Analytics for Banks
November 11, 2016
Outline
About AlgoAnalytics
Problems we can solve for banks
Our experience
Technology
Page 2 AlgoAnalytics All rights reserved
About AlgoAnalytics
Analytics Consultancy
Work at the intersection of mathematics and other domains
Harness data to provide insight and solutions to our clients
Led by Aniruddha Pant
+30 data scientists with experience in mathematics and engineering
Team strengths include ability to deal with structured/ unstructured data, classical
ML as well as deep learning using cutting edge methodologies
Expertise in Mathematics and Computer Science
Develop advanced mathematical models or solutions for a wide range of industries:
Retail, eco o ics, healthcare, BFSI, teleco ,
Work with Domain Specialists
Work closely with domain experts either from the clients side or our own to
effectively model the problem to be solved
Page 3 AlgoAnalytics All rights reserved
Banking Problems We Can Solve
Credit score Customer segmentation Sentiment analysis
Predicting balances in
Analytics and loans Portfolio Anaytics current, savings account
Fraud detection
Operations optimization like
establishing and predicting Risk aggregation reporting for
reducing duplicative systems,
patterns and raising an alert Basel III and Dodd Frank
IT costs and others
when anomaly is noticed
Predicting default Alternative modes of credit Liquidity measures by
probabilities on a particular rating those will enable micro predicting in and out flows of
loan in next 12-18 months loans at lower default rates individual customer deposits
Heat maps suggesting what
loans can be offered in what
areas
Page 4 AlgoAnalytics All rights reserved
Internal Credit Scoring Models
Credit scores are available from commercial credit rating agencies.
However, organizations can significantly improve performance and profit
potential with internal statistical credit scoring modes
Three aspects of credit score Benefits of credit score Technical Aspects
Application scoring - helps Application scoring results in Data set required monthly
decision making regarding granting credit to right income, no. of dependants,
acceptance of an application customer and at right price demographics, open credit
lines, loans, past repayment
Behaviour scoring - predict the Behavioural credit scores of history, etc.
likely default of customers that customers help in early
have already been accepted detection of high-risk accounts Classification models - Decision
and perform targeted. trees/ neural networks
Collection scoring - predict the
likely amount of debt that the Collection scores are also used
lender can expect to recover for determining the accurate
value of a debt book before it is
sold to a collection agency.
Page 5 AlgoAnalytics All rights reserved
Customer Segmentation
Customer segmentation includes using techniques like clustering, decision trees or regression analysis to
divide your customers in key segments that reflect both your current customer base and your targets. If
you can understand qualitatively different customer groups, then they can be given different treatments
(perhaps even by different groups in the company). Answers questions like: what makes people buy, stop
buying etc
Segmentation enables offering right product
to right customer at right time and at right
price. It also enables cross selling and up
selling
It enables company to retain customers by
knowing about churn in advance and
taking necessary steps
Customer segmentation will enable a bank
to design a recommender system which
will suggest products, royalty programs etc
to be designed for valuable customers
Page 6 AlgoAnalytics All rights reserved
Sentiment Analysis
Applies NLP, text analysis and computational linguistics to source material to
discover what folks really think
Capture client feedback
Analyze unstructured voice recordings from all centers
and recommend ways to improve customer relations
Build algorithms around market sentiments data
Track trends, monitor launch of new products, response
issues and improve overall brand perception
Page 7 AlgoAnalytics All rights reserved
Analytics and loans
Predictive analytics helps Predicting if profits would Real estate pricing models and
monitor loan origination and increase by reducing interest impact on delinquency rates
performance activity by product rates for a particular borrower
or region Valuation of loan portfolio
Advanced data science
Using techniques like heat map techniques could enable
loan provider can choose to institutions to improve
concentrate on a particular underwriting decisions and
geographical area increase revenues while
reducing risk costs.
What
Origination Pricing
else?
Page 8 AlgoAnalytics All rights reserved
Our Other Analytics Projects
Algorithmic Trading and Strategy Development
- Quantitative strategies in Indian markets
- Improvements to pre-existing algorithmic strategies
Text Analytics
- News/social media analytics
- Multi-language sentiment analysis
Image analytics using deep learning
- Predicting diabetic retinopathy
- object/ image recognition
Performance Manager
- Forecast various KPIs concerning operational performance
Clickstream Analysis
- inherent features of users based on website logs
Mutual Fund Action Predictor
- Our product predicts the changes a portfolio manager is likely to make to his portfolio
Page 9 AlgoAnalytics All rights reserved
Our Product: The Mutual Fund Action Predictor
The product offers insights into
portfolio changes that MFs are likely
Predictive to make
Algorithm
Intermediaries can use the
information to identify
Trade counterparties for their clients or to
initiation initiate trade between two parties
A brokerage can target different
mutual funds proactively for buying
Reducing or selling a large position in stock
impact cost with minimum market impact
One can also use this to compute
expected buy sell pressure on stocks
Buy Sell in near future
pressure
http://mutualfunds.algoanalytics.com:8181/sharefunds
Page 10 AlgoAnalytics All rights reserved
Portfolio Risk Analytics
This involves reviewing existing portfolio, understanding risks and defining problem areas.
