The problem
Ticket volume outgrew the support team during peak periods. Response times slipped, satisfaction fell, and hiring seasonal agents was expensive and slow.
What we built
- A triage layer that classifies intent and sentiment on arrival
- Auto-resolution for the highest-volume repetitive request types
- Intelligent routing so only genuinely complex tickets reach a human
How we got there
- 1
Clustered a year of tickets to size each request category
- 2
Automated the top categories first and measured containment weekly
- 3
Set escalation rules on low-confidence and negative-sentiment tickets
Results
- •85% ticket auto-resolution rate
- •50% improvement in customer satisfaction
| Metric | Before | After |
|---|---|---|
| Tickets auto-resolved | 0% | 85% |
| Customer satisfaction | Baseline | +50% |
Technologies
NLP
Sentiment Analysis
Knowledge Graphs
What’s next
Feeding resolved-ticket patterns back into product and help-centre content.