Autonomous Incident Intelligence Platform
Smart incident data management, run by agents
From a first ML classifier to a real-time, agent-driven GenAI platform that categorises, groups, enriches and traces the root cause of incidents across 100+ Digital Workplace products.
01Context
Every incident in ServiceNow has to be categorised, matched to related incidents and understood before anyone can fix the underlying problem. Done by hand across 100+ products, that work was slow, inconsistent and expensive — and reporting on it took days.
The challenge: Make incident data management fast, accurate and largely autonomous — so engineers spend their time fixing root causes, not sorting tickets.
02Approach
Prove value with a first ML baseline
Built the first ML-based ServiceNow incident categorisation system — saving 8 days of manual work per 10,000 incidents.
Evolve to transformers
Moved categorisation onto BERT-based models for higher accuracy on noisy incident text.
Re-platform on GenAI and agents
Evolved the system into a real-time GenAI platform where agents handle categorisation, grouping and enrichment.
Reason about root cause with a knowledge graph
Added knowledge-graph root-cause analysis, so related incidents are connected through the products and dependencies they share.
03Architecture
04Key decisions
Evolve the system, don't replace it
- Why
- Each generation — ML, then BERT, then GenAI agents — had to beat the one before it on the same incidents.
- Trade-off
- Moving in stages takes longer than one big rebuild, but every step was measured against a working baseline.
Grouping accuracy as the north-star metric
- Why
- Grouping is what turns thousands of tickets into a handful of real problems.
- Trade-off
- Optimising for one metric means some others wait — the right trade when it is the one that drives the outcome.
A knowledge graph for root cause
- Why
- Incidents are related through the products and dependencies they share, which plain similarity misses.
- Trade-off
- The graph has to be kept current, in exchange for root-cause analysis that follows real dependencies.
05Outcomes
- Real-time, agent-driven incident management across 100+ Digital Workplace products.
- Recognised with the American Express DW Project Award.