| --- |
| license: mit |
| language: |
| - en |
| tags: |
| - sales |
| - fashion |
| --- |
| # Analyzing Consumption Efficiency Over Time: The Regret Gap and Logistical Burden in Global E-commerce |
|
|
| ## Project Presentation |
| <div align="center"> |
| <h1>Video Presentation</h1> |
| <video controls width="100%"> |
| <source src="https://huggingface.co/datasets/rubinshahaf/globalconsumptioneffectiveness/resolve/main/shahaf.f.mp4" type="video/mp4"> |
| Your browser does not support the video tag. |
| </video> |
| </div> |
| |
|
|
| ## Research Question |
| > **"Analyzing Consumption Efficiency Over Time: What are the peak effective hours for e-commerce activity, and what is the logistical cost of high-risk nocturnal purchases?"** |
|
|
| --- |
|
|
| ## Project Overview |
| This project investigates the psychological and logistical boundaries of modern e-commerce. The research utilized a complex relational dataset consisting of 7 different tables. Through a process of strategic selection, the 4 most relevant tables (Orders, Order_Items, Products, and Users) were integrated to create a unified master dataset containing over 181,000 records. |
| |
| The primary objective was to identify the "Regret Gap": a phenomenon where late-night, impulsive purchases—driven by lowered cognitive barriers—lead to higher return rates and significant logistical inefficiencies. |
| |
| ## Methodology and Data Pipeline |
| To transform raw data into actionable business insights, a rigorous cleaning and normalization process was implemented: |
| |
| 1. Data Integration and Integrity: Joined 4 tables using unique identifiers (Order ID, User ID). |
| 2. Missing Value Imputation: |
| - User Identity Logic: Identified identical User IDs to complete missing demographic details. |
| - Temporal Completion: Derived day of the week and specific dates where timestamp data was incomplete. |
| 3. Local Time Normalization: Converted UTC timestamps into the user's local time based on their geographic coordinates. This step was critical for enabling behavioral analysis relative to the user's actual time of day. |
| 4. Outlier Removal: Filtered the top 2% of price points and removed illogical data points (e.g., ages 0 or 120, and transactions with zero profit). |
| |
| ## Statistical Overview |
| Broad statistical analysis revealed a significant trend: while afternoon hours (14:00-17:00) record the highest volume of "safe" transactions, the post-midnight window (00:00-05:00) accounts for a disproportionate percentage of total return costs. |
| |
| ## Visual Analysis and Findings |
| |
| ### 1. 24-Hour Business Health: Volume vs. Risk |
| - Visualization: A dual-axis chart comparing total order volume against the probability of return over a 24-hour cycle. |
| - Analysis: While sales volume remains relatively steady throughout the day, the probability of risk (cancellations and returns) spikes significantly between 00:00 and 06:00. |
| - Conclusion: Nocturnal sales often represent "illusory revenue"—they appear positive on real-time dashboards but translate into logistical burdens within the following week. |
| |
|  |
| |
| |
| ### 2. Global Order Density: The Biological Clock |
| - Visualization: A geospatial Choropleth map normalized to local time zones. |
| - Analysis: Regardless of geography or specific market, return density correlates with the local "night-time wave." |
| - Conclusion: Purchase efficiency is tied to the human biological clock rather than geography. Night-time regret is a universal behavioral trait. |
| |
|  |
| |
| ### 3. Statistical Validation: Chi-Square Test |
| - Visualization: A statistical summary table/heatmap of the Chi-Square contingency test. |
| - Analysis: Tested the independence between the purchase time window and final order status. |
| - Conclusion: With a p-value < 0.05, the time of day is a statistically significant predictor of order success. This confirms the patterns observed are not coincidental. |
| |
|  |
| |
| ### 4. Behavioral Mirror: Complex Fit vs. One-Size |
| - Visualization: A "Mirror" Lollipop chart comparing different product categories. |
| - Analysis: Categories such as Jeans and Sweaters (Complex Fit) experience a massive spike in returns at night due to sizing errors made under fatigue. Conversely, Accessories (One-Size) remain stable. |
| - Conclusion: Night-time risk is category-dependent. Retailers should avoid promoting products requiring high cognitive effort (such as precise sizing) during low-energy hours. |
| |
|  |
| |
| ### 5. Purchase Price Distribution and the Regret Gap |
| - Visualization: Analysis divided into two parts: Distribution Density and Median Price Trends. |
| - Analysis: Paradoxically, median prices are higher at night, as consumers tend to make more expensive impulse purchases. |
| - Conclusion: This represents the core of the Regret Gap. Consumers commit to expensive items when their cognitive barriers are low, leading to "buyer's remorse" the following morning. |
| |
| **Distribution Density** |
| |
| |
|  |
| |
| ## Strategic Business Conclusions |
| - Advertising Optimization: Reallocate marketing budgets away from nocturnal hours for categories requiring complex fit. |
| - Cooling-off Period: Implement a 4-hour processing delay for high-value night orders to allow for morning cancellations before the logistics chain is activated. |
| - Dynamic User Interface: Adjust the digital storefront to feature low-risk items (such as accessories) during late-night hours. |
| |
| ## Future Research Questions |
| - Social Media Influence: How do specific platforms (TikTok vs. Instagram) influence impulse purchasing behavior across different time windows? |
| - Demographic Deep-Dive: Does the Regret Gap vary significantly based on the shopper's age or gender? |
| - AI Intervention: Can real-time AI sizing assistants mitigate the night-time return spike in the apparel category? |
