AI operations2024 - Present

AI Ticket Triage

GPT-powered Freshservice incident triage that categorizes tickets, detects catalog requests, writes troubleshooting notes, and drafts requester responses before an agent begins work.

impact.

5K+tickets processed yearly

420+hours saved yearly

3category levels

750+projected phase 3 hours

case study.

how it works.

Every new support ticket is intercepted, summarized, classified, and enriched with private agent notes. The system coordinates focused helper flows for categorization, catalog matching, access detection, internal guidance, and public replies.

Three-level categorization using strict taxonomy matching
Catalog item detection for application access requests
Private HTML troubleshooting notes generated for agents
Requester-facing replies drafted from the ticket context

model selection.

The current flow uses OpenAI with structured lookup tables for the service catalog and category hierarchy, keeping the model constrained to approved labels and public-safe response formats.

roadmap.

Phase 3 targets automatic resolution for repetitive issues such as MFA resets, lost phone workflows, reopened ticket cleanup, and common access problems with human escalation for ambiguous cases.

tech stack.

OpenAI / Freshservice / Workato / Service Catalog / Prompt Engineering / HTML Notes / REST APIs

Workflow visual

Freshservice AI intake, step by step.

A public-safe walkthrough of the two intake patterns: new incidents that need triage and reopened tickets that need a close-or-continue decision.

Workflow

New incident triage path.

Trigger

Freshservice sends a new incident

A real-time webhook delivers the ticket ID, subject, description, requester context, catalog list, and category hierarchy.

Ticket IDSubjectDescriptionRequesterCatalogTaxonomy

Context

Automation builds the AI packet

The ticket is wrapped with approved lookup data so the model can classify against known labels instead of inventing new ones.

Helper flows

Focused AI helpers analyze the request

Separate helper flows handle categorization, catalog matching, access detection, internal troubleshooting guidance, and requester messaging.

CategorizeCatalog matchAccess intentIT noteUser reply

Decision

The flow decides the ticket path

The output separates normal IT work, access requests that need catalog links, duplicate or missing context, and non-IT redirects.

Write-back

Freshservice gets structured notes

The ticket is updated with category fields, detected applications, a private agent note, and a public HTML reply.

L1-L3 categoryCatalog appsPrivate notePublic reply

Samples

Pick a ticket and see what the flow writes.

OpenAI helper flows

Ticket sample

Requester needs access to Docker Desktop and Zoom Business for a new dev setup.

Context read

Two application names appear in the description and both have catalog matches.