Pranav Mulgund

Enterprises have spent two decades and millions of dollars on ERP, procurement suites, and RPA – yet roughly 20% of invoices still fall into exceptions that land on human desks in shared services. Blocked invoices, failed 3-way matches, missing POs, and credit memo disputes remain stubbornly manual because resolving them requires something software has never had: your company's tribal knowledge, spread across siloed systems.

The conventional answer is to stabilize processes and rebuild data foundations before deploying AI. This session makes the case for a different path: agents that deploy onto enterprise systems exactly as they are – messy data, siloed sources, undocumented tribal knowledge included – and start resolving exceptions end-to-end in days. Drawing on live deployments at Fortune 500 manufacturers, we'll walk through what production-grade agentic AI looks like in exception-heavy source-to-pay work, and how shared services leaders should evaluate it for their own organization.

Key takeaways:

• Why exceptions persist despite ERP, P2P suites, and RPA – and why the last 20% of invoices consumes a disproportionate share of shared services capacity
• What separates agents that autonomously resolve exceptions end-to-end from copilots and workflow tools that still route work back to humans
• Why agentic AI doesn't have to wait on a multi-year data cleanup: how agents deploy onto existing systems (SAP, Ariba, Snowflake) in hours, with no rip-and-replace and no data migration
• Real results from Fortune 500 deployments