<aside> 💡
The problem: QA reviewed support calls manually — listening, scoring subjectively — with no consistent standard between reviewers and no reliable way to catch a bad call before the client complained about it directly.
What it looked like before: Bad calls often went unnoticed until damage was already done, and there was no structured process for turning a problem call into an apology and a resolution for the client.
What we built: A system that transcribes and scores every call across 8 QA categories automatically, flags anything below threshold, and drafts a client apology email that a manager has to explicitly approve before it goes out.
Result: Every call now gets scored consistently with zero manual listening for calls that pass, problem calls get caught and escalated before a client has to complain, and nothing reaches a client without human sign-off — with a full audit trail logged automatically.
</aside>
AI system that automatically transcribes, evaluates and escalates problematic customer support calls, with a person checking before any letter to the customer.
Context:
Problem:
<aside> 🧠
A Google Drive folder watch triggers transcription and AI scoring on every new call recording. Calls scoring below threshold automatically escalate to a manager via Telegram with a pre-drafted client email — nothing reaches the client without explicit human approval.
Tools used: n8n, Google Drive, Whisper-compatible transcription, Groq (Llama 3.3), Google Sheets, Telegram, Gmail.
</aside>
Solution logic: