Automated Classification, Routing & Analytics for a Bag Store

🏢 Context

Company: Hush - B2C premium bag e-commerce Industry: Online Retail (Fashion Accessories) Target clients: End consumers purchasing bags online Support volume: 15 -30 emails/day Team: 1 owner managing support manually Core problem: Every email handled manually - no triage, no prioritization, no analytics

❌ Problem

Metric Value Issue
Email handling 100% manual No triage or prioritization
Response time Hours to days Urgent cases missed in queue
Escalation detection None Legal threats not flagged immediately
Support analytics None No data on categories, sentiment, volume
Human time wasted 2–4 hrs/day Standard FAQs answered manually every time

The owner was answering every email manually — from “what’s the zipper material?” to legal threats — with no system to separate urgent from routine. At 15–30 emails per day, this consumed a significant part of the working day with no prioritization or visibility into patterns.

✅ Solution

Built a 4-workflow AI email support system that classifies, routes, responds, and analyzes:

Gmail (unread) → Filter → AI Classifier → Parser
                                              ↓
                                        Switch Router
                                              ↓
              ┌───────────┬───────────┬───────────┬───────────┐
              P1          P2          P3          P4        Default
         Draft+Telegram  Draft     Auto-Reply   Tag       Needs-Human
              └───────────┴───────────┴───────────┴───────────┘
                                              ↓
                                     Google Sheets Log
                                              ↓
                                  Analytics Agent (Chat UI)

WF1 — Email Support Automation

Processes every incoming email automatically. AI classifies priority (P1–P4), category, sentiment, and generates a suggested response. Routes to correct action based on priority and requires_human flag.

WF2 — Report for Period

On-demand analytics tool. Accepts a date range and returns 4 metrics: total emails, % negative sentiment, % requiring human intervention, top 3 categories.

WF3 — Comparison of Two Periods

Compares two date ranges side by side. Returns % change for each metric with plain-language conclusions: improvement, deterioration, growth, decline, or stable.

WF4 — Customer Support Analyst (AI Agent)

Natural language chat interface. The owner asks questions in plain text — the agent detects dates in any format, selects the right tool, and returns a readable summary.