Copyright Skyline Labs 2016-17
Copyright Skyline Labs 2016-17
Hackathon 2017
GoPay Credit based e-wallet payment system
Copyright Skyline Labs 2016-17
Why this idea?
GoHack – GoJek services
Current Hyper-local transport & logistics scene in Indonesia
GoJek is expert in services it provides
Minor changes in logistics or backend not enough
To beat the competition – Unique feature
Using existing infrastructure
Copyright Skyline Labs 2016-17
GoCredit – The idea
GoPay : Credit based e-wallet
Revolutionary concept for e-wallets
Allow transactions without requirement to pay immediately
0 GoPay account balance and still need to use services –
Use ‘GoCredits’ for payment (Negative GoPay balance)
Payments clear at end of month
- Through cash or digital payment
Copyright Skyline Labs 2016-17
GoCredit calculation
Sources of data for credit score calculation
1. Bank transactions
2. Timely repayment of bank loans
3. Social media profile
4. GoJek app activity
5. Timely repayment of GoCredits
Machine learning based determination of max GoCredits allocated
Copyright Skyline Labs 2016-17
Data collection
What’s unique – No need to approach banks for data!
SMS Mining through android app will provide complete profile
Research paper on the proposed model accepted to
Peer reviewed IEEE International conference
Copyright Skyline Labs 2016-17
Data collection
Traditional credit score calculation
Copyright Skyline Labs 2016-17
Data collection
Proposed credit score calculation
Copyright Skyline Labs 2016-17
Data mining
Keyword extraction based NLP
Your SBI account 123456789 has been credited/
Debited with Rs 100 for transaction at Flipkart.
Current balance: 56
Example of data captured through SMS
Copyright Skyline Labs 2016-17
Solution architecture
GoCredit
Copyright Skyline Labs 2016-17
Solution architecture
R studio – Data pre-processing
and processing
PHP – Android to MongoDB
REST APIs
MongoDB
Android database NodeJS application for
Web-dashboard
Apache server for
REST APIs
Materialize CSS
ExpressJS
Twitter APIs
Android application
Application uses
Keyword word extraction
Based NLP algorithm
Copyright Skyline Labs 2016-17
What have we made?
Android application
1. Replicates e-wallet transactions
2. SMS data extraction
Apache server with PHP
1. Android to MongoDB data collection and extraction
MEAN Stack web application
1. Graphical data insights
R Server
1. Data pre-processing
2. Twitter data collection and sentiment analysis
3. Credit score calculation Machine learning algorithms
Copyright Skyline Labs 2016-17
Future prospects
GoCredit system opens up a completely new potential domain
For GoJek (PayTM Bank – GoJEK bank)
Credit score as a service – Use the calculated credit score for enabling
Various services
Data collected by the system helps GoJek create detailed model
Of every user
1. Use for targeted advertisements
2. Chatbot (Using contextual data collected , NLP algorithm created)
1. Automated helpdesk chatbot
2. Interactive chatbot for using GoJek services
Copyright Skyline Labs 2016-17
Machine learning
• Data unavailable for accurate prediction as of now
• Sample data trained over linear regressions to calculate weightage of
parameters
• To train the backend – Limited release of GoCredit feature to collect data
• Code for calculating weightage using ML is ready to demo
Copyright Skyline Labs 2016-17
Privacy issues
• Ask user permission to access data – If denied, GoCredits not available
• Many applications (Including Google) access SMS data
• Hash code the data captured - GoJek provides GoCredit without knowing
who the user is.
• Don’t send SMS data to server. Offline logic on smartphone
will calculate the credit score and send to servers.
Copyright Skyline Labs 2016-17
Money security
• GoCredit system is created to be profitable – Model assures minimal risk
• Maximum credit value allocated will not be high (50,000-60,000 Rupiah)
• Credit is given only to the users with average monthly spending of a
certain high amount (For example 1,00,000 rupiah)
• Legally enforceable credit system – defaulting will lead to legal action
• Credit allocation will be based assuming worst case scenarios – The data
trained using ML will allocate 50,000 rupiah if credit worth calculated is
of 1,00,000 rupiah.
• Typical value for customer acquisition - ~30,000-40,000 rupiah.
• Risk involved in GoCredit per user – equal to customer acquisition cost
Copyright Skyline Labs 2016-17

Kym - GoJek GoPay integration

  • 1.
