Agentic AI rebooking orchestration for ‘PNR Not Generated’ failures across CRM, booking and payment systems
An AI agent that resolves failed-booking cases end to end for a large airline’s payment desk - orchestrating CRM, booking engine and payment gateway from one conversational interface, cutting resolution from 15-30 min to under 5.
Business Challenge
The airline’s Customer Payment Desk handles 700+ ‘PNR Not Generated’ cases a day - customers who paid but received no booking. Agents resolved each case by manually toggling between three disconnected systems: CRM (case details), the booking engine (flights, seats, passengers) and the payment gateway (transactions, payment links). Resolution took 15-30 minutes per case; agents managed only 20-25 cases a day Seats were often lost while agents switched systems Manual re-entry caused errors (~15% error rate): wrong passenger details, missed senior-citizen discounts, infants allocated to the wrong adult No unified view or audit trail across systems, and poor CSAT from long waits
Solution Delivered
Built an AI agent on DevRev Computer that orchestrates the full rebooking flow from a single conversational interface for payment-desk agents. Architecture: API integrations with CRM (case retrieval/update), booking engine (flight search, seat sell, passenger, infant, contact and PNR endpoints) and payment gateway (transaction verification, payment-link generation) Workflow orchestration layer with static-IP HTTP nodes for allow-listed calls; agent uses a planner LLM, skill/tool repository and short-term session memory Input/output guardrails, RBAC and field-level access control, SSO, full audit trail written back to the CRM Flow: agent pulls case context and order ID -> verifies payment status -> searches flights and reserves seats -> enrolls passengers in a loop (multi-passenger, infant-to-adult allocation, automatic senior-citizen fare code) -> adds contact details -> generates and sends payment link -> retrieves PNR post-payment and closes the case. Built error handling and retries, and validated 4 passenger scenarios (2 adults; adults + infant; senior citizen; 4 adults + infant + 2 seniors) with human confirmation at key steps.
Outcomes Achieved
Case resolution time cut from 15-30 minutes to under 5 minutes (~80% reduction) Zero manual system switching: 3 systems per case reduced to 1 conversational interface All 4 passenger scenarios validated end to end, including infant allocation and senior-citizen discount logic Automated payment-link generation and post-payment PNR retrieval Complete audit trail in the CRM for every action (PNR regeneration, booking, payment link, case status) Projected agent productivity increase from 20-25 to 150-180 cases/day Faster processing improves seat-availability capture and CSAT Lesson learned: the model was the easy part - the value came from hardening multi-passenger edge cases, session state across systems and enterprise security (SSO, static IPs, RBAC)
Measurable Business Outcome
Reduced ‘PNR Not Generated’ resolution time from 15-30 minutes to under 5 minutes per case (~80% reduction) across a 700+ cases/day queue; eliminated manual switching across 3 systems; validated infant-allocation and senior-citizen discount logic across 4 passenger scenarios; projected agent throughput from 20-25 to 150-180 cases/day.