Customer Signal AI:

Automated Call Scoring, Agent Coaching & Client Retention for a Premium Bag Brand

🏢 Context

Company: HUSH - B2C premium bag e-commerce Industry: Online Retail (Fashion Accessories) Target clients: End consumers purchasing premium bags online Support volume: Inbound customer service calls (daily) Team: Customer service agents + 1 manager Core problem: Zero visibility into call quality - no QA, no coaching, no retention follow-up

❌ Problem

Metric	Value	Issue
Call QA coverage	0%	No calls reviewed systematically
Agent feedback	None	Coaching was ad-hoc and reactive
Bad call detection	Never	Poor calls went completely unnoticed
Client retention	None	At-risk customers never followed up
Manager time on QA	0 (not done)	Too time-consuming to do manually

The customer service team handled inbound calls daily with no quality control in place. Managers had no way to know which calls went badly, which agents needed coaching, or which customers were at risk of churning. Every poor call was a silent loss — no detection, no recovery.

✅ Solution

Built a 4-workflow AI automation system that transcribes, scores, alerts, and recovers:

Call recording → Google Drive
                      ↓
            WF01: Transcribe & Score
                      ↓
         Filter: Audio → Whisper → GPT QA Score
                      ↓
              Log to Google Sheets
                      ↓
                 Score < 3.0?
                      ↓
        ┌─────────────┴─────────────┐
     YES                           NO
Slack Alert                       End
+ Call WF03
        ↓
WF03: Draft Retention Email
        ↓
  AI analyzes call → drafts email
        ↓
  Save to Sheets → Telegram to Manager
        ↓
WF04: Manager Decision (Telegram buttons)
        ↓
   ┌────┴────┐
 Send      Refuse
   ↓          ↓
WF02:      Mark as
Get email  Rejected
+ Send

WF01 — Call QA: Transcribe & Score

Picks up every new call recording automatically. Transcribes using Whisper, scores the agent across 8 QA criteria (1–5 scale) using GPT-5-mini, logs full results to Google Sheets, and fires a Slack alert for any call scoring below 3.0.

WF02 — Call Recording URL to Email

Sub-workflow triggered on manager approval. Extracts the Google Drive file ID from the recording URL and retrieves the client email for sending the retention email.

WF03 — On Bad Call: Notify Manager

Triggered automatically when score < 3.0. Reads the call transcription, runs a second AI agent to draft a personalized retention email and write a 2–3 sentence call summary for the manager. Sends everything to the manager via Telegram with Send/Refuse buttons.