Why grand unified data architecture kills AI projects before they start

Most organizations over-invest in model sophistication and under-invest in decision architecture. Anusha Dandapani, Chief Data & AI at UNICC, builds AI systems where a wrong output can mean a family loses humanitarian aid. She shares why she would rather deploy a less sophisticated model with rigorous decision architecture, how her team built a minimum wide interoperability layer across UN agencies, and why data lineage is the investment nobody wants to fund. Chapters 00:00 | Introduction 02:30 | Asymmetry of consequences in AI 05:30 | Multilingual AI safety evaluation 08:00 | Minimum wide interoperability layer 14:00 | Responsible AI from design phase 20:00 | AI as capability, not a project 24:30 | Data lineage as foundation 31:00 | Accountability in AI decisions 36:30 | Decision architecture over models 38:30 | Moving decisively but reversibly About the guest: Anusha Dandapani is Chief Data & AI at UNICC (United Nations International Computing Centre), where she leads the AI Hub. She brings roughly 20 years of experience in data science and AI, including serving as VP Data Science Lead at Barclays. She is an adjunct professor at NYU and Fordham University. Anusha was featured in the Observer's AI Power Index 2025 and has been named to the Global Data Power Women List from 2022 to 2024.