to 2024.
This study employs two-way fixed effects models and observes firm performance at one， three， five， seven， and nine years after AI adoption. Pre-adoption trend tests and industry fixed effects robustness checks are also conducted. The empirical results suggest that AI adoption does not immediately and comprehensively improve firm performance. Instead， its effect is more consistent with a pattern of short-term adjustment costs and partial medium- to long-term improvement. Profitability indicators， including gross margin and operating margin， show short-term pressure followed by later recovery. Asset efficiency improves more clearly in the medium term， while sales growth does not show robust evidence. ROA declines significantly in the short term， whereas ROE does not present consistent results. In terms of cost structure， labour cost ratio increases more consistently in the early stage of adoption. Overall， the findings indicate that the effects of AI adoption are likely to be delayed， as firms may first bear implementation costs and organisational adjustment pressure before performance gains gradually appear in selected indicators.
