德勤中国TMT行业主管合伙人程中：企业的AI化已成趋势，重构过程要经历四个阶段

Group 1 - The core viewpoint emphasizes the imbalance between value extraction and risk management in the context of generative AI, highlighting the need for companies to act quickly to harness AI's potential for sustainable business value [1] - Deloitte China suggests that effective generative AI governance requires clear responsibilities, enhanced skills, and a comprehensive risk management process throughout the AI lifecycle [1][2] - Companies are shifting from questioning whether to adopt generative AI to focusing on how to implement it effectively, with a critical need to prioritize investments in verifiable return cases [1][2] Group 2 - The AI transformation process in enterprises typically involves four stages: establishing an AI strategic vision, pilot exploration, deep integration into core business processes, and financial mapping [2] - During the pilot exploration phase, companies face challenges in quantifying value and ensuring collaboration between IT and business departments [2][3] - The final financial mapping stage connects AI investments directly to financial metrics, although companies still encounter difficulties in quantifying indirect benefits like customer satisfaction [2] Group 3 - A "light architecture" approach is recommended, which encapsulates AI capabilities as API services to reduce the burden of core system reconstruction [3] - Companies should maintain flexibility in technology selection and establish agreements with technology vendors to mitigate costs associated with potential technology shifts [3] Group 4 - The concept of "illusion" in AI outputs, which can mislead business decisions, is identified as a significant risk, necessitating the establishment of a "trustworthy AI framework" to implement multi-layered defenses [4][5] - Structural illusions, which manifest in seemingly accurate outputs that are actually based on flawed data, should be prioritized for resolution due to their high risk [5] Group 5 - The board of directors is advised to redefine the value boundaries of "human-machine" collaboration, positioning AI as an "enhancement tool" rather than a replacement [6] - In media applications, AI should initially be deployed in low-risk scenarios, with ongoing training and feedback mechanisms to improve its reliability [6] Group 6 - Effective AI governance should transition from a passive to an active approach, involving clear strategic goals, dedicated governance teams, and the use of explainable AI technologies [7] - Companies are encouraged to integrate AI into their core operations rather than treating it as an add-on, which can enhance revenue through improved customer satisfaction and compliance [8]