Building an Agentic Collections Workflow
Designed an AI-assisted collections flow with customer escalation and compliance guardrails.
- Agentic AI
- Collections
- Compliance
- Customer operations
Siddharth Mishra
Lead Product Manager at Zeta
From their time as
Senior Product Manager
PayPal • 2025 - 2025
Overview
Siddharth designed an agentic AI workflow for PayPal collections. It redirected sensitive cases to live agents and reduced operations costs by 25%.
The Story
Siddharth worked on PayPal's collections charter, where operations agents handled delinquent-customer calls in a tightly regulated environment. The brief was to improve customer handling while using agent capacity more selectively.
He reviewed agent SOPs, listened to live customer calls, and mapped the use cases driving inbound contact. He then defined what the AI workflow could handle and where a live agent had to take over.
The product included a customer-controlled route to a live agent at each step. It also added guardrails for cases such as bankruptcy and possible hallucinations, directing those conversations to trained human agents.
The resulting agentic workflow reduced overall operations costs by 25% while preserving escalation paths for regulated collections cases.
Ownership Snapshot
Broad role
Senior Product Manager for Pay Later and digital collections.
Goal
Improve recovery operations while protecting customers and meeting compliance requirements.
Direct ownership
Defined use cases, requirements, escalation paths, and product guardrails.
Team execution
Worked with operations, legal, and technology teams to deliver the workflow.