What this automation does
This automation collects customer feedback from multiple sources — reviews, NPS surveys, support tickets, social mentions — and runs each piece through AI sentiment analysis. It scores sentiment (positive, neutral, negative), extracts the main topics mentioned, and logs everything to a dashboard.
You get a daily or weekly summary showing sentiment trends, common complaints, and individual customers who are at risk of churning. Instead of reading hundreds of reviews manually, you spend five minutes scanning a dashboard.
Tools you need
- Feedback sources: Google Reviews, Typeform surveys, support tickets, social media
- OpenAI API: For sentiment scoring and topic extraction ($0.005-0.01 per item)
- Make or n8n: Orchestrates the data flow from sources to analysis to dashboard
- Google Sheets or Airtable: Stores results and serves as a simple dashboard
How to set it up
Step 1: Identify your feedback sources. Connect each one to Make or n8n — new Google reviews, new Typeform submissions, new support tickets. Set each trigger to fire on new entries.
Step 2: For each source, add an OpenAI step that analyzes the text. Ask for a JSON response with: sentiment (positive/neutral/negative), score (-10 to +10), main topics (array of 1-3 keywords), and a one-line summary.
Step 3: Log every result to a Google Sheet or Airtable base with columns for date, source, original text, sentiment, score, topics, and summary. Add a chart showing sentiment trend over time.
Step 4: Set up alerts — if sentiment score drops below -5, or if three or more negative reviews come in within 24 hours, send a Slack notification to your team. This catches problems early.
Cost breakdown
| Item | Cost | Notes |
|---|---|---|
| Make or n8n | $10-$25/mo | n8n self-hosted is free |
| OpenAI API | $5-$15/mo | Very low cost per analysis |
| Google Sheets | Free | Or Airtable at $20/mo for better views |
| Total monthly | $15-$40/mo | Replaces manual review reading |
Frequently asked questions
GPT-4 achieves 90-95% accuracy on straightforward sentiment. It occasionally struggles with sarcasm or mixed sentiment ('The food was great but the service was terrible'). For mixed reviews, ask the model to break sentiment down by topic.
Yes. Log every analysis to a spreadsheet with timestamps. Use pivot tables or charts to spot weekly/monthly trends. You'll see patterns like 'shipping complaints spike every Friday' or 'product quality sentiment dropped after the last update.'