Order Cancellation Analytics — Chi-Square, Risk Segmentation & Pareto ROI in Python

A end-to-end business analytics project investigating order cancellation drivers, customer risk segmentation, and ROI-driven intervention recommendations for an electronics retailer.

---

🔍 Project Overview

This project analyses 20,000 electronic sales transactions (September 2023 – September 2024) to answer three business-critical research questions:

#	Research Question	Method
RQ1	What product, shipping, and price characteristics are associated with higher cancellation rates?	Chi-square tests, Mann-Whitney U
RQ2	Which customer segments face the highest cancellation risk, and what is the financial impact?	Risk matrix heatmaps, revenue quantification
RQ3	What operational interventions could reduce cancellations, and what is the projected ROI?	Pareto analysis, ROI modelling

Key finding: Cancellation is not a transactional phenomenon — it is a behavioural one. No product, price, or shipping variable significantly predicts cancellation (all p > 0.05). Customer segmentation reveals that loyalty members aged 36–50 and young male customers (18–25) are the highest-risk segments, and a four-intervention portfolio could recover an estimated $4.8M in annual revenue at a combined cost of $30,000.

---

📁 Repository Structure

sales_analysis/
│
├── analysis_2.py        # RQ1 — Chi-square tests + Figures 1–3
├── analysis_3.py        # RQ2 — Segmentation heatmaps + Figures 4–5
├── analysis_4.py        # RQ3 — Pareto analysis + ROI projection + Figures 6–7
├── tables.py            # Table 1 & Table 2 generation
│
├── Figure1_RQ1_cancellation_by_variable.png
├── Figure2_RQ1_payment_method.png
├── Figure3_RQ1_monthly_trend.png
├── Figure4_RQ2_risk_matrices.png
├── Figure5_RQ2_revenue_lost.png
├── Figure6_RQ3_pareto.png
├── Figure7_RQ3_roi.png
│
├── Table1_RQ1_results.png
├── Table2_RQ3_interventions.png
│
└── Individual Assignment Data File Electronic_sales.xlsx

---

📈 Key Visualisations

RQ1 — Cancellation Rate by Transaction Variable

RQ2 — Customer Segment Risk Matrices

RQ3 — Pareto Analysis & ROI Projection

---

🛠️ Tech Stack

Tool	Usage
Python 3	Core analysis language
pandas	Data cleaning, transformation, pivot tables
scipy	Chi-square tests, Mann-Whitney U test
matplotlib	All charts and visualisations
seaborn	Heatmap risk matrices
numpy	Cramér's V calculation, array operations

---

🔬 Methodology

Data Preparation

• Standardised inconsistent categorical values (PayPal capitalisation)
• Modal imputation for single missing Gender value
• Created four derived variables: Cancelled (binary), Has Add-on (binary), Age Group (binned), Month (extracted)
• Created Price Band variable for chi-square compatibility

RQ1 — Chi-Square Analysis

• Tested 9 categorical variables + 1 continuous variable (Mann-Whitney U)
• Reported χ², degrees of freedom, p-value, and Cramér's V for each
• Result: No significant transaction-level predictor found (all p > 0.05)
• Payment Method closest to significance (p=0.077, V=0.021) — negligible practical effect

RQ2 — Customer Segmentation

• Built pivot table risk matrices: Age Group × Loyalty Member, Age Group × Gender
• Quantified revenue loss per segment using Order Value = Total Price + Add-on Total
• Key finding: 36–50 loyalty members cancel at 34.1% — higher than non-loyalty peers (32.8%)
• Non-loyalty segments aged 51+ represent over $12M in combined revenue loss

RQ3 — Pareto Analysis & ROI Modelling

• Proposed four data-driven interventions targeting inferred cancellation drivers
• Applied conservative recovery rate assumptions (15–25%)
• Result: Pre-cancellation popup alone recovers 59.7% of addressable revenue
• All four interventions ROI-positive even at 10× assumed costs

---

📊 Results Summary

RQ1 — No Transaction-Level Predictors Found

Variable	χ²	p-value	Cramér's V	Significant
Product Type	0.87	0.928	0.007	No
Shipping Type	3.50	0.478	0.013	No
Payment Method	8.42	0.077	0.021	Marginal
Price Band	1.47	0.833	0.009	No
Has Add-on	0.36	0.547	0.004	No
Quantity	6.32	0.707	0.018	No

RQ3 — ROI Projection

Intervention	Cost	Recoverable Revenue	ROI
Pre-cancellation Popup	$5,000	$3,207,353	64,047%
Targeted Discount	$25,000	$1,608,981	6,336%
Young Male Nudge	$10,000	$256,716	2,467%
Loyalty Re-engagement	$15,000	$295,076	1,867%

---

💡 Business Recommendations

1. Implement a pre-cancellation confirmation popup — lowest cost ($5,000), highest return ($3.2M recoverable), targets the behavioural root cause
2. Launch targeted retention discounts for Non-Loyalty customers aged 51–65 and 65+ — highest absolute revenue at risk ($8M+)
3. Investigate loyalty programme effectiveness for 36–50 age group — loyalty is not functioning as a retention mechanism for this segment
4. Deploy young male engagement nudges for 18–25 male customers — largest gender-based cancellation gap in the dataset

---

👤 Author

Sundus Afreen MSc Business Analytics — BU7160 Individual Assignment Tools: Python · pandas · scipy · matplotlib · seaborn · Statistical Analysis · Pareto Analysis

---

📄 Licence

This project is for academic and portfolio purposes. The dataset is adapted from the Electronic Sales Dataset publicly available on Kaggle.