A confectionery company receiving 500+ customer calls per day. Call recordings were stored in Google Drive. Two employees manually listened to recordings, assessed tone, and compiled weekly reports — slow, expensive, and inconsistent.
The business had no real-time visibility into customer satisfaction. By the time negative feedback was identified, clients had already churned. Manual analysis cost $1,400/month in salaries and still delivered late, incomplete insights.
Specifically:
A fully automated AI pipeline that handles everything — from raw audio to actionable analytics — without human involvement.
The system automatically picks up new call recordings every 2 minutes, transcribes speech with 95%+ accuracy, classifies tone and issue category, saves structured data to Google Sheets and a stats database, and answers management questions in a chat interface instantly.
The manager simply asks: "What's happening with customer calls this week?"
The AI answers in seconds — total call volume, sentiment breakdown, and performance by category. Not just how many clients are unhappy, but why and where: delivery complaints spiking on Thursdays, product quality issues concentrated in one item, payment friction showing up repeatedly.
Instead of guessing what to fix, the manager sees exactly which area is underperforming — and can act on it the same day.
Step 1 — Prompt Design Designed a dual-prompt system: one for call classification (fixed business categories + tone), one for the AI Agent (date-aware, natural language queries in Ukrainian/Russian/English).
Step 2 — API Integration