Matt Yu

Problem

A two-sided staffing marketplace loses site trust when last-minute shift cancellations cannot be refilled in time. The task was to find out why some cancellations get recovered and others do not, and turn it into a fix an executive could act on.

Dataset

Around 146,000 shift, booking and cancellation records from a single regional market (2021). Raw data is proprietary and withheld; the repo ships the anonymised report and aggregated model.

Tools Used

Process

• Tiered every cancellation by lead time to measure damage, not just volume
• Tested and rejected the intuitive "remove the repeat offenders" fix
• Segmented by role, price and time of day, controlling for severity
• Traced the failure to a morning shift reopening into an empty overnight market

Key Insights

• Recovery, not cancellation, is the real problem
• Damage concentrates in one worker role and one shift band
• Price is not the lever — recovery is flat across charge bands

Business Recommendations

Offer the unfilled morning shift to the on-site overnight worker at a premium, opt-in and triggered at the end of their shift — a costed, testable intervention worth under 1.2% of market revenue.

What I Learned

How to pick the right denominator, control for confounders, falsify an intuitive hypothesis, and size an intervention with guardrails and explicit kill criteria.

Problem

The business needed to know which customers actually drive revenue, and why others stop buying, so retention and marketing effort could be focused where it pays off.

Dataset

Two years of transactional e-commerce data covering orders, customers and revenue.

Tools Used

Process

• Cleaned and prepared the raw transaction data
• Wrote SQL to calculate lifetime value, repeat rate and churn
• Segmented customers by value and visualised revenue contribution in Excel

Key Insights

• Segmented customers by lifetime value
• Identified high-value customers driving most revenue
• Analysed churn and repeat purchase behaviour

Business Recommendations

Prioritise retention and upselling for high-value segments, and target re-engagement at customers showing early churn signals.

What I Learned

How to turn raw transactional data into meaningful customer segments and translate SQL output into commercial recommendations.

Problem

Understand what drives sales performance, and how customer behaviour and category mix affect revenue across the year.

Dataset

A synthetic e-commerce dataset of 34,500 transaction records.

Tools Used

Process

• Cleaned and prepared the data in Pandas
• Engineered customer and category-level metrics
• Visualised trends and comparisons with Matplotlib and Seaborn

Key Insights

• Cleaned and prepared 34,500 transaction records for analysis
• Compared one-time and repeat customers to understand revenue contribution
• Analysed category revenue, profit margin, return rates and monthly trends

Business Recommendations

Focus on nurturing repeat customers and the strongest categories, and review categories with high return rates that erode margin.

What I Learned

Strengthened my Python data-wrangling and visualisation skills, and how to structure an analysis end-to-end.

Problem

Sales data sat in raw tables, making it hard to see what was actually profitable and how discounting was affecting margin.

Dataset

The Superstore retail sales dataset covering orders, products, regions and profit.

Tools Used

Process

• Cleaned and structured the raw sales data
• Built PivotTables and formulas to summarise sales and profit
• Designed an interactive dashboard with slicers and data validation

Key Insights

• Compared sales and profit trends
• Identified high and low-performing product categories
• Analysed how discounting affected profitability

Business Recommendations

Reduce heavy discounting on low-margin categories and concentrate effort on consistently profitable products and regions.

What I Learned

How to design a clear, interactive Excel dashboard and communicate profitability in a way stakeholders can act on.

Problem

Spreadsheet-based reporting didn't scale and didn't allow fast, interactive review of sales and profit performance.

Dataset

The Superstore retail sales dataset covering sales, profit, discounts and regional performance.

Tools Used

Process

• Modelled the data with a dedicated date table
• Wrote DAX measures for sales, profit, margin and shipping
• Built KPI cards, slicers, drill-through pages and conditional formatting

Key Insights

• Built KPI cards and slicers so users could quickly review performance and filter the report.
• Created a date table and DAX measures for sales, profit, margin and shipping analysis.
• Added drill-through pages, tooltips and conditional formatting to make weak-margin areas easier to spot.

Business Recommendations

Use the dashboard for regular performance reviews and act on the weak-margin areas highlighted by conditional formatting.

What I Learned

Data modelling and DAX, and how to build a scalable, interactive BI report from spreadsheet data.

Problem

Identify the patterns that separate fraudulent transactions from legitimate ones, to support monitoring and prioritise investigation.

Dataset

A digital transaction dataset with payment channel, device type and behavioural fields.

Tools Used

Process

• Explored and aggregated transactions in SQL
• Compared fraud rates by payment channel and device type
• Analysed transaction velocity across legitimate and fraudulent activity

Key Insights

• Analysed fraud by payment channel and device type
• Identified customers and merchants linked to higher fraud activity
• Compared transaction velocity between legitimate and fraudulent activity

Business Recommendations

Prioritise monitoring on the highest-risk channels and devices, and flag high-velocity accounts for review.

What I Learned

How to use SQL for risk analysis and pattern detection, and how to frame findings for a fraud-monitoring audience.

Problem

Understand housing affordability across London and what drives property value from borough to borough.

Dataset

London housing data covering property prices, salaries, crime and housing demand by borough.

Tools Used

Process

• Imported and cleaned the data with Power Query
• Modelled the relationships with Power Pivot
• Visualised prices, affordability and crime trends

Key Insights

• Compared property prices against salary trends
• Analysed housing affordability by borough
• Explored the relationship between crime and property value

Business Recommendations

Highlight the boroughs offering the best affordability and factor crime levels into value assessments for buyers and investors.

What I Learned

Power Query and Power Pivot for larger datasets, and how to present complex market data in a clear, decision-ready format.