Advancing data science & technology

to deliver financial access for all

Who is Trusting Social?

2
Who we are

● We are an AI company focusing on solving

About Us financial inclusion in the emerging markets
● Currently assessing 470 million credit profiles in
Vietnam, Indonesia & India
● 2019 objective: 1 billion
● Our score is the most accurate large-scale
alternative credit score globally
● Present in Vietnam, Indonesia, India and soon to
launch in Philippines and Bangladesh

3
 our team

Bharath Dinh Phung

Nguyen Nguyen, Rohit Narang, Sidharth Kohli, Vasudevan, Ronald Kasim, Johnny Escaler, Tuyen Huynh, Principal,
CEO COO CMO CRO CEO ID CEO PH CTO Scientist

Ph.D. in MBA, MD, MBA MBA (IIM) CFA, Ph.D., MBA, Machine Professor,
Econometrics Goldman Head of Chief Risk President Co-Country Learning, UT Machine
Credit Risk, Sachs Vodafone Officer, Bajaj PEFINDO Head, Austin Learning,
Barclays Enterprise Finserv Credit Bureau Credit Suisse Stanford Monash
Philippines University
 Our Advisors

DR. Burhanuddin DR. Djoko Sarwono, Nelson Tampubolon, DR. Abdul Salam, SE,
Abdullah SH, MPA SE, MSc MM

Governor Bank Bank Indonesia, Bank Indonesia, Bank Indonesia,

Indonesia Aberdeen, Otoritas Jasa PNM, Bank BRI,
PEFINDO, OCBC Keuangan (OJK) Indonesian Banking
NISP Sekuritas, School
Bank QNB
Indonesia

5
Problem Statement

6
 problem
banks and lenders are keen to expand
their customers but lack information
Consumer lending
in emerging markets - No credit bureau data
is tough - Income data not available
- Manual processes (especially KYC)
- High acquisition cost vs lifetime value

7
How we can help?

8
 solution Full-Stack AI for Lending
Credit
Enable banks to Score
tailor financial Targeted
products for the E-Com Fraud Customer
unbanked, using AI Scores Acquisition
and new data
streams AI Digital
Onboarding

Portfolio Affluence
Management Score

and
Score more...
Scrub

9
Why TELCO Credit Score?

10
 Partnership with Telco

Telco data offers the GINI reaches mid 60s,

broadest coverage means very high level of
across data sources separation of good/bad

360-degree view
of customer: Risk,
the business benefits of Our models capture novel
aspects of customer
Fraud, Affluence the score are manifold behaviors which have never
been recorded before

The breadth of data, the

frequency of updates and TS models are built on
the volume = very rich near universal samples
data set and are thus “Generic”
models
11
 ● Model performance significantly superior to alternatives
Telco Credit Score ● Validated in over 60 clients across countries and products
And over 20 client-products in Indonesia

Telco private cluster

Anonymized Call & Location Transaction VAS transaction

SMS History history history

Component feature sets

Income Employment Financial Consumption Social Life

skills profile capital habits 300-850

12
Credit Scores: A risk scoring solution for lenders

Product Description Use Cases

Trusting Social’s Credit Scores are built using ● On the fly underwriting
alternate data sources, Telco data in this case
using state of the Art AI & ML techniques. ● Swap In Swap Outs

Our Credit Scores predicts probability of default by ● Pre-Approvals & Eligibility List Generation
an individual.
● Risk Based pricing
Our Scores Ranges between 300-850, 850 being
the lowest risk customer and 300 being the ● Cross Selling
highest risk customer

13
 ● Score Scrub Services are available for Lenders to check the
Score Scrub scores of their existing customers.
● Example: Saving and Current Account customers of the bank, Rejected
Customers in last 3 month, Dormant customers, Existing Borrowers etc.

Portfolio management on existing Credit Card

portfolio is massively profitable - Credit line adjustment, block/unblock,
refining cross sell funnels, stimulation of
- Acquiring a new customer can cost five offers on credit card to drive usage
times more than retaining an existing
customer Cash Loan
- Line Assignment
- The success rate of selling to an existing - Top up/cross sale product, overdraft,
customer is 60-70%, while the success rate assist with risk-related decisioning and loan
of selling to a new customer is 5-20% offers
- Risk Management and early triggers of
- Lenders need to identify high value and risk for existing loan portfolio
low risk customers

14
How to Access the Telco Score

input output

score Phone number of the Credit score of validation

range applicant. applicant results
purpose at the requested date
300 - 850. with API/Webscore Model performs extremely well
The higher the Provide Credit Score of an individual when they apply with Gini coefficient reaching
better. for a loan & support decision making on Personal Loan .48
products.

15
Backtest Process

16
 Data Set
A
Required Information – to be submitted by financial institution

Mobile Number Loan Origination / Loan Type Performance Income

Application Date

▪ Ideally one file per telco ▪ The preferred format would ▪ i.e. new car financing, used ▪ Days-Past-Due (DPD) of the ▪ Income as specified in loan
operator be yyyymm”. motorcycle financing, first six instalments, application forms (in IDR
unsecured personal loans, separately for each units) or verified
▪ May be encrypted as per ▪ If the loan origination date is credit cards, micro loan instalment. If this is not
client’s request not available then we would
possible, then:
need the loan application
date. ▪ The highest DPD of the first
six instalments

B C
Sample Size Observation Period

Bad Record: Good Record: ▪ Require at least one months of incoming application samples per product type
▪ All bad records for each loan type ▪ Ideally, we would need all of the ▪ Ideally the periods are Jan 2018 - March 2019
ideally be submitted good records (zero DPD) per product
or loan type.
▪ Minimum number of bad record is
1,000 for all months observed period
Backtest Results – Actual Data from FI

18
Lift Table – Actual Data from FI

Score Range Population Bad Rate Bad Percentage Cumulative Bad Bottom 20% of the
300-591 10% 16.7% 35,10% 35.1% Telco scores captures
591-640 10% 7.6% 16,00% 51.1 % 51% of bad customers
640-674 10% 6.0% 12,90% 64.0%
674-700 10% 4.2% 8,70% 72.6%
700-724 10% 3.9% 8,50% 81.2%
724-747 10% 3.1% 6,40% 87.5%
747-770 10% 1.7% 3,50% 91.0%
770-795 10% 1.7% 3,60% 94.7%
795-824 10% 1.5% 3,20% 97.9%
824-850 10% 1.0% 2,10% 100%
TOTAL 5% 100%

19
Swap in - Swap Out

Sub-segments from existing rejected population

could be approved using Telco Score

Approval rate optimization could be driven by

testing
20
Thank you

21