SlideShare a Scribd company logo

Machine learning for retail banking

•
•
Rudradeb Mitra
Rudradeb Mitra

This is a talk given to bankers at CCX Forum where I share how Machine Learning products can be built for retail banking sector, what are the challenges and how can they be overcome.

1 of 41
Download to read offline
Machine Learning for Retail Banking Rudradeb Mitra | Serial entrepreneur, Writer and Mentor (Google Launchpad) http://www.linkedin.com/in/mitrar/ CCX Forum 18th May 2018, London, UK
make you all Machine Learning experts in 30 minutes! My goal is to ..
Unlearn Machine Learning
Real power of Machine Learning is in predicting human behavior
Relearn Machine Learning
"Communication and making people feel valued are the most important for banking CX" Drivers of customer experience in banking sector But how to make someone valued? What to communicate? Machines help us to know the answers!
I. Sceptic's ask Why Machine Learning: Arguing with a puzzle
Puzzle I Imagine you bought a wine for $20. It now sells for $75. If you decide to drink the wine, what cost will you assign to the bottle? 1. $0 2. $20 3. $20 + interest 4. $75 5. $-55 (made profit of $55) Source: Misbehaving: The making of Behavioral Economics by Richard Thaler.
Most people said $0 (30%) or $-55 (25%) Answer Source: Misbehaving: The making of Behavioral Economics by Richard Thaler.
Puzzle II Imagine you bought a wine for $20. It now sells for $75. And you broke the wine bottle, so what cost will you assign to the loss? 1. $0 2. $20 3. $20 + interest 4. $75 5. $-55 (made profit of $55) Source: Misbehaving: The making of Behavioral Economics by Richard Thaler.
Most people said $75 Answer Source: Misbehaving: The making of Behavioral Economics by Richard Thaler.
Normal people do not behave like economical theory.
Machine Learning can model behaviors Machine Learning algorithms can be used to learn patterns from data including human behavior without explicitly being programmed.
II. What problems to solve using Machine Learning?
Problems where Bayes Error rate is >80% and which have high cost
Patterns, Patterns, and more Patterns!
• What we buy, when we buy? • What makes us engaged? • What makes us move to another product? Everywhere....
• Solving problems that were thought unsolvable (For ex, Anticipation of clients needs, Loans to people without bank accounts) • Solving problems that were thought not a problem (For ex, customer acquisition, retention) • Improving upon existing systems (For ex, Increase transparency and frequency of communication, risk assessment) Three groups of problems
III. How to build products using Machine Learning?
Step 1: Intuitive Thinking to decide what data to collect to train your model
Example : Modeling "gamblers"
Puzzle III You won $30. Which of the following you are likely to take? 1. 50% chance of gaining $9 or 50% chance of losing $9 2. No further gain Source: Misbehaving: The making of Behavioral Economics by Richard Thaler.
70% choose option 1 (50% change of win or lose $9) Answer Source: Misbehaving: The making of Behavioral Economics by Richard Thaler.
Puzzle III You lost $30. Which of the following you are likely to take? 1. 33% chance of gaining $30 or 67% chance of nothing 2. Sure $10 Source: Misbehaving: The making of Behavioral Economics by Richard Thaler.
60% choose option 1 (33% to gain $30) Answer Source: Misbehaving: The making of Behavioral Economics by Richard Thaler.
Model If you want to model people's next behavior you need to have data about their past looses or gains.
Step 2: Collect data (GDPR)
For business - Often data is public
B2C - How to make people share their data? (GDPR)
1: Build trust and likeability: Create community and gamification Goals & challenges Rewards
2: Cannot force to adopt and let users be in control vs
3: Do not try to change behaviors https://techcrunch.com/2013/07/13/why-behavior-change-apps-fail-to-change-behavior/
4: Educate your customers/users
Machine learning for retail banking
Step 3: Algorithm
Algorithms • Group people (risk profiling, communication): Classification and Clustering • Individual future behavior (what, when someone will act): word2vec and LSTM
Step 4: Development
Open source Libraries
• Intuitive thinking • Collect data • Select algorithms • Development Summarize - How to build ML products?
Cost of not doing! • Amazon becoming partially a bank! • New 'payment banks' are already in Asia (India, Vietnam). • Existential crisis in era of Internet and Globalization.
Machine Learning is NOT rocket science Adoption How to collect data? Intuitive Thinking Feel free to contact: https://www.linkedin.com/in/mitrar/ mitra.rudradeb@gmail.com As ML experts answer the following Algorithm What algorithm to use? What data to collect?

