Grameen Foundation’s
Savings Seminar
Data Analytics:
Answering business questions with data
Oct 22nd, 2013
Speakers
p
Tanaya Kilara, Financial Sector
Analyst at CGAP
Jacobo Menajovsky, Senior Data
Analyst at Grameen Foundation
Analyst at Grameen Foundation
The Role of Data
Grameen Foundation Savings Seminar October 22, 2013
Warm-up Quiz
•
How long does it take Google to get 2 million
i
?
queries?
•
How much do consumers spend on web
shopping in a an hour?
•
How many emails sent in a minute?
More Data with Every Passing Day
Big Data Analytics Modelling Data Mining 5Significantly Better Analytical Capacity
Implications
Gleaning Customer Insights Fitting More Data Capacity to Analyze g Products to Needs Managing g g Risk Designing Customer Experience Optimizing Channel 7 ChannelChallenges in Financial Inclusion
•
Banks
Have customer data, need to build analytical capacityMFI
Need to build systems to•
MFIs
capture and analyze datay•
Telcos
Have the capacity, need to use it to generate insights relevant to financial services
Asking the Right Questions
•
What is the problem I am looking to solve?
•
What types of data do I need to answer my
question?
question?
•
How do I get the mix of data right (quant vs qual,
g
g
(q
q
,
internal vs external)?
•
Data gives me the ‘how’. What methods to
answer the ‘why’?
Advancing financial access for the world’s poor
Agenda
Some guiding principles for doing Analytics
Data is everywhere. Why?
A li d i i 101 d
Applied statistics 101, concepts and most common problems and mistakes
Using, mixing, benchmarking, visualizing and testing data to support decisions and respond to business questions
A few guiding principles
Not all products are created equal.
Not all customers have the same needs.
Discovering customers profiles and usage patterns can support product and service (re)design.
Understanding big trends and patterns in the portfolio can
Understanding big trends and patterns in the portfolio can help orgs to drive change and take decisions.
Data is everywhere
Start small
Think data as signs and indicators, not as numbers in an excel file
excel file
All of us are using and modelling data all the time to make even the simplest decisions
Put your questions first and then go to the data
Don’t overcomplicate things, but be careful because it is really easy to “lie to yourself” with statistics
Statistical lies? Are you sure?
The average annual salary of a Lakeside school graduate is e a e age a ua sa a y o a a es de sc oo g aduate s around 2,000,000 per year.
What a class!
How many households below the poverty
line does your organization reach?
Find out with the Progress out of Poverty
What is the PPI?
Find out with the Progress out of Poverty
Index® (PPI®)
A poverty measurement tool for organizations with a mission to serve the poor
10 easy-to-answer questions and a scoring system
Provides the likelihood that the survey respondent’s
Provides the likelihood that the survey respondent s household is living below the poverty line
Country-specific; there are PPIs for 45 countries
Why use the PPI? Why use the PPI?
With the PPI, your organization can:
21
To download the PPI and learn more, visit:
PPI as a segmentation tool
-Survey for the Philippines
Survey for the Philippines
Segmentation Segmentation Family size Schooling Educational level Employment
For the complete survey and look up tables go to: progressoutofpoverty.org
About the data we used
From partners and public sources
Financial, demographic and poverty data
Transactional level
Customer levelCustomer level
Aggregated level
Data comes under different formats, dirty and dispersed
What are we doing with the data?
Measuring poverty outreach and benchmarking against national figures.
Tracking main trends like product performance penetrationTracking main trends like product performance, penetration, uptake, and dormancy levels.
Discovering behavioral patterns and interactions in the data.
Running models to discover main drivers of certain events.
Partner’s overview and poverty outreach
benchmarking
benchmarking
India Philippines India Cashpor 100K+ active savers R.232 (US$3.50) average Philippines CARD Bank 750K+ active savers Php 2900 (US$65) average g savings balances <1% PAR 30 p ( ) g savings balances <3% PAR 30 96% of Cashpor’s customers are living below the $2 line48% of CARD Bank’s customers are living below the $2.50 line
Scaling up savings - Some initial questions (CARD
Bank)
Bank)
What did the savings business look like when the project started (and after)?
a) What was their product offering and cross selling product penetration?
b) What was CARD’s strategy for scaling up savings?
I Customer base expansion?
I. Customer base expansion?
II. Product deepening and cross selling?
Product penetration mapping at CARD Bank
Before and after
Before and after
a) Before
I. 300K accounts
II. 97% monoproduct, only 2.5% cross sold into just one
savings product savings product
b) After
I 750K acco nts
I. 750K accounts
II. 84% monoproduct, 15% cross sold into 4 different
savings products targeting 4 different customer
segmentsg
Kids savings, Convenient access, Increased returns, Regular savings
A few business and social questions we wanted to
answer with data
Which should be the main target segment when
introducing a new savings product at CARD Bank
Cross sold profiling and customer lifecycle analysis
g
g
and when?
Cross sold profiling and customer lifecycle analysis
Average savings by tenure (in years) and poverty level Average savings by tenure (in years) and poverty level
PPI
Profile data PPI
Much higher cross sell penetration
Is it possible to launch an aggressive customer
expansion strategy without affecting poverty
outreach?
Is ATM technology a barrier for the poorest
customers?
customers?
Transactional savings volume by channel and poverty level
Is it possible that transactional fees had an effect on
saving behaviors at Cashpor?
saving behaviors at Cashpor?
How much are they saving? (average amount)
Pay as you go
(a e age a ou )
Yearly fee: Unlimited transactions
h sdfkhd khsdfkhd fk dfhds fk h fk dfhds f
Last 12 months of activity Last 12 months of activity
N=64,841 N=21,731,
“H
ypotheses can be rejected or supported, never proven”
P tti d t t t t
Putting your data to test
Why is it important to test hypothesis and assumptions?Why is it important to test hypothesis and assumptions?
What are the data and tools required to do so?
What are the most common methods?
Your questions and data will help you identify which tests you should apply. pp y
Use correlations to look at whether changes in one variable are
accompanied by changes in another variable accompanied by changes in another variable.
Use the chi-square test to look at whether actual data differ from
a random distribution.
Is tenure correlated with the historic total number of
loans disbursed?
Correlation
Correlation refers to any of a broad class of statistical relationships involving Correlation refers to any of a broad class of statistical relationships involving dependence. Dependence refers to any statistical relationship between two random variables or two sets of data.
disbursed
N
umber of loans
Tenure (length as a customer in months)
N
Pearson’s correlation=.789 R2= 62%
Is tenure correlated with the historic total number of
loans disbursed?
Correlation
disbursed N umber of loansTenure (length as a customer in months)
N
Pearson’s correlation=.789 R2= 62%
Is tenure correlated with the historic total number of
loans disbursed?
Correlation
disbursed
Above average loan takers
N
umber of loans
Tenure (length as a customer in months)
N
Pearson’s correlation=.789
Below average loan takers
Pearson s correlation .789 R2= 62%
Hypothesis: Are women in my portfolio poorer than
men?
Chi-Square test
The Chi Square test tests a null hypothesis stating that the frequency
distribution of certain events observed in a sample is consistent with a particular theoretical distribution.
Hypothesis: Are women in my portfolio poorer than
men?
Chi-Square test
The Chi Square test tests a null hypothesis stating that the frequency
distribution of certain events observed in a sample is consistent with a particular theoretical distribution.
Pearson's Chi Square 0 0000000063
Hypothesis supported
Is there a significant difference on declared assets
across poverty segments?
T-tests
T tests can be used to compare two groups or p g p treatments.