Business impact

<aside> 💡

The problem: Customer requests piled up in a shared Gmail inbox with no formal SLA and no analytics — a manager manually judged urgency, tracked deadlines by hand, and overdue tickets were usually only caught after a customer complained.

What it looked like before: Classification depended on one person's subjective judgment, there was no way to see request volume or recurring patterns, and any management report meant manually reading through every ticket.

What we built: An AI classifier that tags and routes every incoming request automatically, a 3-level SLA escalation watcher running every 5 minutes, and an AI agent that answers "analyze the last week" with a full report on demand instead of a manual review.

Result: First-response handling time dropped from several minutes to under 1 minute, missed SLA breaches were eliminated entirely, and management gets metrics + root-cause patterns + recommendations from a single chat request — running on free-tier LLM inference, at zero added cost.

</aside>

An automated system for classifying customer inquiries, monitoring SLA compliance with escalation, and performing AI-driven analytics — both quantitative and qualitative — built with n8n and an LLM.

Project Details

Context:

• Business/industry: IT & SaaS / Customer Support Services
• Current process: Customer requests arrive in a shared Gmail inbox. A manager manually assesses urgency, decides who should receive each ticket, and tracks response deadlines by hand. Management had no tool to analyze request volumes or identify recurring issues — only scattered emails in Gmail.

Problem:

• Classifying and prioritizing requests took too much time and depended on the manager’s subjective judgment.
• There were no formal SLA standards or a mechanism to monitor compliance.
• Overdue requests were discovered by chance, often only after a customer complaint.
• There was no analytics on request volume — how many requests were received, in which categories, and with what sentiment.
• Management had no tool to understand the root causes of recurring issues, and producing any summary report required manually reviewing all tickets.

Solution

<aside> 🧠

The system automatically receives customer requests, classifies them using AI with an explanation of the decision, logs them, and notifies the ticket owner. A separate SLA timer monitors deadlines with a three-level escalation process. On top of this, an analytics layer was built: quantitative metrics and qualitative AI-based pattern analysis, accessible through a conversational AI agent (chatbot) on demand.

Tools used: OpenAI API, n8n, Zendesk API, Slack Webhooks, PostgreSQL.

</aside>