Description
Overview
This customer insights automation workflow facilitates extracting, analyzing, and summarizing Trustpilot reviews via a no-code integration pipeline. Designed for data analysts and customer experience professionals, it converts raw customer feedback into structured insights by leveraging web scraping, vector embeddings, and clustering techniques. The workflow initiates with a manual trigger and uses an HTTP Request node to scrape recent Trustpilot pages.
Key Benefits
- Automates collection and parsing of multiple pages of Trustpilot reviews for thorough analysis.
- Transforms unstructured review text into vector embeddings for efficient similarity searches.
- Clusters related reviews using K-means to identify prevalent themes and feedback groups.
- Generates concise summaries and sentiment analysis with a large language model for actionable insights.
- Exports structured insights and raw data seamlessly into Google Sheets for easy consumption.
Product Overview
This automation workflow begins with a manual trigger node to initiate the process on demand. It sets the target company identifier, which is used to construct a Trustpilot URL to scrape up to three pages of recent reviews via HTTP requests. The HTML extraction node parses multiple data fields including review author, rating, title, text, dates, country, and review URLs using CSS selectors. Extracted arrays are consolidated into structured review objects.
Each review is then processed into vector embeddings using OpenAI’s embedding model after applying a recursive character text splitter to handle long text inputs. The vectors and associated metadata are stored in a Qdrant vector database, enabling efficient similarity searches and filtering. After data storage, a subworkflow is triggered to query reviews within a specified date range and retrieve up to 500 records with their embeddings for analysis.
The workflow applies a K-means clustering algorithm in Python to group reviews by semantic similarity into up to five clusters, filtering out smaller clusters with fewer than three points. It then fetches detailed metadata for each cluster from Qdrant. Using an LLM-powered information extractor node, it summarizes grouped reviews into concise paragraphs, assesses overall sentiment, and suggests potential improvements. Finally, results including insights and raw responses are appended to a Google Sheets document for further use.
Error handling relies on default platform mechanisms; no explicit retry or backoff logic is configured. Authentication uses predefined API credentials for Qdrant, OpenAI, and Google Sheets integrations. No data persistence beyond the vector store and sheet export occurs internally.
Features and Outcomes
Core Automation
The automation workflow accepts a manual trigger initiating data extraction and analysis from Trustpilot reviews through this orchestration pipeline. It consolidates review data arrays into structured objects and applies vector embedding generation followed by clustering to detect thematic groupings.
- Deterministic manual start with explicit company ID setting for targeted scraping.
- Single-pass extraction and transformation of multi-page review data into vectorized format.
- K-means clustering algorithm dynamically adjusts to data size with up to five clusters.
Integrations and Intake
This no-code integration connects to Trustpilot via HTTP requests for review scraping and leverages Qdrant vector database for storage and similarity queries. OpenAI API is used for embedding generation and language model inference. Google Sheets integration facilitates export. Authentication is managed via API keys.
- Trustpilot scraped with up to 3 paginated HTTP requests filtered by company ID.
- Qdrant vector store handles insertion and querying of review embeddings and metadata.
- OpenAI API used for text embedding and summarization with credential-based authentication.
Outputs and Consumption
The workflow outputs structured insights including summarized review paragraphs, sentiment classification, and suggested improvements. These results are compiled alongside raw review data and metadata, then appended asynchronously to a Google Sheets document. Outputs conform to JSON schema generated by the information extractor node.
- JSON format insight objects with fields: Insight, Sentiment, Suggested Improvements.
- Exported data includes review counts, date ranges, and raw concatenated review texts.
- Google Sheets target allows easy access and further analysis or reporting.
Workflow — End-to-End Execution
Step 1: Trigger
The workflow starts manually via the “When clicking ‘Test workflow’” node, which allows controlled execution. It requires setting the target company identifier to correctly form the Trustpilot review URL before proceeding.
Step 2: Processing
Review pages are scraped through HTTP requests with pagination limiting to three pages. Extracted HTML content is parsed via CSS selectors to retrieve multiple review attributes. The data arrays undergo consolidation into structured objects representing individual reviews. Basic presence checks ensure data integrity before further processing.
Step 3: Analysis
Vector embeddings are generated for each review text chunk using OpenAI’s embedding model. Reviews are stored in Qdrant with metadata. The workflow queries these vectors for a defined date range and applies a K-means clustering algorithm in Python to group similar reviews. Clusters with fewer than three points are filtered out to focus on significant themes. An LLM then summarizes and extracts sentiment from grouped reviews.
Step 4: Delivery
Insights and clusters are compiled into a JSON structure and asynchronously appended to a Google Sheets document. The export includes metadata such as company ID, date ranges, number of responses, and raw review texts, enabling downstream analysis or reporting.
