The Context
A fast-growing B2B SaaS company was struggling to scale support. Ticket volumes surged, onboarding new agents took months, and critical expertise lived in scattered systems.
-
15,000+ tickets across 5 years, almost all manually handled
-
New agents took months to train due to inconsistent categorization
-
Institutional knowledge locked in tribal habits, not systems
Despite investing in documentation, the data was too fragmented to learn from. Each new agent restarted the same learning curve from scratch.
The Challenge
Automation attempts had failed before — too complex, too detached from daily operations.
The team needed a solution that:
-
Integrated natively into existing Jira workflows
-
Reduced onboarding time by making patterns visible
-
Proved real accuracy within 8 weeks
-
Required no behavior change from frontline agents
Our Approach
Arti analyzed 15K historical tickets to uncover hidden taxonomy patterns and recurring intents.
We built a human-in-the-loop ML model directly inside Jira:
-
AI suggested categories; agents validated in 5 seconds
-
No new tools or dashboards
-
Retraining every two weeks from validated tickets
-
Champion-led adoption (15 min/day commitment)
The Results
| Metric | Before | After 8 Weeks |
|---|---|---|
| Auto-Categorization | 15% | 68% |
| Validation Accuracy | — | 97% |
| Manual Triage | 255 tickets/mo | 96 tickets/mo |
| Time Saved | — | ~20 hrs/mo |
| New Agent Ramp-Up | 8 weeks | 3 weeks |
“Adoption was seamless. Five seconds per ticket, no new tools. The system learns from our corrections every two weeks.”
— Support Operations Lead
The Impact
Routine triage work dropped by two-thirds, onboarding accelerated, and knowledge capture improved. The team scaled faster without burning out — and built a foundation for continuous learning inside Jira itself.