Provision of insights and suggestion for improvement
Automated analysis and suggestion to a client to undertake an appropriate course of action for
eg. What area should I focus on to drive my growth
Scalable and cost effective solution for variety of individual portfolios
Page 11 AlgoAnalytics All rights reserved
CEO Profile
Aniruddha Pant
CEO and Founder of AlgoAnalytics
PhD, Control systems, University of California at Berkeley, USA 2001
Highlights
20+ years in application of advanced mathematical techniques to academic and enterprise problems.
Experience in application of machine learning to various business problems.
Experience in financial markets trading; Indian as well as global markets.
Expertise
Experience in cross-domain application of basic scientific process.
Research in areas ranging from biology to financial markets to military applications.
Deep experience in building and guiding 20+ people teams working in quantitative applications.
Close collaboration with premier educational institutes in India, USA & Europe.
Active involvement in startup ecosystem in India.
Prior Experience
Vice President, Capital Metrics and Risk Solutions
Head of Analytics Competency Center, Persistent Systems
Scientist and Group Leader, Tata Consultancy Services
Page 12 AlgoAnalytics All rights reserved
SOME RELEVANT CASE-STUDIES
Page 13
Analytics and Gold Loan
Gold Loans Characteristics Problem Statement
Associated with unorganized sector Using data across branches, it could
Required for short duration be predicted if profits would increase,
Amount of loan required is usually by decreasing interest rate for a
small borrower meeting certain standards
This can be used for any
consumer credit in order to
increase the profitability.
Page 14 AlgoAnalytics All rights reserved
Application of Our Experience For Banks
Place clients at risk within their
A e a ple of the flow of a al tics corresponding segments to view
where they stand in the firm
Get the probability of each client Predicting customers who might
becoming dormant this can be remain good and turn bad
used to predict defaults for FCs
Obtain
Dormancy Prediction predictions
Customer Identify clients Identify clients
Segmentations at high risk not at high risk
High probability
Personalized High probability
of high loss to
Provide usual
of dormancy support
Recommendations the firm
Employ recommender systems using customer
profiles to identify appropriate products to suggest-
can be used to enable a firm to develop
personalized collection effort
Page 15 AlgoAnalytics All rights reserved
Recommender System
What is RecSys? Value of Recommendation
Aims to predict user preferences based on What I really
historical activity and implicit / explicit What I think want What the site
feedback I looki g for wants to show
Helps in presenting the most relevant
information (e.g. list of products / services)
RecSys Modeling and Applications
Nearest Neighbor modeling * Movies, music, news, books,
Collaborative filtering: Users
search queries, social tags, etc.
behavior, similar users
Matrix factorization and
factorization machines * Financial services, insurance
Content-based filtering: using
Intel business unites (BUs), sales
discrete characteristic of items
Classification learning model and marketing
Page 16 AlgoAnalytics All rights reserved
Session Based Recommender System
Use historical sessions data from all customers to recommend products
Items
Sessions
Deep Learning
Page 17 AlgoAnalytics All rights reserved
Use Cases
Product Recommendation Stock Recommendation
Similar
Transactions
Your cart
Similar users and
their stocks
You may also like: Recommended Stocks
Page 18 AlgoAnalytics All rights reserved
Retail Analytics: Other areas
Customer Retention and Loyalty Marketing Analytics
Analytics Understand customer behaviour, predict
Identify customers who will contribute to likelihood of success
sales and build high value relationships with Determine optimal communication channel,
customers and reward loyalty develop optimal promotion strategy to reach
customer
This is a case of recommender
This is a churn problem problem, This can be used in
which is similar to the one we designing customized loan
will need to solve in FC packages for customers
Operational Analytics Risk Management
Credit score to improve decision making at
Fraud reduction, less shrinkage individual level
Store analytics site selection etc Reduce transaction cost, time lags
Determine whether a COD customer will pay
or not
These examples are similar to portfolio
analytics functions required in FCs. Classification and score
Predictive analytics will be able to problems, significant from point
advise about which area and which of view of deciding customers
loans to concentrate on who might default, for FCs
Page 19 AlgoAnalytics All rights reserved
METHODOLOGY AND TECHNOLOGY
Page 20
The Analytics Process
Once a client requirement comes in:
Define and outline the Understand data Develop models
problem statement Data preparation Evaluate performance
Business Requirement Data Situation Analytics Solution
Value-adding Results Seamless Integration
Explainable results Work with clients to
Operational outcomes integrate solution
Page 21 AlgoAnalytics All rights reserved
Machine Learning Techniques
Decision Trees Kernel Learning
Flow-chart like structure SVM extension different kernel
Maps observations of an item to functions for feature subsets
conclusion on items target value Effective when data comes from
a variety of sources
Random Forests Deep Learning
Extension of classification trees Model high level abstractions
High accuracy and efficient on Architecture composed of
large databases multiple non-linear
transformations
Artificial Neural Networks Clustering
Idea analogous to biological Unsupervised learning to group
neural networks data into 2+ classes
Used to discover complex Clustering based on similarity or
patterns in data dissimilarity between data points
Logistic Regression Optimization
Probabilistic statistical Modifying a system to make
classification model some aspect of it work more
Binary predictor efficiently or use fewer resources
Page 22 AlgoAnalytics All rights reserved
Technology:
Page 23 AlgoAnalytics All rights reserved
