Business Outcome

<aside> 💡

Problem: Yabko runs 130+ Apple electronics stores across Ukraine, each with hundreds of Google Maps reviews. Reading them manually to find out what customers complain about most, and where, is not realistic.

Before: Reviews were handled one by one: each got a reply, but there was no overall picture by topic or by location.

What we built: An n8n system that collects public reviews for every store in the chain, uses AI to tag each one by topic and sentiment, calculates the negative share per topic and per store in code, and has a second model write a plain-language report for the marketing director.

Result: Collected 5,794 reviews across 133 stores in 47 cities; 3,712 reviews with text went through AI classification. Found that more than half of the chain's replies to negative reviews follow two templates. Top 3 complaint drivers: [...].

</aside>

Customer experience research with no access to the company's internal systems: public Google Maps reviews only, with AI classification and analytics combined in a single n8n workflow.

Project Details

Context:

• Business/industry: Retail chain of Apple electronics stores and service centers (Yabko), 130+ locations in Ukraine.
• Format: Free research project agreed with the Yabko team via LinkedIn; results reviewed by the marketing director.
• Current process: The team replies to reviews, but there is no systematic analysis of which topics and which stores drive negative feedback.

Problem:

• Thousands of reviews are scattered across 130+ Google Maps listings.
• Manual analysis takes weeks and goes stale quickly.
• A 4.5–5.0 average rating hides local problems at specific stores.

Solution

<aside> 🧠

Apify collects Google Maps reviews for every store in the chain over the last 12 months. n8n filters out competitor stores with similar names and reviews without text, groups reviews into batches of 15 and sends them to OpenAI. The model returns a topic, sentiment and short summary for each review as JSON. A Code node calculates the negative share per topic and per store, and a second model turns the numbers into a plain-language report in Google Docs. Raw classified data is stored in Google Sheets for manual verification.

Tools used: n8n, Apify (Google Maps Scraper), OpenAI (via n8n built-in credits), Google Sheets, Google Docs, JavaScript.

</aside>

Solution logic: