Crisis Recovery Analytics – QuickBite Express

📌 Project Overview

This repository contains a comprehensive analytics solution developed to diagnose and address the operational and reputational fallout from a dual-crisis event at QuickBite Express (food safety incident + delivery outage). The project integrates customer segmentation, sentiment modelling, operational diagnostics, and incentive-based recovery simulations to guide strategic decision-making.

🧠 Objectives

• Quantify customer churn and behavioural shifts across Recency, Frequency, and Monetary (RFM) dimensions
• Identify operational breakdowns in SLA compliance, delivery delays, and cancellation patterns
• Model return probability for lapsed users based on incentive type and behavioural profile
• Simulate ROI across recovery strategies segmented by customer archetypes
• Benchmark Customer Acquisition Cost (CAC) against competitors (Swiggy, Zomato)

🧰 Tools & Technologies

• Python: Data preprocessing, modelling, and visualisation
• Pandas, NumPy: Data wrangling and feature engineering
• Matplotlib, Seaborn: Visual storytelling and trend analysis
• Scikit-learn: Predictive modelling and probability estimation
• NLTK: Sentiment extraction and keyword parsing

📊 Key Deliverables

• RFM segmentation and churn diagnostics
• Monthly ratings and sentiment trend analysis
• SLA and cancellation heatmaps
• Restaurant-level performance decline
• Return probability model with incentive mapping
• ROI simulation by customer segment and strategy
• CAC benchmarking dashboard

📈 Methodology Highlights

• RFM Scoring: Used to segment customers based on pre-crisis engagement
• Sentiment Modelling: Monthly score trends extracted from review text using NLP
• Return Probability Estimation: Logistic regression and decision trees applied to predict reactivation likelihood
• ROI Simulation: Incentive cost vs reactivation yield modelled across archetypes
• Benchmarking: CAC multipliers derived from simulated ad inflation, seasonal demand, and saturation factors

📌 Limitations

• Post-crisis data beyond September 2025 not included
• Sentiment analysis is limited by unstructured or missing review text
• ROI estimates based on simulated incentive costs and modelled probabilities
• SLA attribution not vendor-specific due to data granularity constraints

📣 Author

Bitan Sarkar MSc Business Analytics, University of Bristol Data Analyst | Strategic Storyteller | Dashboard Designer