The Challenge
Running AI-driven ticket triage in a healthcare support environment only works if the system can handle sensitive data safely and apply complex routing logic reliably. A healthcare company had a support team of 10 analysts processing 350–500 Jira Service Management tickets per week from more than 1,000 Community Health Centers, and every single ticket required manual triage.
Two critical problems emerged: the classification and routing process consumed the equivalent of one full-time employee's capacity, and the routing logic lived in tribal knowledge across the team rather than in any structured system. On top of that, tickets frequently contained PHI —including screenshots and attachments—making automation a compliance challenge, not just a technical one.
We began with a deep discovery process, mapping the company's routing rules, SLA requirements, and PHI handling constraints before writing a line of code. This allowed us to define a rules-first architecture that would encode the organization's own logic — rather than relying on a black box — with LLM fallback only for edge cases.
The Approach
Our engineering team designed and implemented an event-driven integration between Jira Service Management and Azure that ingested unassigned tickets in real time and triggered automated triage instantly. A PHI-aware preprocessing and OCR pipeline was built to sanitize ticket text and attachments before any LLM processing — ensuring 100% HIPAA compliance throughout.
Classification and assignment agents were deployed to apply the company's routing rules, regional logic, capacity caps, and product specialization — with deterministic rules taking priority and the LLM stepping in only when the rules didn't cover a case. Every decision came with an audit trail and a plain-language explanation.
At the same time, a human-in-the-loop layer was embedded into the workflow. Analysts review AI-generated assignments before they go live, and their feedback is captured and stored — creating a continuous improvement loop that makes the system more accurate over time.
The result was a live AI triage agent running inside Jira, processing 500+ tickets per week in 3–4 minutes per ticket instead of 15+, while maintaining the organization's 4-hour SLA across all clients.
The Solution
We delivered the full system within 3 months, built and deployed by a cross-disciplinary team of a Data Engineer, ML Engineer, Project Manager, and Product Manager. The agent operates entirely within the company's Azure tenant—no sensitive data leaves their environment.
The system freed the equivalent of one full-time support role—approximately 50 hours per week—previously spent on manual triage, reallocating $70,000 in annual capacity toward higher-value client resolution work.
The Outcomes
The triage system is now running in production across the company's full support operation. As the feedback loop matures and the system learns from human corrections, assignment accuracy continues to improve — without requiring manual retraining.
