As part of a real business project in the media and journalism industry, I developed an interactive NPS (Net Promoter Score) dashboard in Power BI that integrates both quantitative metrics and qualitative customer feedback.
The goal was to analyze reader feedback on digital news content and platform experience - a critical factor in understanding audience satisfaction and trust within the publishing sector.
Since the underlying data was spread across multiple Excel files, I built a Python-based preprocessing pipeline to:
- Automatically read and merge all Excel datasets
- Clean and standardize free-text comments
- Use OpenAI’s GPT-4 model to classify open responses into thematic categories
(e.g., Cost, App performance, Journalistic quality, etc.)
The dataset consisted of German-language customer feedback, making this a multilingual NLP use case.
GPT-4 successfully handled the zero-shot classification task without translation, demonstrating strong multilingual generalization capabilities even in non-English contexts.
This automated text classification enabled direct filtering of qualitative feedback by topic within the Power BI dashboard, linking it to quantitative KPIs such as NPS scores.
As a result, the dashboard provides actionable insights into reader satisfaction, content perception, recurring issues, and key improvement areas - without the need for manual comment coding.
The Power BI dashboard includes several key metrics to monitor customer sentiment and operational performance:
- Net Promoter Score (NPS) – overall customer loyalty and satisfaction
- Response Rate – share of users providing feedback
- Topic Frequency – most mentioned categories from comment classification
- Sentiment Trends – evolution of promoter, passive, and detractor shares
- Category-based NPS – breakdown of NPS by main feedback themes
- Temporal Insights – Time-based performance comparison
These KPIs allow management to quickly identify areas for improvement and track the impact of implemented changes.
The project initially included a sentiment analysis step to identify whether user comments were positive, neutral, or negative.
However, since the NPS score (0–10) already reflects the respondent’s satisfaction level, an additional sentiment layer would have been redundant.
Instead, the focus was shifted to a zero-shot topic classification approach using GPT-4.
This method complements the NPS data by revealing what users are talking about rather than how they feel - uncovering the main drivers behind positive or negative feedback (e.g., price, app usability, journalistic quality, etc.).
Additionally, the zero-shot approach allowed each comment to be assigned to multiple relevant categories simultaneously, reflecting the multidimensional nature of real customer feedback.For example, a single comment could relate both to Cost and App Usability, which would be lost in single-label classification methods.
- Python → Data preprocessing, merging, GPT-4 text classification
- Power BI → Data modeling, visualization, KPI tracking
- Excel / CSV data sources
- Zero-shot text classification with GPT-4
A fully functional NPS insights dashboard that combines:
- Structured NPS and score data, and
- Automatically categorized qualitative feedback
to enable data-driven decision-making in customer experience management.
Here’s a preview of the NPS dashboard in Power BI:
Note: The screenshots are intentionally blurred to protect confidential business data and comply with data privacy requirements.
The image above shows how the classified comment data, NPS, and other KPIs integrate visually in the dashboard.