Agentic AI for In-Flight Services Complaint Resolution

Project Overview

This customer is a leading Indian airline operating a high-volume, multi-base In-Flight Services (IFS) organization spanning Customer Experience (CX), Customer Relations (CR), and Base Operations teams across metro and regional airports.

Customer complaints — relating to IFS, i.e. staff behavior, policy and procedure, or operational error — arrived through email, call center logs, social media, and weekly consolidated exports from the airline's CRM. Resolving them required CR staff to manually read unstructured email threads, cross-reference flight and crew information across multiple internal systems, populate a standardized investigation grid, route it to the relevant base location, and consolidate the response back to CX — all before a 48-hour turnaround-time target that was often missed.

Minfy Technologies designed and deployed a two-agent Agentic AI system, built natively on Google Cloud Platform, to automate this email-driven investigation and response workflow end-to-end while keeping the airline's existing business process and organizational structure intact. The engagement progressed from an initial proof of concept through hardening, UAT, and a sandboxed pilot deployment into the customer's own GCP environment.

Customer Challenges

• High-Effort Manual Email Interpretation: CR staff manually read and interpreted lengthy, unstructured complaint emails to extract complaint context, flight details, and customer information before any investigation could begin.
• ‍Fragmented, Multi-System Investigation: Identifying the responsible base location, operating crew, and flight context required manually querying multiple internal systems (crew rosters, flight/operations records) with no unified view.
• ‍Repetitive Grid & Template Filling: Investigation grids were built and shared by hand between CR and Base Operations teams over email, with no standardized automation for population, validation, or consolidation.
• ‍Missed Turnaround-Time Targets & Limited Traceability: The airline's 48-hour resolution TAT was frequently missed, and there was no documented, auditable rationale linking investigation findings to final customer responses — a gap for a regulated, safety-conscious operator.

Solution

Minfy deployed a purpose-built, two-agent Agentic AI system on the customer's Google Cloud Platform environment, orchestrated end-to-end via event-driven cloud infrastructure and designed to be explainable, auditable, and reversible at every step.

Key Solution Components

• CX Email Interpretation Agent: Triggered automatically when a complaint email lands in the CX/CR mailbox (via Microsoft Graph/Outlook integration), this agent reads and interprets the email, extracts complaint and operational context, queries flight and base-operations reference data, and auto-populates a standardized investigation grid routed to the correct base location.
• ‍Base Operations Response & Confidence-Scoring Agent: Triggered by the Base Operations team's response, this agent validates the completeness of the investigation grid, applies confidence-scoring logic to the proposed resolution, and consolidates a structured, standardized response. Scoring uses a retrieval-augmented (RAG-based) comparison against historical complaints and their resolutions — a proposed resolution that aligns more closely with a positive historical outcome receives a higher confidence score. ‍
• Event-Driven Orchestration: Cloud Functions and Pub/Sub (with dead-letter queues and retry/backoff) connect email ingestion, agent invocation, and response delivery into a single automated pipeline, deployed on Vertex AI Agent Engine and powered by Gemini models.
• ‍Contextual Session Memory: Firestore maintains per-complaint session context (keyed by Complaint ID), allowing the two agents to share state across the CX → Base Ops → CX response cycle.
• ‍Security & Governance: Credentials and API keys are managed via Secret Manager, access is scoped through least-privilege IAM service accounts on a custom VPC with private access, and all agent actions are captured through structured Cloud Logging for audit and traceability. No production PII was persisted during the POC/pilot phase. ‍
• Staged, Production-Grade Rollout: The solution progressed through architecture validation, core agent build, security hardening, and UAT before a sandboxed deployment into the airline's own GCP project — supported by a formal Go/No-Go review and structured hypercare support.

Business Benefits

• Faster, More Consistent Complaint Handling: Automated email interpretation and grid population remove the most repetitive, time-intensive steps from the CR team's workflow, targeting improved and more consistent turnaround against the airline's TAT commitments.
• ‍Confidence-Scored Resolution Logic: The Base Operations Agent applies confidence scoring to proposed resolutions, laying the groundwork for straightforward cases to move forward with minimal manual touch while ambiguous cases are routed for human review — the precise escalation thresholds were still being finalized as of the last project update, so this reflects the design intent rather than a fully confirmed, deployed rule.
• ‍Auditable, Explainable Investigations: Every automated step — email interpretation, grid population, and response consolidation — is logged, giving the airline a documented, traceable record of how each complaint was investigated and resolved.
• ‍Cloud-Native, Governed Architecture: Built entirely on the airline's own GCP environment with least-privilege IAM, private networking, and centralized secret management, the solution met the airline's data-handling and tenant-governance requirements from the outset. ‍
• Aligned with a Broader IFS Transformation Direction: The airline separately commissioned a strategic roadmap for a centralized, AI-augmented IFS platform spanning crew scheduling, compliance monitoring, and incident tracking. This complaint-automation build was scoped and delivered as its own standalone workstream, but sits within that same broader agentic-AI direction for IFS — it was not built as a technical extension of that roadmap engagement.