More Related Content

Machine learning for retail banking

  • 1. Machine Learning for Retail Banking Rudradeb Mitra | Serial entrepreneur, Writer and Mentor (Google Launchpad) http://www.linkedin.com/in/mitrar/ CCX Forum 18th May 2018, London, UK
  • 2. make you all Machine Learning experts in 30 minutes! My goal is to ..
  • 4. Real power of Machine Learning is in predicting human behavior
  • 6. "Communication and making people feel valued are the most important for banking CX" Drivers of customer experience in banking sector But how to make someone valued? What to communicate? Machines help us to know the answers!
  • 7. I. Sceptic's ask Why Machine Learning: Arguing with a puzzle
  • 8. Puzzle I Imagine you bought a wine for $20. It now sells for $75. If you decide to drink the wine, what cost will you assign to the bottle? 1. $0 2. $20 3. $20 + interest 4. $75 5. $-55 (made profit of $55) Source: Misbehaving: The making of Behavioral Economics by Richard Thaler.
  • 9. Most people said $0 (30%) or $-55 (25%) Answer Source: Misbehaving: The making of Behavioral Economics by Richard Thaler.
  • 10. Puzzle II Imagine you bought a wine for $20. It now sells for $75. And you broke the wine bottle, so what cost will you assign to the loss? 1. $0 2. $20 3. $20 + interest 4. $75 5. $-55 (made profit of $55) Source: Misbehaving: The making of Behavioral Economics by Richard Thaler.
  • 11. Most people said $75 Answer Source: Misbehaving: The making of Behavioral Economics by Richard Thaler.
  • 12. Normal people do not behave like economical theory.
  • 13. Machine Learning can model behaviors Machine Learning algorithms can be used to learn patterns from data including human behavior without explicitly being programmed.
  • 14. II. What problems to solve using Machine Learning?
  • 15. Problems where Bayes Error rate is >80% and which have high cost
  • 16. Patterns, Patterns, and more Patterns!
  • 17. • What we buy, when we buy? • What makes us engaged? • What makes us move to another product? Everywhere....
  • 18. • Solving problems that were thought unsolvable (For ex, Anticipation of clients needs, Loans to people without bank accounts) • Solving problems that were thought not a problem (For ex, customer acquisition, retention) • Improving upon existing systems (For ex, Increase transparency and frequency of communication, risk assessment) Three groups of problems
  • 19. III. How to build products using Machine Learning?
  • 20. Step 1: Intuitive Thinking to decide what data to collect to train your model
  • 21. Example : Modeling "gamblers"
  • 22. Puzzle III You won $30. Which of the following you are likely to take? 1. 50% chance of gaining $9 or 50% chance of losing $9 2. No further gain Source: Misbehaving: The making of Behavioral Economics by Richard Thaler.
  • 23. 70% choose option 1 (50% change of win or lose $9) Answer Source: Misbehaving: The making of Behavioral Economics by Richard Thaler.
  • 24. Puzzle III You lost $30. Which of the following you are likely to take? 1. 33% chance of gaining $30 or 67% chance of nothing 2. Sure $10 Source: Misbehaving: The making of Behavioral Economics by Richard Thaler.
  • 25. 60% choose option 1 (33% to gain $30) Answer Source: Misbehaving: The making of Behavioral Economics by Richard Thaler.
  • 26. Model If you want to model people's next behavior you need to have data about their past looses or gains.
  • 27. Step 2: Collect data (GDPR)
  • 28. For business - Often data is public
  • 29. B2C - How to make people share their data? (GDPR)
  • 30. 1: Build trust and likeability: Create community and gamification Goals & challenges Rewards
  • 31. 2: Cannot force to adopt and let users be in control vs
  • 32. 3: Do not try to change behaviors https://techcrunch.com/2013/07/13/why-behavior-change-apps-fail-to-change-behavior/
  • 33. 4: Educate your customers/users
  • 36. Algorithms • Group people (risk profiling, communication): Classification and Clustering • Individual future behavior (what, when someone will act): word2vec and LSTM
  • 39. • Intuitive thinking • Collect data • Select algorithms • Development Summarize - How to build ML products?
  • 40. Cost of not doing! • Amazon becoming partially a bank! • New 'payment banks' are already in Asia (India, Vietnam). • Existential crisis in era of Internet and Globalization.
  • 41. Machine Learning is NOT rocket science Adoption How to collect data? Intuitive Thinking Feel free to contact: https://www.linkedin.com/in/mitrar/ mitra.rudradeb@gmail.com As ML experts answer the following Algorithm What algorithm to use? What data to collect?