Applied Analytics

Part 1 : Applied Analytics

by analytical verticals
Part 2 : Case studies
from industries
• Non mathematical
• Give an idea on how to apply analytics, look for
opportunities to apply analytics

Created by – Gopal Prasad Malakar

PART 1
Applied Analytics
• View the connections – bird’s eye view of business & goals
• Know about opportunities to apply analytical techniques
• Understand basics of risk, marketing, collections analytics...
• What kind of skills is needed primarily in each vertical
• Get some aspects of Marketing, Risk etc. analytics vertical
work like
• Their objective
• The tools used by them
• The techniques adopted by them
Bird’s eye view

BIG PICTURE
Big picture
• The profit equation
• Profit = revenue – cost – loss
• Big goal  increase profit
• Sub goals
– Increase revenue
– Reduce cost
– Reduce loss
 Analytical verticals
Profit = revenue – cost – loss
– Financial Planning and Analytics Risk Analytics / Collections Analytics / Frau
Marketing Analytics Operations Analytics

Overlap

MBA / MBA /
CA / MBA Econometricians / Statistician MBA Econometricians / Statistician

Excel SAS / R / Excel Excel SAS / R / Excel

Some common industry terms
Industry Lingo
• Secured loan
– Home loan
– Vehicle loan
• Unsecured loan
– Personal loan
– Credit card
• Open ended loan
– Credit card
• Closes ended loan
– Personal loan
• Any similarity between credit card and post paid connections?
• Post paid connections are like loan only..
• Credit line  how much the customer is allowed to use.. Applicable for
both telephone connections as well as credit card
 Definitions

20 Feb 20 Mar 10 Apr 20 Apr

Bill Cycle Bill Cycle

Due date
Bill date

Bill date
• This is applicable in post paid connections (Mobile etc.) and credit card
• Person not paying by due date becomes delinquent
• If you have missed complete 30 days after first bill, you are called 2 due (in
general)
• Reason, you two bills would have got generated and needs payment
• If you don’t pay for another 30 days, you are called 3 due
• Someone who doesn’t pay for 180 days, is written off as loss
Revenue Targets

MARKETING ANALYTICS
Revenue goal
• Revenue = number of customer * revenue / customer
• Increase Revenue goal leads to
– Increase number of customer
– Increase revenue / customer
• Increase number of customer – add new customer
• Maintain customer base – reduce attrition

Customer

Sales gets  Applications / Leads

R
i
s
k

A
p
p
r
o
v
a
l
/
D
e
c
l
i
n
e
Increase / maintain base
 Increase customer base
• Increase number of customer
– Liaison with risk for more approval (reduce leakage) of customers
– Liaison with sales for more applications / leads
– Help sales teams to learn about product feature / plus point for better sales
– Help them to know the profile, which are more suitable for product
• Two important metrics to be watched in this space
No of Applications / Leads Approval rate
26000 42%
24000 40%
22000 38%
20000 36%
34%
18000
Aug-15 Sep-15Jan-15

32%

ug-15 Sep-15Jan-15
 Maintain customer base
• Maintain number of customer
– Reduce attrition
– Liaison with customer services / operations team to reduce customer
pain point
• Two important metrics to be watched in this space

No of complains / customer Attrition rate

0.023 1.8%
0.021 1.6%
0.019 1.4%
0.017 1.2%
0.015 1.0%
5 Aug-15 Sep-15Jan-15

5 Aug-15 Sep-15Jan-15
Find reasons of attrition
• Find reason of customer attrition
– Talk to few customer
– Listen to call recording
– Analyse customer service comments
• Ensure action on customer pain areas
• Set up tracking mechanism to ensure improvement sustains overtime
Analytical tool for
reducing attrition
• Classification tree / logistic regression is a very handy tool in this respect
• Note  Objective function is binary here, and tool is generic in nature
• It can help business to know about which profile has high probability
of attrition
• So that efforts can be made to keep them
• Refer to logistic regression / classification (decision tree) tree course
Increase / maintain  revenue per customer
 Increase revenue per customer
• Goal  increase Revenue per customer
• Track revenue / customer
• If customer are not using the product, find reason for them being
inactive
• Address the issue
• Give them xsell offer
Celebrate Christmas with us
 10 GB data absolutely free for this cycle**
Revenue per customer
2600
2400
2200
2000
1800
-15 Sep-15Jan-15

