Junlong Li

Background
Collaborated with the Quantium Retail Analytics team to support a major supermarket chain’s chips category manager, delivering consumer insights and evaluating in-store layout experiments to inform strategic planning and promotional decisions.

Consumer Analysis
Cleaned and analyzed large-scale transaction datasets to identify core consumer segments; discovered that budget-conscious senior households were the primary buyers, with strong preferences for the Kettle brand and 175g pack size—providing data-driven direction for targeted marketing.

Controlled Experiment
Used Pearson correlation and Magnitude Distance methods to successfully match optimal control stores (233, 155, 237) for treatment stores (77, 86, 88). Evaluated experimental performance and confirmed overall positive impact of the layout change, with Store 86 showing the strongest uplift.

Outcome
Developed an executive-ready presentation using the Pyramid Principle, recommending (1) marketing strategies centered on budget-oriented senior households, and (2) scaling Store 86’s layout configuration across other stores based on its superior experimental results.

Background
Analyzed 120 years of Australian Open records to uncover long-term trends across nationality, gender, match outcomes, time patterns, and top-player performance.

Task
Responsible for preparing the dataset, creating analytical metrics, and developing Tableau dashboards to identify competitive patterns and performance insights for tennis audiences.

Action
Cleaned and enhanced the dataset with custom fields (Win Ratio, Game Wins/Losses), built 12+ visuals including maps, treemaps, parallel coordinates, scatter plots, and time-series views to support multi-dimensional comparison.

Result
Identified five major insights: Australia/USA historical dominance, rising competitiveness in men’s finals, strong seed predictability in women’s finals, Djokovic’s modern-era consistency, and Federer’s rare low-seed (17) championship; improved clarity and analytical value of the full report.

Background
Developed a customer segmentation model using e-commerce transaction data to classify customers and predict the products that new customers are likely to purchase.

Data Pre-processing
Handled missing and duplicate records, analyzed cancelled orders and special product cases, and produced a fully cleaned dataset to ensure the accuracy of downstream analytics.

Clustering Analysis
Applied text processing (NLTK) to tokenize product descriptions and remove stop words, then constructed a product–keyword binary matrix. Performed K-means clustering on products, merged product clusters into customer order data, engineered features such as purchase price and purchase date, applied PCA for dimensionality reduction, and executed K-means clustering again to segment customers.

Project Outcome
Built multiple machine learning models to predict customer segments, and implemented a soft-voting ensemble achieving 0.93 precision, significantly improving classification performance and enabling both effective customer segmentation and accurate product purchase prediction for new users.