I find the number that changes the decision.

Project 1 · SQL

Where does an online retailer's revenue really come from?

• MySQL 8
• Window functions
• CTEs
• Cohorts
• RFM

Question: a UK online gift wholesaler has two years of transactions (1M rows). Is it growing, who is it depending on, and where should marketing spend its time?

Revenue is flat (+2%) and rests on a shrinking base: the top 10% of customers bring in 63% and first-time customers fell 41% year on year.

• 349 high-value customers have gone quiet. They spent £1.07M. Winning back one in five is worth about £100k a year.
• 44% of new customers reorder within 90 days, and 24% never return, so the second order is the moment to act on.
• Cleaning mattered: two cancelled 75,000-unit orders were entry errors that made random products look like best sellers.

Project 2 · Python

Who should the bank call? Targeting a phone campaign

• pandas
• scikit-learn
• Logistic regression
• Gains and lift

Question: a bank's campaign converted 11.7% of 45,211 calls. If it can only call part of its list, who goes first and what does it lose?

Calling the top half of the list in model order keeps 79% of sign-ups and cuts cost per sign-up from EUR 43 to EUR 27.

• Top 10% of scores convert at 42%, 3.6 times the average, on a held-out test set (ROC AUC 0.747).
• Customers called 5+ times took 41% of all calls and gave 7% of sign-ups. I recommend a four-call cap, tested with an A/B split first.
• I left out call duration because it's only known after the call. Most write-ups of this dataset include it and report inflated accuracy.

Project 3 · Power BI

A retail sales dashboard a manager can open every Monday

• Power BI
• DAX
• Star schema
• Time intelligence
• Drill-through

Question: the same retailer's manager needs one place to see revenue against last year, which customers are slipping, and which products get sent back.

• Star schema with one fact table (1M order lines) and date, customer and product dimensions.
• Measures for year-on-year revenue, new vs returning customers, and cancellation rate.
• Figures reconciled against the SQL results from Project 1.

Dashboard in progress