AI/ML use cases
BFSI industry overview
Lending Products
Underwriting Strategy
Customer Lifecycle Management
How to prepare for becoming a banking analyst
Materials to study for statistics
What is fintech?
What is a Credit Bureau?
Books for statistics
Tools for data science
Techniques for data science
2. How AI/ML have evolved and the different use of AI technology. The recent trends and industry use
cases
Analytics overview
01
Understand the different financial institutions and the product offerings.
Industry Overview
02
Use of analytics in customer lifecycle management. Understand customer life cycle and decision
points where analytics is used today
Use of Analytics in Banking
03
What to read? What are the tools and techniques you should know to get into banking domain. What
is the expectation from a data scientist?
How to Prepare yourself for Analytics job?
04
Agenda
5. AI Timeline
Chatbot
• Chatbot capable of
conversation with
human being
• SAS institute
Robot
Sony launches AiBO
pet Dog
Initial release Scikit
Learn in 2007
Python
Most used
programming
language
AI
John MaCarthy
introduces the term
Artificial Intelligence
Deep Blue
Deep Blue defeated
great chess champion
Garry Kasparov
Alexa
Amazon launches
Alexa
Alpha Go
20192014-182000-201419971960-901956
Alan Turing Test
If a machine tricks
Human in thinking its
human then machine
has intelligence
1950
6. Analytics - Industry use cases
Retail E- comm Manufacturing BFSIHealthcare
Highly used
• Supply chain analytics
• Segmentation and
customer targeting
• Market basket analysis
• Loyalty programmed
• Dashboarding
• Campaign management
• Churn Analytics
AI & Analytics
• Recommender system
• Next best offer
• Inventory management
• Marketing analytics
• Reporting
• Forecasting
Analytics : Medium
• Clinical trial data analysis
• Disease pattern analysis
• Drug discovery and
development
• Epidemiology modeling
AI: High
• Robotics
• Maintenance
prediction
• Process automation
• Demand forecasting
• Warranty reserve
analytics
AI : Medium
• Credit risk modeling
• Pricing analytics
• Customer service
• Automation
• Alternate data
• New to credit analytics
• Fraud & Collection
8. Type of Financial Institutions
Banks, NBFC’s , MFI’s
Retail banking
Private Banking
Investment Banking
Group Loans
Small Finance Banking
Payment Banks,
All insurance related companies
Health Insurance
General Insurance companies
Insurance brokers
Third party providers
Financial Technology companies
• Technology and platform provider
• Collection house
• Aggregators
• Alternate Data
• Analytics company
• AI companies
All type of Investment companies
• Wealth Management
• Mutual Fund house
• Broker
• Traders
Banks FinTech's
Insurance Investment
9. Banking Products
Only lending products is shown here. Savings account is banking product but not a loan
Loan to buy a house or renovate a
house. The interest rate is lower. LTV
is key term. Foreclosure if the loan is
not paid.
Home Loan/Mortgage
To buy the vehicle. It is a depreciating
asset and hence the risk is more
Car Loan
Low risk product.
Loan against FD
Cash Credit, Line of Credit, Overdraft
etc. Some of the loans can be partially
secured
Business Loan
Loan to meet the personal need such
as marriage, children education, travel
and any other need.
Personal Loan
Plastic with a credit limit. The interest
rate is generally higher
Credit card
Secured Un-secured
Un-secured loan credit risk is higher compared to the Secured loan as there is higher recovery in case of
delinquent and NPA. The interest rate for secured loan is generally lower compared to un-secured loan
10. Customer Life Cycle Management
Targeting
Target the best
customer who are more
probable to take the
products
Portfolio Management
Manage the customers
and loans to reduce the
losses and increase the
revenue
Fraud
Identify and reject the
fraudulent transactions
and application
Origination
Approve the customers
which are more like to
give profit. Process
automation
Collection
Collect the loan amount
from the delinquent
customer without
incurring much expense
Regulatory
Adhere to the regulatory
requirement of loss
provisions, Basel norms
and other Indian
Accounting standard
11. Definition
Credit score is a numerical score range to measure the credit risk of
customers. The higher the score better is the customer.
Features
Techniques used to create and validate credit scoring models include:
• Logistic regression and linear regression
• Machine learning and predictive analytics
• Binning algorithm (i.e., monotone, equal frequency, and equal width)
• Cumulative Accuracy Profile (CAP)
• Receiver Operating Characteristic (ROC)
• Kolmogorov-Smirnov (K-S) statistic
Areas it is used
Credit approval, Cross sell and up sell
What is a Credit Score?
The measure of Credit Worthiness
12. Predictive Model Development
Historic
Vintage
selection
Snapshot
Period
selections
Independent
Variable and
Target
Variable
Variable
Reduction
Binning &
WOE
Model
Fitting,
Validation
Independent Variables:
Demographic Data: Qualification, Income, House owner
Bureau Data: # of delinquency in past 6 month, Total balance,
DBR, Vintage in bureau, Payment%
Dependent Variables:
90+ DPD in 12 month
60+ DPD in 9 month
13. Example1: Underwriting Strategy
Approving the Marginal population which
are profitable for business with higher
interest rate
If the customer doesn’t pay, it is called delinquent customer.
After missing 4 payment , its called 90+DPD.