Use Cases
Scenario 1
A customer experience analyst needs to understand recent feedback trends from Trustpilot reviews for a retail company. This workflow scrapes, clusters, and summarizes reviews monthly, providing clear insights and sentiment analysis. The result is a structured report exported to Google Sheets, enabling evidence-based decision-making.
Scenario 2
A product manager wants to identify common pain points in customer reviews to guide feature improvements. The automation pipeline groups similar reviews via clustering and generates concise summaries with suggested improvements. This reduces manual review reading, focusing efforts on actionable feedback.
Scenario 3
A marketing team requires ongoing monitoring of brand sentiment from online reviews. Using this orchestration pipeline, they extract and vectorize reviews, apply sentiment classification, and export results to a shared sheet. This ensures consistent tracking of customer opinion without manual intervention.
How to use
After importing the workflow into n8n, set the target company ID in the designated variables node. Authenticate API credentials for Qdrant, OpenAI, and Google Sheets integrations. Trigger the workflow manually to start the extraction and analysis process. The workflow will scrape recent reviews, generate embeddings, cluster similar feedback, summarize insights, and export results to the configured Google Sheets document. Results include sentiment and suggested improvements, refreshed per execution.
Comparison — Manual Process vs. Automation Workflow
| Attribute | Manual/Alternative | This Workflow |
|---|---|---|
| Steps required | Multiple manual steps including scraping, parsing, clustering, and summarizing. | Single automated pipeline combining scraping, vectorization, clustering, and insight generation. |
| Consistency | Subject to human error and variable interpretation of reviews. | Deterministic extraction and clustering with standardized LLM summarization. |
| Scalability | Limited by manual reading capacity and data handling. | Handles hundreds of reviews with vector search and clustering algorithms. |
| Maintenance | Requires frequent manual updates and tool switching. | Automated with configurable parameters and reusable API integrations. |
Technical Specifications
| Environment | n8n automation platform with Python support for code nodes |
|---|---|
| Tools / APIs | Trustpilot (web scraping), Qdrant vector database, OpenAI embeddings and chat APIs, Google Sheets API |
| Execution Model | Manual trigger with synchronous scraping and asynchronous vector storage and export |
| Input Formats | HTML pages from Trustpilot, JSON metadata for reviews |
| Output Formats | JSON insights and review metadata appended to Google Sheets |
| Data Handling | Transient review text chunking, vector embedding generation, metadata persistence in Qdrant |
| Known Constraints | Relies on Trustpilot HTML structure stability and external API availability |
| Credentials | API keys for Qdrant, OpenAI, and Google Sheets OAuth2 |
Implementation Requirements
- Valid API credentials configured for Qdrant vector store, OpenAI API, and Google Sheets OAuth2.
- Network access allowing HTTP requests to Trustpilot and API endpoints.
- Properly formatted company identifier matching Trustpilot URL schema.
Configuration & Validation
- Set the company ID in the “Set Variables” node to match the Trustpilot URL format.
- Verify API credential connections for Qdrant, OpenAI, and Google Sheets nodes are active and authorized.
- Run the workflow manually and confirm that review data is scraped, processed, clustered, and exported without errors.
Data Provenance
- Trigger node: manualTrigger initiates the workflow on demand.
- Extraction node: HTML node uses CSS selectors to parse Trustpilot review elements.
- Embedding and analysis nodes: OpenAI embeddings node and Langchain information extractor node generate vector data and insights.
FAQ
How is the customer insights automation workflow triggered?
The workflow is triggered manually via a dedicated manual trigger node, allowing controlled execution whenever analysis is required.
Which tools or models does the orchestration pipeline use?
The pipeline integrates HTTP requests for scraping, Qdrant for vector storage and similarity search, OpenAI’s embedding model for vectorization, and a GPT-based language model for summarization and sentiment analysis.
What does the response look like for client consumption?
The response consists of structured JSON with summarized insights, sentiment classification, suggested improvements, and raw review data, exported to Google Sheets for easy access.
Is any data persisted by the workflow?
Review metadata and vector embeddings are persisted temporarily in the Qdrant vector database, while final insights and raw data are exported and stored in Google Sheets. The workflow itself does not persist data internally.
How are errors handled in this integration flow?
Error handling relies on default n8n node behaviors; explicit retry or backoff mechanisms are not configured within this workflow.
Conclusion
This customer insights automation workflow provides a systematic approach to extract, cluster, and summarize Trustpilot reviews for any specified company using no-code integration. It delivers structured sentiment analysis and actionable feedback by combining web scraping, vector embeddings, clustering, and large language model inference. While it depends on external API availability and the stability of Trustpilot’s HTML structure, it enables repeatable, scalable analysis with minimal manual intervention. The workflow exports comprehensive insights to Google Sheets, facilitating ongoing review monitoring and decision-making.
Reviews
There are no reviews yet.