← Back to case studies
Confidential Client
AI-Powered Incident Classification & Prediction for ITSM
Triage time
manual
→
significantly reduced
Classification accuracy
above UAT threshold
ITSM UI changes
none
A confidential enterprise IT client needed to reduce the manual triage burden on their service desk by automating incident classification and routing using machine learning.
What changed
Before
Ticket
Raised in ITSM
Triage
Manual categorisation
Route
Queue by hand
After
Ticket
ServiceNow
unchanged UI
Classify
DNN
transfer learning
Integrate
REST API
Maintain
Retraining pipeline
drift as patterns change
Classification sits beside the existing ITSM rather than inside it — a REST call in, a category back, no platform change.
What we built
The client's service desk was handling high volumes of IT incidents, with manual triage consuming significant analyst time and introducing inconsistent categorization. The solution needed to ingest historical incident data, train classification models, and integrate predictions directly into the ITSM workflow.
Key Features
Data engineering pipeline covering ingestion, cleansing, feature extraction, and data profiling from ITSM system APIs.
Comparative ML evaluation: supervised and unsupervised approaches benchmarked before model selection.
Deep Neural Network with Google transfer learning for multi-class incident classification.
Feature engineering pipeline extracting text, categorical, and temporal signals from incident records.
REST API delivery for real-time prediction integration into the existing ITSM workflow.
Model monitoring and retraining pipeline to maintain accuracy as incident patterns evolve.
Results
AI-powered incident categorization integrated into the ITSM workflow, reducing manual triage time significantly.
Multi-class classification accuracy exceeded client-defined acceptance threshold in UAT.
REST API integration required no changes to the existing ITSM platform UI.
Retraining pipeline ensured model performance was maintained as ticket patterns changed over time.
Stack
Engagement: 6 months · Real-time & Industrial Data Platforms, Product Engineering, AI Infrastructure & MLOps