The problem
A SaaS company's support team of eight was handling approximately five hundred tickets per week. Every ticket was manually triaged — categorised, prioritised and assigned — before any resolution work began. Triage consumed an estimated forty per cent of the team's time, and response times were suffering as volume grew.
The triage work was necessary but mechanical. It involved reading the ticket, determining what it was about, assessing urgency and routing it to the right person. It did not require the deep product knowledge or customer empathy that were the team's real strengths.
What we built
We built an AI support triage assistant that handled the initial categorisation, prioritisation and routing of support tickets:
- Categorisation. The assistant read each incoming ticket and determined what it was about — bug report, feature request, how-to question, account issue.
- Prioritisation. Urgency was assessed based on the issue type, the customer's tier and any explicit urgency signals in the ticket.
- Context gathering. The assistant gathered relevant context before the agent engaged — customer history, recent activity, related tickets.
- Routing. Tickets were routed to the appropriate agent or team based on issue type, workload and expertise.
- Initial response drafting. For routine tickets, the assistant drafted a response for agent review.
The results
Three months after deployment:
- Triage time reduced by over eighty per cent — from manual review of every ticket to exception-only handling
- First response times decreased significantly because tickets were routed immediately rather than waiting for manual triage
- Agent capacity increased — the team handled more tickets without adding headcount
- Agent satisfaction improved because they spent less time on triage and more on resolution
How it worked
The assistant handled the triage, which was the bottleneck. Agents received tickets already categorised, prioritised and enriched with context. They could immediately begin working on resolution rather than spending the first portion of their time on triage.
The assistant's categorisation improved over time as it learned from agent feedback. When an agent corrected a category or priority, the assistant incorporated that feedback into future decisions. The accuracy was good at launch and improved steadily.
What we learned
The most important lesson was that triage automation improves the entire support workflow, not just the triage step. Faster triage means faster resolution. Better categorisation means more appropriate routing. More context means better first responses. The improvement compounds through every subsequent step.
We also learned that support teams are often enthusiastic about automation when it handles the work they find least satisfying. Triage was universally seen as necessary but unfulfilling. Freeing the team from triage freed them for the problem-solving and customer interaction that were why they chose support work in the first place.
For an overview of the assistant type, see AI customer support assistant. For how AI handles the email component, see AI email automation.
This case study describes a composite of real implementations. Results vary based on the specific process, team and context.