MIRKO

Workflow Agents That Triage 60,000 Service Tickets a Month With Human Approval

LLM workflow agents for an industrial equipment maker: they read service tickets, telemetry and manuals, propose a diagnosis and parts list, and route anything irreversible to a human, auto-resolving 71% of tickets.

Industry
Manufacturing
Country
Industrial Equipment Maker, Austria
Team Size
9
Services
LLM & AI Orchestration, AI Integration, IoT Development
Technologies
Python, LangGraph, Claude, GCP, BigQuery, ServiceNow
Workflow Agents That Triage 60,000 Service Tickets a Month With Human Approval

Overview

About The Project

An industrial equipment maker with 40,000 machines in the field received 60,000 service tickets a month from operators, dealers and the machines themselves. First-line engineers spent most of their day finding the right manual section, checking telemetry and asking for missing information.

We built a set of workflow agents inside their ServiceNow process: a triage agent classifies and enriches tickets, a diagnosis agent correlates telemetry with known failure modes and manuals, and a resolution agent proposes a fix and a parts list. Engineers approve or edit; the agents learn from every decision.

The Challenge

Volume, VarietyAnd Safety

Tickets arrived in nine languages, often as a photo and one sentence. Telemetry lived in a separate data warehouse, manuals in a document system, and parts in SAP. No single engineer had all four open at once.

Some resolutions are safety-relevant. Sending a remote reset to a running machine or shipping the wrong hydraulic part cannot be left to an unsupervised model.

Volume, Variety And Safety

Our Solution

Three Agents,One Approval Flow

Each agent has a narrow tool set exposed through typed interfaces: ticket read and update, telemetry queries in BigQuery, manual retrieval with citations and parts lookup in SAP. A manager workflow in LangGraph coordinates them and keeps a ledger of what was checked and why.

Resolutions are tiered: informational answers go out automatically, parts orders under a threshold notify the dealer, and remote actions or safety-relevant parts wait for an engineer's approval in ServiceNow. An evaluation set of 3,000 historical tickets gates every release, and production tickets are sampled weekly.

Three Agents, One Approval Flow

Key Features

What WeDelivered

The Impact

Results ThatMatter

Engineers now handle the hard tickets; the agents clear the rest with a full evidence trail.

71%

Tickets auto-resolved

60K

Tickets per month

-54%

Mean time to resolution

11 months

Discovery to full rollout