** You need to get wi-fi connections

or 4G dongle for the same
Analytical tool for xsell
• Classification tree / logistic regression is a very handy tool in this respect
• Same chart  but objective function is different
• The tool is generic and has wide usage
• It can help business to know about which profile has high probability of
taking your xsell product
• So that effort can be optimized for better gains
Tool to increase revenue / customer

MARKETING CAMPAIGNS
Campaigns
• Campaigns are important marketing tool
• Here you give targeted offer i.e. customer specific offer
• And track result to know, where you gained good response
• Classification tree can be used to find, for which profile, we are getting
the best response.
Campaigns
• At times, we are interested in knowing, for each portfolio subsection,
how much increase in revenue in expected
• Regression tree is very useful tool for such situation
• Here the objective function is numeric

Created by – Gopal Prasad Malakar

Campaigns effectiveness for behaviour
change – did campaign work, really?
• At times, we are interested in knowing, whether campaign made a
lasting impact on customer behaviour or not
• For example, did credit limit increase, increase the customer spending
patters on credit card or not?
• If you need to do anything like this, you can take last 90 days spend as
past spend
• And next 90 days spend as next spend
• Adjust both by seasonality.. Reason if past / next 90 days contains
festive season, it will bound to result into more spend
• Apply pair t test on past and next spend
• This gives confidence interval for account level increase / decrease in
spend pattern

Created by – Gopal Prasad Malakar

Loss reduction Targets

RISK ANALYTICS
Goal  Loss reduction
• Activities
– Approval  not let the bad prospect to be customer
– Authorization  not let the exposure go beyond limit / manage fraud
– Exposure management  manage credit line
– ( The exposure should be good enough so that the customers make good
purchases but also for bad customer it will try to reduce the credit limit so it
will not go beyond the limit we should see the delinquency of customer to limit
credit line for those customers)
Sales gets  Risk  Approval/
Applications / Decline
Leads
(Customer is someone who is already on our company but prospect is someone who tries to be our
customer)

Not let the bad prospect be the customer

APPROVAL
Approval
• Objective  find who will turn out to be loss making customer
• Not let them be part of business
• Credit risk  risk of not been able to get the amount back, given on
credit to customer
• Tools  logistic regression (primarily) / classification tree
• It is called application score
• It is also called bureau score as there will be hardly any behaviour for
such prospect
• Bureau keeps the record of financial transactions of customers
Approval – application fraud
• Objective  find which applications are not genuine
• These can be account take over / identity theft cases
• Loss is bound to happen in such cases
• Ensure that these applications don’t get approved
Prevent fraud / bad usage

AUTHORIZATION
Authorization
• Monitor transaction
• Objective 
– Ensure that total exposure is within control
– So not allow high risk customers / delinquent customers to purchase further
– A block is put into the card details, so the transactions gets declined
– For other accounts (unblock accounts), ensure outstanding + current
purchase <= customer specific factor * credit limit
– For medium risk customer, customer specific factor =1
– For low risk customer, customer specific factor =1.2 (generally)
• For example
– If outstanding balance was 45000, credit limit is 50000
– Risky customer will be allowed to make maximum purchase of 5000
– For good customer, it will be allowed up to 15000
 Fraud prevention
• Objective 
– Check if purchase pattern appears genuine or not
– Think of a stolen credit card
– The person will tend to buy too many things quickly
– His behaviour will be quite abnormal
– At times it is part of fraud analytics / fraud management team
• For example
Average of last three purchase
volume Difference in time between two
30000 consecutirve 1500 + purchase
6000
20000
4000
10000
2000
0
02 L_01L_12

0
L_02L_12
Reduce loss from bad customer

MANAGE EXPOSURE
Credit line management
• Objective 
– Reduce exposure on those who are risky
– This is called credit limit decrease (CLD)
– Increase exposure on safe customer (revenue goals)
– This is called credit limit increase (CLI)
• Tools  classification tree / logistic regression
• Score  behaviour score
• Reason, it is based on customers behaviour on account like
– He is paying on time or not
– How is customer utilising his exposure
– How much he is paying
Collect delinquent amount

COLLECTIONS ANALYTICS
Goals
• Collect as much as possible with least cost to reduce losses
• Reduce loss
– Collections strategy: targeted and optimal collections effort to reduce loss
and maximize receivables (someone who pays on the right time on their
own but at sometimes they will forgot so with even small reminder but
there are some people who will not pay unless u explain the
consequences of not paying that’s why we make collection strategy)
• Reduce cost
– Reduce expenditure
– Collections strategy for efficient and cost effective placement of accounts
– (agency A is cost effective than agency B then we choose agency A)
– Efficient resource planning and better utilization of resources
(if we need more people and someother agency fill that gap that’s also right way
to reduce cost)
– Technology based cost effective solutions
Whom to contact for collections

