Title: AI-powered automation for client request filtering.

Context: Business.

Problem: As the store scales, a challenge arises in maintaining the quality of client request handling.

Solution: The system consists of 5 interconnected workflows in n8n that together form a complete audio call processing cycle.

Overall logic:

  1. Handle Audio Queue — the main orchestrator. Pulls calls from the queue (Google Sheets), routes them for processing, and manages the flow between all other workflows.
  2. AI Audio Assistant — an AI agent powered by an LLM (OpenRouter / Nemotron) that accepts user requests via chat and returns call statistics for any given time period. It understands imprecise date expressions — such as "last week" or "in March."
  3. Get Call Stats — a sub-workflow for statistics aggregation. Retrieves data from the spreadsheet and returns the total number of calls, the number of negative calls, and their percentage.
  4. Status Tone / Status Category — two separate workflows that write analysis results (call tone and category) back to Google Sheets. Both include overwrite protection — if a field is already populated, no re-writing occurs.
  5. Status Audio "Done" — marks the completion of processing: sets the status to done and records the processing timestamp. Also protected against duplication.

Key feature: Each workflow is an independent module with a clearly defined input and output. This allows the system to scale — adding new analysis types or data sources without rewriting the entire logic.

Process:

1. Prompt Design

A system prompt was written for the AI agent, establishing its role as a customer request statistics assistant. The prompt dynamically passes the current date to the agent via $now — enabling it to correctly interpret relative time-based queries such as "this week" or "in the last 3 days" without requiring additional clarification from the user.

2. API Integration

The system integrates three external services:

3. n8n Workflow