AI-Driven Retail Insights Dashboard

AI-Driven Retail Insights Dashboard (Power BI + ChatGPT)

1. The Mission (Project Context)

The goal of this project was to solve the 'Interpretation Gap'—the time it takes for an executive to look at a chart and understand the necessary action. By integrating ChatGPT via Python scripts within Power BI, I built a dashboard that not only visualizes retail performance but also generates natural language narratives and sentiment-driven strategy recommendations in real-time.

2. The Interactive Stack (The Visuals)

Screen 1: Overview

Screen 2: Tooltip - Insight 1

Screen 3: Tooltip - Insight 2

Screen 4: Tooltip - Insight 3

Screen 5: Tooltip - Insight 4

Screen 6: Tooltip - State Insight

Screen 7: Tooltip - Category Insight

Screen 8: Tooltip - Sub-Category Insight

Screen 9: Guided Insights (Executive Summary)

Screen 10: Insight Deep Dive (Detailed Analysis)

3. Technical Architecture

AI Integration: Implemented a Python transformation layer using the OpenAI library to send aggregated sales data and customer comments to the GPT-4 API.
Data Modeling: A classic Star Schema optimized for AI-querying. I created a "Summarization Table" that stores pre-processed AI insights to ensure the dashboard remains fast for the end-user.
Geospatial Intelligence: Integrated a custom TopoJSON map (usa.states.topo.json) to correlate AI-detected regional sentiment with physical sales performance across the US.

4. Key Business Insights (“The Discovery”)

Predictive Sentiment: The AI identified that a 10% drop in customer sentiment for the 'Electronics' category was a leading indicator of a sales slump two weeks later—providing a "pre-warning" system that traditional BI would miss.
Automated Root-Cause: While sales were up in the West region, the AI flagged that profit was lagging due to "hidden" shipping costs mentioned in customer feedback regarding late deliveries and damaged goods.
Clustered Topic Analysis: Using AI, I clustered 5,000+ customer comments into three main buckets: Price Sensitivity, Product Quality, and Shipping Reliability, allowing the business to target specific operational fixes.

5. Strategic Recommendations

Sentiment-Based Pricing: Implement a dynamic pricing floor for products with high positive sentiment scores to maximize margin while customer "hype" is at its peak.
Automated Feedback Loop: Use the AI-generated sentiment clusters to automatically route high-priority "Quality" complaints directly to the Manufacturing team, bypassing manual sorting.
AI-Driven Inventory: Reduce stock levels for product codes where the AI-detected sentiment is trending downward, even if current sales are still positive.

6. Let’s Talk About the Data

"The technical challenge was integrating an LLM into the Power BI refresh cycle. Since calling an API for every row is inefficient and costly, I engineered a Grouping & Batching strategy. I wrote a Python script that grouped data by Category and Week before sending it to ChatGPT. This reduced API calls by 95%, ensuring the project stayed within budget while still delivering high-quality, human-like summaries of the business performance."

AI-Driven Retail Insights Dashboard (Power BI + ChatGPT) Build Guide

A streamlined guide for replicating the project:

1. Prepare the Data

Load dataset into Power BI
Create a proper Date table
Build relationships (Date → Orders, Category → Products, etc.)

2. Create Core Measures

Total Sales
Total Profit
Profit Margin %
Category Contribution %
Sub-Category Rank
Unprofitable Product Count
Narrative Insight Measures (State_Message, Category_Message, Sub-Category_Message)