Public projects · Banking · Credit risk
The Lending Machine
Is Buy Now, Pay Later really riskier than a bank loan?
Personal project on public data. Not client work. An independent analysis, not linked to any company mentioned.
307,511loan applications analysed
1.00odds ratio at the same loan size
17report pages
The data
307,511 loan applications and 767,000 past loans from an anonymised public lending dataset; the AML part uses simulated transactions.
How it was built
Power BI: 17 pages, more than 160 DAX measures, a credit scorecard, a cut-off simulator, IFRS 9, collections, a Customer 360 page and a full Arabic version.
What I found
- At first BNPL-style loans looked riskier: 2.42% went 30+ days late, against 1.49% for cash loans.
- Compared at the same loan size, the gap disappears (odds ratio 1.00). The risk is in the first small loan to a new customer, which is mostly what BNPL sells.
- 88% of accounts that reached 30 days late showed a warning signal about two months earlier.
- A smarter AML alert rule caught real cases at 76% precision, against 218 false alerts per case for a simple big-transfer rule.
How I tested it
10 main results were tested: 5 held, 2 only partly, 2 need care, and 1 broke. The one that broke was my first headline, and it stays in the report because that test led to the real finding.
Screens
Next project
The Quiet Freeze
Want a report like this on your own data?
Get in touch