  • 2.
    Copyright Skyline Labs2016-17 Hackathon 2017
  • 3.
    GoPay Credit basede-wallet payment system Copyright Skyline Labs 2016-17
  • 4.
    Why this idea? GoHack– GoJek services Current Hyper-local transport & logistics scene in Indonesia GoJek is expert in services it provides Minor changes in logistics or backend not enough To beat the competition – Unique feature Using existing infrastructure Copyright Skyline Labs 2016-17
  • 5.
    GoCredit – Theidea GoPay : Credit based e-wallet Revolutionary concept for e-wallets Allow transactions without requirement to pay immediately 0 GoPay account balance and still need to use services – Use ‘GoCredits’ for payment (Negative GoPay balance) Payments clear at end of month - Through cash or digital payment Copyright Skyline Labs 2016-17
  • 6.
    GoCredit calculation Sources ofdata for credit score calculation 1. Bank transactions 2. Timely repayment of bank loans 3. Social media profile 4. GoJek app activity 5. Timely repayment of GoCredits Machine learning based determination of max GoCredits allocated Copyright Skyline Labs 2016-17
  • 7.
    Data collection What’s unique– No need to approach banks for data! SMS Mining through android app will provide complete profile Research paper on the proposed model accepted to Peer reviewed IEEE International conference Copyright Skyline Labs 2016-17
  • 8.
    Data collection Traditional creditscore calculation Copyright Skyline Labs 2016-17
  • 9.
    Data collection Proposed creditscore calculation Copyright Skyline Labs 2016-17
  • 10.
    Data mining Keyword extractionbased NLP Your SBI account 123456789 has been credited/ Debited with Rs 100 for transaction at Flipkart. Current balance: 56 Example of data captured through SMS Copyright Skyline Labs 2016-17
  • 11.
  • 12.
    Solution architecture R studio– Data pre-processing and processing PHP – Android to MongoDB REST APIs MongoDB Android database NodeJS application for Web-dashboard Apache server for REST APIs Materialize CSS ExpressJS Twitter APIs Android application Application uses Keyword word extraction Based NLP algorithm Copyright Skyline Labs 2016-17
  • 13.
    What have wemade? Android application 1. Replicates e-wallet transactions 2. SMS data extraction Apache server with PHP 1. Android to MongoDB data collection and extraction MEAN Stack web application 1. Graphical data insights R Server 1. Data pre-processing 2. Twitter data collection and sentiment analysis 3. Credit score calculation Machine learning algorithms Copyright Skyline Labs 2016-17
  • 14.
    Future prospects GoCredit systemopens up a completely new potential domain For GoJek (PayTM Bank – GoJEK bank) Credit score as a service – Use the calculated credit score for enabling Various services Data collected by the system helps GoJek create detailed model Of every user 1. Use for targeted advertisements 2. Chatbot (Using contextual data collected , NLP algorithm created) 1. Automated helpdesk chatbot 2. Interactive chatbot for using GoJek services Copyright Skyline Labs 2016-17
  • 15.
    Machine learning • Dataunavailable for accurate prediction as of now • Sample data trained over linear regressions to calculate weightage of parameters • To train the backend – Limited release of GoCredit feature to collect data • Code for calculating weightage using ML is ready to demo Copyright Skyline Labs 2016-17
  • 16.
    Privacy issues • Askuser permission to access data – If denied, GoCredits not available • Many applications (Including Google) access SMS data • Hash code the data captured - GoJek provides GoCredit without knowing who the user is. • Don’t send SMS data to server. Offline logic on smartphone will calculate the credit score and send to servers. Copyright Skyline Labs 2016-17
  • 17.
    Money security • GoCreditsystem is created to be profitable – Model assures minimal risk • Maximum credit value allocated will not be high (50,000-60,000 Rupiah) • Credit is given only to the users with average monthly spending of a certain high amount (For example 1,00,000 rupiah) • Legally enforceable credit system – defaulting will lead to legal action • Credit allocation will be based assuming worst case scenarios – The data trained using ML will allocate 50,000 rupiah if credit worth calculated is of 1,00,000 rupiah. • Typical value for customer acquisition - ~30,000-40,000 rupiah. • Risk involved in GoCredit per user – equal to customer acquisition cost Copyright Skyline Labs 2016-17