A full end-to-end retail analytics project simulating a professional data consultancy engagement for Quantium — one of Australia's leading analytics firms.
This project was completed as part of the Quantium Data Analytics Virtual Experience Programme during my MSc at the University of Birmingham, where I led a team of 7 members through a full retail analytics workflow — from raw data ingestion to strategic client recommendation.
The analysis covers 264,836 transaction records across 271 stores, 71,287 customers and 20 chip brands over the financial year July 2018 to June 2019.
The Category Manager for chips at a major Australian supermarket chain needed to:
- Understand which customer segments were driving chip sales
- Identify what was driving spend within each segment
- Assess whether a new in-store chip layout trial was effective
- Receive a data-driven recommendation for the next half-year category strategy
| Notebook | Description | Key outputs |
|---|---|---|
| 01 — Data Cleaning | Load, validate and clean raw transaction and customer data | Clean merged dataset, audit log |
| 02 — Customer Analysis | Segment metrics, visualisations, brand and pack size analysis | 8 charts, segment summary CSV |
| 03 — Trial Store Analysis | Control store selection, uplift testing, statistical significance | Trial results CSV, recommendation |
- Older Families — Budget is the highest revenue segment at $156,864 (8.7% of total)
- Mainstream Young Singles/Couples pay the highest unit price at $4.07 — a classic impulse buying profile
- Mainstream Young Singles/Couples are 23% more likely to purchase Tyrrells chips vs the broader population
- Family segments drive revenue through frequency (4.4–4.7 trips/year) while singles drive it through unit price
- Kettle dominates with $390,240 — 21% of total category revenue
- 175g is the #1 pack size at $485,000 — 26% of all chip sales
- Bottom 6 brands contribute less than 6% combined — ranging review recommended
| Store | Control | Sales Uplift | Customer Uplift | Significant | Verdict |
|---|---|---|---|---|---|
| 77 | 233 | +27.8% | +25.2% | No (p=0.13) | Extend trial |
| 86 | 155 | +11.6% | +12.7% | Yes (p=0.012) | Roll out |
| 88 | 91 | +8.6% | +7.2% | No (p=0.24) | Re-trial |
The new layout consistently drives more customers into the chip category — not more spend per visit.
Four prioritised actions for the Category Manager:
- Off-locate Tyrrells and small packs in discretionary space near Young Singles/Couples zones — capitalise on impulse buying behaviour
- Roll out the new layout to stores similar to Store 86 — the only statistically confirmed trial success
- Protect Kettle and 175g shelf space — together they represent the backbone of category revenue
- Target Older Families with volume promotions — highest frequency, highest total spend segment
| Chart | Preview |
|---|---|
| Segment sales heatmap |  |
| Monthly sales trend |  |
| Brand revenue ranking |  |
| Trial store uplift |  |
| Category | Tools used |
|---|---|
| Language | Python 3.10 |
| Data manipulation | pandas, numpy |
| Visualisation | matplotlib, seaborn |
| Statistical testing | scipy.stats (Pearson correlation, t-test) |
| Notebook environment | Google Colab |
| Dashboard | Tableau Public |
| Presentation | PowerPoint (PptxGenJS) |
| Version control | Git / GitHub |
- Data cleaning — outlier removal, date format correction, regex feature extraction
- Feature engineering — brand name and pack size derived from product name string
- Customer segmentation — 21 segments across lifestage × premium tier
- Metric definition — total sales, avg spend per customer, purchase frequency, avg unit price
- Control store selection — Pearson correlation + normalised magnitude distance scoring
- Uplift testing — scaled control comparison + independent samples t-test
- Executive reporting — Pyramid Principle framework
Full data transformation log available in docs/data_audit_log.md
| Phase | Action | Rows before | Rows after |
|---|---|---|---|
| Raw load | Transaction file loaded | — | 264,836 |
| Phase 2 | Non-chip products removed | 264,836 | 264,834 |
| Phase 2 | Outlier customer removed | 264,834 | 246,740 |
| Phase 4 | Merged with customer data | 246,740 | 246,739 |
| Phase 4 | Unmatched row dropped | 246,739 | 246,739 |
- Open each notebook link above
- Click Open in Colab
- Upload the data files when prompted
- Run all cells in order
# Clone the repository
git clone https://github.com/YOURUSERNAME/quantium-retail-analytics.git
cd quantium-retail-analytics
# Install dependencies
pip install -r requirements.txt
# Open notebooks
jupyter notebook notebooks/| Detail | Information |
|---|---|
| Programme | Quantium Data Analytics Virtual Experience |
| Institution | University of Birmingham — MSc |
| Team size | 7 members |
| Role | Project Lead & Technical Lead |
| Timeline | 2026 |
| Status | Complete |
quantium-retail-analytics/
│
├── data/
│ ├── README.md
│ └── sample/
│ └── sample_transactions.csv
│
├── notebooks/
│ ├── 01_data_cleaning.ipynb
│ ├── 02_customer_analysis.ipynb
│ └── 03_trial_store_analysis.ipynb
│
├── outputs/
│ ├── charts/
│ │ ├── chart1_sales_heatmap.png
│ │ ├── chart2_sales_by_segment.png
│ │ ├── chart3_avg_spend.png
│ │ ├── chart4_purchase_frequency.png
│ │ ├── chart5_monthly_trend.png
│ │ ├── chart6_brand_sales.png
│ │ ├── chart7_pack_size_sales.png
│ │ ├── chart8_avg_unit_price.png
│ │ ├── uplift_summary.png
│ │ └── final_recommendation_chart.png
│ │
│ ├── report/
│ │ └── Quantium_Chip_Category_Report.pdf
│ │
│ └── data/
│ ├── quantium_segment_summary.csv
│ └── quantium_trial_results.csv
│
├── docs/
│ ├── data_audit_log.md
│ └── pyramid_principle_report.md
│
├── README.md
├── requirements.txt
└── .gitignore
** Amay Sonar**
MSc Student — University of Birmingham
LinkedIn |
Tableau Public |
Email
This project was completed as part of the Quantium Virtual Experience Programme. Raw data files are not included in this repository in accordance with Quantium's data usage terms. A 50-row sample is available in the data/sample/ folder for reference.