Quick Commerce Analytics - Blinkit ML Project

Developed as part of the EQ-REV's recruitment process

---

Project Overview

This project analyzes Blinkit quick commerce data (August – October 2025) to optimize inventory and business decisions. It focuses on recommending stock quantities, identifying products that need attention, and analyzing root causes of sales changes.

---

Summary of Work

• Built a system to predict restock quantities and maintain optimal inventory levels.
• Designed a framework to flag products needing attention with clear actions and priorities.
• Performed root cause analysis on sales fluctuations and developed a Q&A tool for business insights.

---

Dataset

• Data Period: August 1 – October 31, 2025
• Rows: 56,376
• Columns: 36
• Granularity: Product × City × Day

Includes stock, pricing, sales, and purchase order data.

---

Tasks & Implementation

Task 1: Restock Quantity Prediction

• Predicted exact units to restock for each product-city-day.
• Used stock levels, sales trends, and lead time.
• Output: recommended_restock_qty for test period.

Task 2: Action Suggestions

• Flagged products needing attention.
• Assigned actions, reasons, and priority levels.
• Helped brand managers quickly identify operational issues.

Task 3: Root Cause Analysis

• Analyzed sales spikes/drops using multiple signals.
• Identified causes such as stock issues, pricing gaps, and demand changes.
• Built a Q&A tool for business queries using data + LLM.

---

Data Validation & Preprocessing

• Missing values handled:
  • rating_stars → median imputation
  • benchmark_selling_price → product-level median / MRP
• No duplicate rows found
• PO and darkstore totals validated
• DOH inconsistencies (~56%) resolved by recalculating using: