Learn how to automate the sentiment analysis process of customer reviews by combining Make.com, Airtable, and AI models like GPT-4. This system enables continuous feedback monitoring, immediate response to issues, and effective online reputation management.
Client and Business Context
The project was implemented for a medium-sized company in the diet catering industry (box diets). The company was growing dynamically but lacked an automated way to track and respond to the increasing number of customer reviews online.
Challenge: Effective Monitoring and Response to Feedback
The main challenge was to create a system for continuous monitoring of online reviews appearing across various channels (social media, review portals, website forms). It was crucial to quickly identify problems (e.g., regarding meal quality, delivery punctuality, customer service), understand overall satisfaction, and leverage positive feedback. Manual tracking of reviews was time-consuming and did not allow for a quick response.
- Need to aggregate reviews from multiple online sources.
- Necessity of automatic classification of review sentiment (positive, negative, neutral).
- Requirement for quick notification of relevant departments about problems or praise.
- Desire to use data to improve service quality and marketing efforts.
- Limited human resources for manual monitoring.
Solution: Automation with Airtable, Make.com, and AI
An automated system based on the Make.com platform and the Airtable database was implemented. The system aggregates reviews from selected sources, performs sentiment analysis using OpenAI models from the GPT-4 family (called via API), and then classifies and saves the results in Airtable. Depending on the detected sentiment, the system automatically initiates appropriate actions (e.g., alerts, task creation). The solution was also successfully tested with other language models like Gemini and Claude.
Technologies Used:
- Make.com (formerly Integromat): An automation platform serving as the "brain" of the operation. It is responsible for:
- Integration with data sources (social media APIs, Google Reviews, forms).
- Process orchestration: fetching reviews, sending for analysis, saving results, triggering actions.
- Invoking sentiment analysis: Passing the review text to the OpenAI models API (GPT-4).
- Conditional logic defining actions based on the analysis result.
- Airtable: A flexible cloud database serving as:
- A central database of reviews (content, source, date, author, sentiment, keywords, status).
- An interface for data management and analysis for teams.
- A dashboard for visualizing sentiment trends.
- OpenAI Models (GPT-4) and a Properly Prepared Prompt: Sentiment analysis is performed by calling the GPT-4 models' API with a precisely constructed prompt that instructs the model on how to classify the emotional tone of the text (positive, negative, neutral) and identify key topics or words in the review.
Data Flow Process:
- Monitoring Sources: Make.com scenarios periodically check the sources (social media, reviews, forms).
- Data Retrieval: Make.com retrieves the new review content and metadata.
- Sentiment Analysis: Make.com sends the review content along with the appropriate prompt to the OpenAI API (GPT-4).
- Receiving Results: The OpenAI API returns the result (e.g., 'Negative', 'Positive', 'Neutral') and possibly identified keywords/topics.
- Saving in Airtable: Make.com creates a new record in the "Customer Reviews" base with all the data.
- Triggering Conditional Actions: Based on the sentiment, Make.com initiates further actions.
Response Method Based on Detected Sentiment:
- Negative Sentiment:
- Immediate alert (Slack/e-mail) to customer service/quality manager.
- Automatic creation of a high-priority task (e.g., in Airtable or Trello).
- Marking the record in Airtable as requiring intervention.
- Positive Sentiment:
- Notification to the marketing team (Slack/e-mail) for inspiration or social proof.
- Marking the record in Airtable as 'Positive' (for easy filtering of testimonials).
- Optionally: Flagging to send a thank-you message.
- Neutral Sentiment:
- Archiving in Airtable for trend and context analysis.
- Optionally: Identifying questions/suggestions and flagging them for review.
Key Benefits for the Client
- Speed of Response: Instantly identifying and reacting to negative reviews, minimizing crises.
- Operational Efficiency: Automation of monitoring and classification, saving employee time.
- Better Customer Understanding: Easy analysis of trends and issues thanks to data centralization in Airtable.
- Marketing Support: Quick access to positive reviews and testimonials.
- Quality Improvement: Systematic feedback supporting service enhancement.
- Scalability: The ability to easily expand with new sources and features.
Frequently Asked Questions (FAQ)
Thanks to the flexibility of Make.com, the system can be integrated with many sources that have an API or data export capability. Typical sources include: Facebook business pages, Instagram profiles, Google My Business reviews, Trustpilot, Capterra, contact forms on a website, dedicated email addresses for collecting feedback, and even internet forums or discussion groups (depending on API availability).
Models from the GPT-4 family offer very high accuracy in sentiment analysis, including in Polish. The key, however, is a well-formulated prompt that instructs the model. Despite high accuracy, models can still struggle with linguistic nuances like sarcasm or irony. The system allows for easy review of results in Airtable and potential manual corrections or future prompt adjustments.
Yes, OpenAI's GPT-4 models handle the processing and analysis of the Polish language perfectly. There is no need for additional translation tools before the analysis.
Airtable offers very flexible data visualization and analysis capabilities. You can create various views (e.g., grid, calendar, kanban, gallery), filter and sort reviews by sentiment, source, date, or keywords. Built-in Airtable Apps allow you to create charts (e.g., pie, bar) showing the distribution of sentiment over time, the most common topics of negative reviews, etc. Data can also be easily exported to other analytical tools.
Yes, this is one of the main advantages of using Make.com. This platform has hundreds of ready-made connectors to popular CRM (e.g., HubSpot, Salesforce, Pipedrive) and Helpdesk (e.g., Zendesk, Jira Service Management) systems. We can configure scenarios that, for example, automatically create a ticket in the Helpdesk system based on a negative review or update a customer card in the CRM with information about their latest feedback.
The implementation cost depends on the number of sources to integrate and the complexity of the required response logic. Maintenance costs include subscriptions for Make.com and Airtable (both offer free and paid plans) and OpenAI API costs, which are billed based on the number of processed tokens (usually very competitively priced). The total cost is often more affordable than dedicated media monitoring platforms or the cost of manual labor.
Yes. A properly constructed prompt for the GPT-4 model can not only classify sentiment but also identify and extract key topics, keywords, or specific reported issues (e.g., "meal quality," "delivery," "service"). This information can be saved in separate fields in Airtable, which facilitates filtering, analysis, and reporting of specific types of problems.