Support
Clearing a support queue without a night shift
A consumer app with 40,000 users let a Ticket Triage Agent watch Zendesk and WhatsApp around the clock. Humans still answer; they just start from a draft.
· 7 min read
Support at a growing consumer app came through two doors: Zendesk tickets and a WhatsApp business number. The team of four could keep up during the day and fell behind every night. They did not want a chatbot that argued with customers. They wanted the queue to be sorted and half-written when they arrived.
One Agent, two channels
The Ticket Triage Agent signed into Zendesk and WhatsApp from its Apps settings. Each connection belongs to that Agent alone. It reads new tickets and WhatsApp conversations on a cadence, tags them by topic and urgency, and merges duplicates.
Drafts, not auto-replies
For every ticket it writes a suggested reply using the help centre and past answers. The team sees the draft in the console thread, next to the Agent's note on why it chose that answer. Sending is a human click. Refund and account changes are flagged for a person, never performed.
Escalation rules
Anything mentioning safety, legal threats or payments goes to a Slack channel immediately, with a link to the ticket. The Agent stays quiet when the queue is clean, so the channel is not noisy.
Results after a month
First response time at 9 a.m. dropped because the backlog was already sorted. The team spends its time on the hard cases. The Agent's history page shows every model call and tool action it made, which is how the team audits it.
Takeaways
- Triage and drafting run all night on the customer's own node.
- No message leaves without a human send.
- Every tool action is logged in the Agent's history.