Leveraging Big Data For Payment Risk Management€¦ · Big Businesses (Enterprise) Medium-sized...
Transcript of Leveraging Big Data For Payment Risk Management€¦ · Big Businesses (Enterprise) Medium-sized...
Payments for platform businesses.
Leveraging Big Data For Payment Risk ManagementJohn Canfield, VP Risk Management, WePay@JCRisk"
June 11, 2014© 2014, WePay Inc.QCon New York, 2014
New York 2014
Outline1. Payments opportunity for the bottoms-up economy"
2. Requirements"
3. Solution"a) Big data collection"
b) Decisioning using machine learning, rules, and expert staff"
c) Metrics and feedback
2© 2014, WePay Inc.QCon New York, 2014
3© 2014, WePay Inc.QCon New York, 2014
What is the “Bottom Up” Economy?
Big Businesses (Enterprise)
Medium-sized Businesses
Small Businesses
Bottom Up Economy
‣ According to IRS data, there are over 25 million businesses in the United States with less than 5 employees. "‣ They collectively take in over
$2 trillion annually - a tremendous opportunity for electronic payment conversion. "‣ Services (non-retail) is most of
the market: 22m businesses & $1.7 trillion in income.
Marketplaces
5© 2014, WePay Inc.QCon New York, 2014
Marketplace A
Buyers
Merchants (vertical 1)
Marketplace B
Buyers
Merchants (vertical 2)
Marketplaces bring buyers to merchants in a particular vertical.
The marketplace curates the merchants for quality, controls the buying experience, and facilitates payments
6© 2014, WePay Inc.QCon New York, 2014
$18 Billion
valuation$10 Billion
valuation
$300M
valuation
Crowdfunding sites
© 2014, WePay Inc.QCon New York, 2014
Crowd- funding
site
Donors
Fundraisers for • projects • pre-orders • personal • charity • equity - rule TBD
Crowdfunding sites are similar to marketplaces but instead of bringing buyers to sellers, they bring donors to fundraisers.
The crowdfunding site enables the fundraiser to market their campaign and accept payments
8© 2014, WePay Inc.QCon New York, 2014
Small Business Platforms
© 2014, WePay Inc.QCon New York, 2014
Buyers
Small business platforms offer online services to small businesses like marketing, invoicing, shopping cart, accounting and payments
Small Business
marketing
invoicing
payments
shopping cart
accounting
website builder
10© 2014, WePay Inc.QCon New York, 2014
Requirements
Traditional payments do not work for bottoms-up economy
12© 2014, WePay Inc.QCon New York, 2014
Long application form + day/weeks to approve + approval difficult for individuals
Poor conversion rate for micro-businesses and fundraisers
Payments requirements for bottoms-up economy
• Easy and fast merchant onboarding"• Underwrite individuals or micro-businesses with little or no traditional
business history"• Prevent collusion and takeover fraud
13© 2014, WePay Inc.QCon New York, 2014
Fraud threats are everywhere
Information Week Target Confirms Hackers Stole 40 Million Credit Cards
The New York Times Michaels Stores’ Breach Involved 3 Million
The New York Times Neiman Marcus Data Breach Worse Than First Said
The Register Krebs: Lexis-‐Nexis, D&B and Kroll hacked
Entrepreneur, April 18 2014 Online Debit, Credit Fraud Will Soon Get Much Worse. Here's Why.
