Without a formal "Financial Domain Ontology", agentic systems cannot reliably interpret complex relationships across products, rules, data lineage, and regulatory obligations — resulting in compliance risks, operational failures, and stalled digital transformation. Hong Kong following interconnected barriers when deploying agentic AI: 1. Fragmented Data & Semantic Gaps Core banking, trade platforms, KYC/AML repositories, investment guidelines, and historical memos live in silos. Agents cannot understand that “Rule 18f” treatment differs by product type, booking location (e.g., Hong Kong vs Beijing), or counterparty — leading to incorrect transaction approvals or missed breaches. 2. Hallucination & Lack of Explainability Generative models produce plausible but ungrounded answers. In a regulated environment this creates unacceptable risk: regulators demand full traceability of every decision, which pure LLM-based agents cannot deliver. 3. Regulatory & Compliance Pressure HKMA AI 2 Strategy and GenAI Sandbox++ require robust risk management, bias detection, and human oversight. - PDPO demands strict data privacy and purpose limitation. - Cross-border activities (e.g., futures trading) add strict firewall and data leakage controls. 4. Technical Debt & Legacy Integration Years of accumulated technical debt and point-to-point integrations make it extremely difficult for agents to access reliable, real-time data. Manual reconciliation and exception handling still dominate, consuming 50%+ of operations effort. 5. Talent & Delivery Gap Most teams lack the rare combination of deep financial domain knowledge, ontology engineering expertise, and proven large-scale delivery experience needed to industrialise agentic AI safely. Real-World Impact Institutions waste 6–18 months and significant budgets on pilots that never reach production, while competitors who solve the ontology layer achieve 40–60% efficiency gains and faster regulatory approval.