Bad Good definition
Post manual and STP process
Final Approval
80%
10%
20%
14. Cut-off Strategy
Historic Time
period sample
creation
Target Variable
creation
Parameter
selection along
with score
Strategy
Development/ Back
testing
Apply the
strategy at
current
snapshot
Strategy
Optimization
KRI/KPI
identification
Approve/Decline
Bad Rate <10k 10-50k 50-1lakh >1lakh
<600 20% 17% 13% 8%
600-750 15% 12% 5% 4%
750-850 10% 8% 3% 2%
>800 2% 1% 0.8% 0.2%
15. Propensity Model
Behaviour risk
score
Cross sell
strategy
Example:
1. Selling personal loan to existing credit
card customer
2. Selling a top-up loan to existing Personal
Loan customer
Example2: Cross-sell Upsell Strategy
17. Challenges in adopting AI
https://www.equifax.com/videos/introduction-neurodecision/
Top Reasons:
- ROI
- Lack of expertise
- Black box
- Regulatory
requirement
- Hype vs reality
- Lack of data
- Lack of data engineer
- Use of Jargon
18. Terminology in Lending Business
Approval Rate
STP
Delinquency
NPA/DPD
Roll Rate
Reage
Revenue
NIM/RAM
Vintage Analysis
Foreclosure
Recovery
Application and Behavior score to
predict the future credit risk or
delinquency behavior. Generally a
supervised ML model with
classification technique used
A-score/B-score
A business input based
simulation problem to come up
with the optimum pricing model
Risk Based pricing
To predict the propensity of the
customer to buy or take certain
product. It an again a classification
problem with some time
multinomial logistic model /Neural
network is used
Propensity Model
Business calculation to arrive
at the business value using the
score cut-off and back testing
of the information
Swap-in/Swap-out
Regulatory modeling
categories. The segmentation
methodology is used before
using logistic model to predict
the PD. LGD and EAD uses
regression technique.
PD,LGD,EAD, Basel
Validate the model with respect
the accuracy such as KS, AUC,
Gini, stability such as PSI, CSI
and rank ordering
Model Validation
Banking
Terminology
19. CIR
Stores,
manage and
infer the data
Sell data to
the
participating
institution
needs
Governed by RBI.
Can’t share data
without customer
consent
Credit Bureau is a custodian of credit data.
All the financial institutions share their data
to the credit bureau.
• There are four Credit Bureau in India
• Equifax, Transunion (CIBIL), CRIF,
Experian
• Equifax is the only bureau with Retail,
Commercial, Employment and MFI
bureau
What is a Credit Bureau?
What is a
Credit Bureau?
Making the credit available to deserving
20. Service
provider
Provides
information of
various
products
Generates
Leads
Collection
Customer can
apply for loan,
Insurance and
many other
services
There are many websites which allow an user to
view loan offers from multiple banks and
lenders and also apply for a loan to one of
these lenders. These are known
as loan marketplaces or loan aggregators.
Example:
Paisa bazar ,Credit Mantri ,Bank Bazar , Policy Bazar
What is an account aggregator?
What is an
aggregator?
Serving customer needs real time
21. Problem
statement
Technology is
core Solve
problems
using
analytics.
Financial inclusion
and alternate data
FinTech or financial technology has
emerged as a relatively new industry
in India.[1]
FinTech is an industry comprising
companies that use technology to offer
financial services. These companies
operate in insurance, asset
management and payment, and numerous
other industries.
Fintech:
What is a
Fintech?
Changing face of Finance Industry
22. Thoughts to
grow..
How do you evaluate credit worthiness for a customer who haven’t taken
any loan yet. Specifically customers not having digital presence
25. Techniques used in Banking Analytics
Linear Regression
Logistic Regression
Neural Network
Random Forest
XG Boost
Multinomial Logistic
Regression
Lasso/Ridge Regression
Decision Tree
Supervised
Clustering
K-means Clustering
PCA /Co-relation
Association Analysis
Unsupervised
NLP
Deep Learning
Markov Chain Model
Computer Vision
RPA
Blockchain
AI & Recent technique
Time Series
ARIMA
ARIMAX
Moving Average
Polynomial Smoothing
Exponential Smoothing
27. Type of jobs in Banking Analytics
This is a subjective division of
different roles in Banking
Analytics
Type of jobs
Create models using ML & AI. Good at
coding and statistics. Good at NLP and
data preparation technique.
Data Scientist
Good at coding, data base
management, technical architect. Good
at model deployment and production.
Testing, Cloud knowledge
Data Engineer
Create solution and describe the
problem statements. Good at
understanding analytics technique, use
case and business knowledge. Can
also develop models and strategies
Business Analyst
Tell a descriptive story. Good at coding
and visualization tool
Visualization/Reporting
Good at articulation, understanding
problem and providing solutions. Good in
PPT skill
Consultant
28. Path to Become a Banking Analyst
• Pick one language
among SAS, R or
Python
• Lots of data cleaning,
unstructured data
and
Proficient in
Coding
Proficient in
Business
concepts
Proficient in
Data preparation
Statistics,
Visualization,
Model
Implementation
• Product Knowledge
• Revenue /Loss
matrices
• Terminology
• How to make data ready
for model
• Snapshot period,
Sampling, Bad
definition, Target
variable
• Derived variables,
Monotonicity
• Github
• Model implementation
• API framework
• Scala
• Cloud
• Plumber/Flask
• Types of graphs
• Create story
29. Resources
Statistics Books
Website or Courses02
Python Coding books
Machine Learning Books
Websites
• "Statistics for Machine Learning" by Pratap Dangeti
• Practical Statistics for Data Scientists by Peter Bruce and Andrew Bruce)
• Core Python programming by Dr. R. Nageswara Rao
• Python Data science Handbook by Jake Vanderplas
• Hands-on Machine Learning with Scikit-Learn & TensorFlow by Aurelien
Geron
• Data mining & Predictive Analytics by Daniel T. Larose, Chantal D. Larose
• Github, Kaggle, Towards Data Science, Analytics Vidya,Academia