All case studies
Case Study · 2025 · Internal R&D · 7 engineering days

Inbox Triage

Support-queue classification and routing for teams without a helpdesk platform.

Reference implementation for our reviewable-automation standard.

triage.internal / queues / shared-inboxInbox TriageQueueAssignmentsSLARulesMacrosAnalyticsShared inbox queue184 open · median first response 11 minAUTO-CLASSIFIED91%last 7 daysBREACH RISK7under 30 min to SLABACKLOG184-22% week on weekNext upSUBJECTPRIORITYOWNERSLA LEFTFailed payout retryP1Ops · rota00:18Invoice mismatch queryP2Finance02:41Access request — depotP3IT07:05Duplicate contact mergeP3CRM11:22Refund exceptionP1Ops · rota00:44
The problem

A shared support mailbox holds up until volume grows. Then everything reads as urgent, the same handful of questions are answered repeatedly, and there is no defensible measure of response time — while a full helpdesk platform is more organisational change than the team wants to absorb.

What we built
  • A classifier tagging each inbound message with intent, urgency, and suggested owner, recording its confidence alongside the tag.
  • A rules layer that auto-routes only above a confidence threshold; everything below lands in a human review queue rather than being silently mis-filed.
  • A response-time dashboard reporting first-reply and resolution times by intent, so staffing decisions rest on measurement.
The result

Against a hand-labelled corpus of 500 messages, routing agreed with the human label on 91% of items at the selected confidence threshold, with the remaining 9% held for review rather than mis-routed. Every automated action is logged and reversible.

Want this level of documentation on your project?

Every engagement ships with the same written record: what changed, why, and what it measured.

Start a conversation