On financial modelling, credit scoring, and risk assessment
summary
Explains how lenders turn social, telecom and transaction traces into usable scores: extract ~10,000 features, reduce to ~100–200, train models with bank-backed loan outcomes and backtest, retrain frequently (Simple reports retraining twice monthly) and use alternative data ensembles to cover missing inputs. Concrete anchors: FICO scale 300–900 and default defined at 90 days. Deployments run in Vietnam/Indonesia/India, feed banks or run proprietary lending platforms; issues are regulatory fairness, adversarial fraud and calibration.
AI-written summary · about
chapters (21)
- 0:00Introduction
- 1:57What is a credit score
- 3:13What does a score really mean
- 4:54What is a simple score
- 8:52Student loans in Australia
- 11:50Data sources
- 19:02Alternative models
- 25:55Academic studies
- 28:19Highlevel point
- 29:23Affordability
- 37:54Missing data
- 42:25Prepayment risk
- 43:34Which factors affect scoring
- 45:16Overfitting
- 46:23Features will change
- 49:39Regulatory position
- 50:44India
- 52:49Credit bumps
- 54:10Typical credits
- 55:11Models
- 55:50Interpretation
related talks
58mNew-type Scoring Models: Analyzing Customer Actions
13mMLOps in FinTech: Architecting Responsible AI
02h 01mFostering Trust in AI-Driven Finance
50mData Science in a modern financial institution
17mUsing big data to inspire consumer confidence
25mAnalyzing Daily Bank Transactions: Maximizing Insights
30mBuilding fairer models for finance
19mThe fintech revolution: what's next?