COLLECTIONS STRATEGY
 Medical triad concept
• Tool  classification tree / logistic regression

Accident in remote village

Collections
ll not survive – no matter what you do agents will
Patients will Patients don’t focus their
survive with need doctor’s effort on
medical help who will pay
help with call
Even general
Leave them
Help them villagers can help
them with first
aid
Two competing schools of thoughts
• Target tough one / good will pay on their own
• Target where it make sense rather than wasting money
• Hypothesis for early stage
– Good account pay on their own, chase the bad ones
– Good accounts needs reminder, bad accounts need call
– Good accounts pay with finance charge, late fee hence more revenue,
– Bad one may be left with nothing if delayed
Early stage strategy

Soft Collectabilit Low Med High

Medium y Balance Balance Balance
Aggressive High
Med
Low
Mid stage strategy
• Hypothesis for mid stage
– There is no point wasting money for smaller amount with less
collectible profile
– Target where it yields and take legal root where such means are
not effective ( for people who had delinquent for larger period )
Mid stage strategy

Soft Collectabilit Low Med High

y Balance Balance Balance
Medium
High
Aggressive
Med
Low
See the change
Early stage Collectabilit Low Med High
y Balance Balance Balance
Soft High
Medium Med
Aggressive Low

Mid stage Collectabilit Low Med High

y Balance Balance Balance
(Low balance high here is because High
Of delinquent for 90 days) Med
Low
Late stage strategy
• Hypothesis for late stage
– Get/collect whatever can be obtained by selling toughest accounts
– Work in house where there is best hope
– Agency will work on behalf of us , they will collect

Collectabilit Low Med High

y Balance Balance Balance
In house High
Agency Med
Debt Sale Low
Very Low
How to perform collections

COLLECTIONS OPERATIONS
Decision points
• Strategy 
– Who should be targeted
– When (days past billing, timing of the call) – planning
• Operations  How to execute collections operations
– From which site
– i.e. based on benefit vs. cost
– From which agency
– Where should be targeted (at his residence or office)
• Executing strategy Mode of contacting – SMS / call / letter/ legal letter

45
 Decision points
Payment First Line supervisors: Training
Promise and compliance
Right Party Contact
Production Manager: work
Dials
Production Hours hour scheduling as per month

Paid Hours

• Less of RPC / dial for an agent  he doesn’t know how to open the call /
address the customer
• Less of promise rate  agents needs training to show empathy
• Less of payment as percentage of outstanding  agents needs
training on negotiations skills
Customer voice capture / internal operations efficiency
improvement

CUSTOMER SERVICE /
OPERATIONS ANALYTICS
Customer service analytics
Customer can reveal great information about
• the bottle neck and trouble areas ( why he is not able to use )
• The differences/slack in competitive rate/services provided ( competitors
rate etc)
• What’s the need of customer/market
• About his/other card holders future course of action (such as attrition, if
services doesn’t improve)
Working at appropriate time can help the business the maintain and
improve business. For this, there should be
• a mechanism of capturing the information from customer touch points
and
• a culture to work on that.
 Internal fraud tracking
• Let me explain by example
• In vehicle loan, bank reposes the vehicle, if payments are not coming
• And it resales the vehicle
• It is as costly affair … where will they keep the vehicle? (They try to sell )
• Find average resale price of two year old vehicle by type (say car , farm
land equipment)
• If you see below chart, you should find if there is collusion
Avg price of 2 year old Ford Figo
petrol titanium
220000
210000
200000
Meaning – is the vehicle recovery unit head is 190000
selling the cars to the person known to him 180000
L_01L_12

and hence reducing the price purposefully

Example of analytics
applications for
business
1. For insurance analytics - Application of
analytics in the insurance industry. It is
linking to my youtube video, which will
explain you
-Customer lifecycle in insurance business
-Fundamental of insurance business
-Application of various analytics techniques
along with the customer life cycle
1. Use of predictive analytics / supervised
machine learning algorithms
2. Use of collaborative filtering etc.

https://www.youtube.com/watch?
v=nAYcec5ja_E