Payments data is fragmented due to inconsistent terminology across processors. This structural issue, compounded by volume and token limits, makes it difficult for current LLMs to reliably process raw, mismatched data at scale without specific business context. As AI adoption for operational and agentic workflows increases, the need for verified, reliable payments data is critical. The gap between asking a payments question and receiving a trustworthy AI-driven answer is an escalating operational cost.
Key Points:
- The Payments Data Fragmentation Problem: Raw payments data is inconsistent and lacks standardized terminology across major processors. This fragmentation creates a structural barrier that current Large Language Models (LLMs) cannot overcome at scale, leading to unreliable AI output for critical operational workflows.
- The Need for a Semantic Layer: To move beyond “mostly correct” AI results and enable trustworthy, agentic workflows, a specialized semantic layer is required. This layer must provide a structured taxonomy, logical constraints, and continuous updates based on deep payments domain expertise to normalize and give real-world meaning to the ambiguous, complex data.
- Why Human Expertise Still Trumps AI Hype: Automated entity resolution and ontology generation are complex problems that remain unsolved by simple LLM application. Payments success requires human domain experts to account for hidden “gotchas” (like cost report duplication or cross-PSP transaction matching) and assess the risk of ambiguous data, providing the foundation an AI can actually trust.
Presenter
Name: Andy Barker and Dr. Robert Clewley
Title: Senior Product Lead (both)
Company: Pagos