Context and problem

In a typical SMB environment, requests come from multiple sources: contact forms, email, and sometimes multiple inboxes. Without structure, agents manually read messages, classify them, set priority, and forward them, often in Slack.

The goal was to build a reliable intake and triage system that works with shared-hosting constraints, scales through n8n, and persists tickets in a database so a future UI and audit trail can be added cleanly.

Solution overview

The system is split into three layers: intake through webhook and IMAP, AI triage through OpenAI structured output, and an ops layer with MySQL persistence and Slack notifications. Each request is normalized into one JSON shape, triaged by AI, and stored as a ticket.

Instead of free-form text, OpenAI returns strict JSON according to an agreed schema. That avoids fragile parsing, gives the workflow a stable contract for downstream steps, and makes confidence gating easier to apply.

• Webhook and IMAP intake are merged into one consistent input contract.
• OpenAI returns strict JSON for category, priority, team, and confidence.
• MySQL persistence creates a stable ticket record and future audit path.
• Slack routing gives the ops team clear context while preserving a human review path for ambiguous cases.
• Confidence below the threshold goes to triage instead of being auto-routed.

Why it works

The HelpBot Ticket Desk pipeline provides a fast, consistent, and extensible way to handle requests. Each input becomes a structured ticket in MySQL, the ops team gets clearer Slack notifications, and confidence gating enables safer automation without pretending the system should make every decision on its own.