Loading_
Loading_
A drop-in classification and routing layer that sits in front of an existing ITSM tool and fixes intake without replacing it.
Measured outcomes
−34%
Reassignment rate
1.1s
Classification latency
22%
Duplicates caught at intake
6 weeks
Time to production
Not every organisation can replace its service management platform. This engine attaches to the one already in place and fixes the part that hurts most: intake quality.
It reads the ticket, assigns category, priority, urgency and assignment group, links duplicates, attaches relevant knowledge, and writes the result back through the vendor API.
Headline result
0.0%
routing accuracy
Tags
Five capabilities that define the system. Each one exists because something specific was broken.
ServiceNow, Jira Service Management, Freshservice and Zendesk supported out of the box.
Low-confidence predictions pass through untouched rather than guessing — precision over coverage.
Every human reassignment becomes a training signal; the model retrains weekly against recent ground truth.
Relevant KB articles are attached at intake, deflecting a further slice of tickets.
Run predictions alongside humans for weeks and compare before enabling write-back.
A stateless inference service behind a webhook adapter layer, with a feedback loop that closes on human corrections.
3 components
One mapping file per tenant translates vendor schema to the canonical ticket shape.
3 components
Stateless and horizontally scaled; p99 under 1.4 seconds.
3 components
No model ships without beating the incumbent on a held-out set.
No mystery components. Everything below is either open source or a platform you already own.
Interactive mock-ups of the shipped interface. The live environment is available during a demo session.
Accuracy and coverage by category over time
The running environment is available during a booked session — including a sandbox tenant you can drive yourself.
The real sequence, in order. Steps with a command are copy-pasteable.
Single Helm release; no database of record required.
$helm install ticketing aiinfraengine/ai-ticketingAuthor the tenant mapping file against your ITSM schema.
Export 12 months of resolved tickets and run the training pipeline.
$aiinfraengine ticketing train --input ./export.jsonlTwo weeks in shadow mode, review the comparison report, then turn on write-back.
Published rather than hidden behind a call. Volume and multi-year terms move these numbers.
Engine
$0.04per classified ticket
Onboarding
from $24kone-time
Need this scoped against your estate? We will size it properly, in writing, within a week.
Request a quoteWe will walk you through the architecture, the trade-offs we made, and what would change for your environment.