The Context
A mid-sized software company provides automation software for machinery and asset management to industrial clients. Demand for their solution was rising fast, but growth created a structural problem: they couldn’t hire and onboard engineers quickly enough to keep pace.
New hires took 18–24 months to reach full productivity, largely because of the complexity of both the software and the customer environments. The software was deployed in air-gapped sites and involved coordination between multiple stakeholders—software vendors, IT service providers, in-house IT teams, project managers, and end-users. Inquiries came from anywhere, and often required navigation across several parties before resolution.
The Challenge
Knowledge lived in two places: eight years of Jira issues (over 50,000 tickets) and thousands of Confluence pages, often just screenshots with minimal description. For new hires, this was a “data graveyard”—hard to search, harder to learn from.
A single “human router” spent the day cleaning, clarifying, and assigning tickets. Senior engineers spent half their time firefighting—answering repetitive questions, clarifying incomplete tickets, and helping juniors. This slowed innovation and kept the team in constant reactive mode. Categorization accuracy hovered around 15%, and documentation coverage around 10%.
Our Approach
Arti began with a deep analysis of 50,000 Jira issues. The results revealed concentration points: just 3 of the 12 software modules accounted for over 50% of the workload. This provided a starting point for high-impact automation.
We then built an AI-driven router:
-
Achieved 75% auto-categorization accuracy out of the box, trained with a human-in-the-loop pipeline to push toward 80%.
-
Automatically checked tickets for completeness, requesting missing details before submission.
-
Surfaced the three most relevant historical issues, along with concise summaries of problem and resolution, to guide engineers.
In parallel, AI agents extracted recurring problem/solution clusters. With senior engineers in the loop, these were structured into a new, usable knowledge base—one that actively supported engineers instead of sitting unused.
Results
The impact was immediate and measurable:
-
Onboarding time dropped from 18–24 months to just 4–6 months.
-
Automated categorization climbed to 75–80% accuracy.
-
The human router role was eliminated, freeing capacity.
-
Senior engineers reclaimed time for innovation instead of firefighting.
Impact
By structuring their “data graveyard” into a usable asset, the company created a foundation for scale. New hires could ramp up in months, not years. Inquiries were routed cleanly through a complex multi-stakeholder environment. And most importantly, the company could continue to grow without linear headcount expansion.
