AI Help Desk Agent for Jira Service Management

What was built

A Glean agent rebuilt as a conversational front door for Jira Service Management (JSM) issue logging. It routes requests, deflects self-serve issues via RAG, and creates tickets with required custom fields on the fly through MCP.

Why it mattered

Employees think app-first (e.g. Teams). JSM organizes work team-first (e.g. CorpIT). That mismatch caused slow issue logging, wrong queues, employee frustration, and missed self-service deflection.

At a glance

2-3

prompts to ticket

20+

request types routed

15 min

time reclaimed per ticket

GleanJiraMCPRAGInsurance

The Unlock

The team was already trying to solve issue intake with an agent, but the workflow had been blocked for 3 months by three failure points: Inconsistent routing, Unsupported JSM ticket creation, and Ungrounded deflection.

01

Inconsistent Routing

Semantic matching Deterministic mapping

A flat mapping table replaced agent guesswork, sending each request to the right application, service desk, and queue.

02

Brittle Integration

Static REST payload Runtime field discovery

MCP reads required JSM fields at runtime, creating each issue type with the right custom fields without manual payload upkeep.

03

Ungrounded Deflection

Web search Tagged knowledge base

RAG checks a curated, tagged knowledge base first, surfacing relevant self-serve answers before creating a ticket.

Three architecture moves turned a stalled prototype into a maintainable enterprise intake system.

How It Works

AI Help Desk Agent System Flow Color-coded workflow showing Glean conversation, Confluence RAG lookup, and Jira Service Management issue creation. Source system Glean Confluence Jira Yes No Yes No Yes No Employee Starts Chat Context Capture Subject and description Mapping Table Issue keyword matching Confident Match? Follow-Up Questions Agent asks targeted questions Issue Type Resolved RAG Lookup Search grounded knowledge base Self-Serve Article Found? Present Article User Satisfied? Self-Service Resolved MCP Query Issue type required and custom fields Populate Payload Mapped values User Validation Confirm ticket details Issue Created

How the agent routes an employee request from chat to self-service help or the right Jira queue.

Impact

EMPLOYEE EXPERIENCE

One front door, not a maze

Employees describe the problem in plain words. The agent finds the desk, type, and fields. No wrong queues.

KNOWLEDGE BASE

One source of truth, two wins

One tagged library. Self-serve issues get deflected; everything else reaches the admin with the relevant article attached.

ARCHITECTURE

Built to not break

A flat mapping table and runtime field discovery absorb change. New fields or issue types need no developer.