CyberSource Online Fraud Report EsEmated $3.5 Billion Lost to Online Fraud
Fraudsters want to monetize their stolen cards
Physical retailer
Back of a truck
Online retailer
Online Mktplace
Resell Drop ship 2-sided platform
Solution outline1. Lots of risk data from many sources"
a) What data"
b) How to collect"
c) What infrastructure"
2. Multi-level risk decisioning"a) Machine learning"
b) Rules"
c) Manual"
3. Metrics & feedback16© 2014, WePay Inc.QCon New York, 2014
Data
Data approach
• Predictive of loss / fraud
18© 2014, WePay Inc.QCon New York, 2014
data
• No silver bullet, so move towards big data
• More data is better if you have scalable data arch and decisioning
• Compliance
Know-Your-Customer (KYC) checks
19© 2014, WePay Inc.QCon New York, 2014
data
API
User
vendor
Typical vendors: • Experian • Equifax • Lexis Nexis • ID Analytics • IDology
Traditional business credit reports
20© 2014, WePay Inc.QCon New York, 2014
data
APIvendor
Typical vendors: • D&B • Experian • Equifax
Business Incorporation Docs
21© 2014, WePay Inc.QCon New York, 2014
data
User manualAPIvendor
Business social media
22© 2014, WePay Inc.QCon New York, 2014
data
manual
API
Editorial Reviews and Ratings
23© 2014, WePay Inc.QCon New York, 2014
data
manual
spider
Maps / Street View
24© 2014, WePay Inc.QCon New York, 2014
data
manual
API
Facebook profile
25© 2014, WePay Inc.QCon New York, 2014
data
manual
API
26© 2014, WePay Inc.QCon New York, 2014
data
manual
API
Device ID
27© 2014, WePay Inc.QCon New York, 2014
data
Typical vendors: • ThreatMetrix • iovation • Experian / 41st Parameter
APIvendor
Control Verification
28© 2014, WePay Inc.QCon New York, 2014
data
+
APIvendor
Typical vendors: • Authentify • Telesign • Twilio
Transaction History
29© 2014, WePay Inc.QCon New York, 2014
data
DB
Data from Partners
30© 2014, WePay Inc.QCon New York, 2014
Marketplace
Buyers Merchant
WePay
Payments APIRisk API
data
Risk API Account Information
31© 2014, WePay Inc.QCon New York, 2014
data
Risk APIpartner
• person • email • business_name • address • phone • tax_id • website_uri • employment • industry_code • business_description • risk_score • comment • project • fundraising_event • fundraising_team • acquisition_channel
• partner_service • member_to_member_message • external_account • editorial_review • other_web_content • revenue • conversation • business_legal • business_report • other_document • device_info • control_verification • risk_review • risk_review_steps • transaction_details
Example Risk API data types
Risk API Transaction Information
32© 2014, WePay Inc.QCon New York, 2014
data
Risk APIpartner
How to effectively organize this data?
33© 2014, WePay Inc.QCon New York, 2014
data
301 duncan loop dr #04, Dunedin, FL 34698
3605 Alt 19 North Palm Harbor, FL 34683
1621 Central Ave Cheyenne, WY 82001
1330 Avenue of the Americas New York, NY 10019
AB Online Training, LLC
EIN: 46-099XXXX
Website: www.ABOnlineTraining.com
Carrie Christie Head of Bus Development
Data storage systems
34© 2014, WePay Inc.QCon New York, 2014
data
SQL databases: • MySQL • Oracle • …
No-SQL databases: • Document (MongoDB…) • Key-value (Redis…) • Graph (Neo4J…) • Column (Cassandra…)
Other: • Hadoop • …
Decisioning
What is Machine Learning?
36© 2014, WePay Inc.QCon New York, 2014
+
1) Supervised learning
Machine Learning Model
Training
2) Production use
decisioning
How does machine learning work?
37© 2014, WePay Inc.QCon New York, 2014
decisioning
Linear regression classifier
Support Vector Machine
38© 2014, WePay Inc.QCon New York, 2014
decisioning
Decision Trees
39© 2014, WePay Inc.QCon New York, 2014
IP dist < 80
Txn <= $250
Txn > $250
IP dist > 80
BL accts <= 0
BL accts > 0
ID3 Training algorithm Recurse and choose attribute that minimizes entropy:
Fraud
Fraud
No fraud
No fraud
decisioning
Neural Network models
40© 2014, WePay Inc.QCon New York, 2014
Gradient descent training
decisioning
Meta-algorithms: grid search, bagging, etc
41© 2014, WePay Inc.QCon New York, 2014
Decision Tree
You can do: Or you can do:
Decision Tree 1
Decision Tree 5
Decision Tree 4
Decision Tree 3
Decision Tree 2
SVM1 SVM 5 SVM 4 SVM 3SVM 2
Neural Network 1
Neural Network 5
Neural Network 4
Neural Network 3
Neural Network 2
Logistic Regression 1
Logistic Regression 5
Logistic Regression 4
Logistic Regression 3
Logistic Regression 2
decisioning
Rules Engine
42© 2014, WePay Inc.QCon New York, 2014
Rule 1 (Transaction threshold): If (signal.Group_30Day_txn_volume > account.tpv_review_threshold_t30d and fact.tpv_review_threshold_t30d > 0) then Refer_for_Review
Rule 2 (Model Score): If (model.fraud_score > 0.85) then Refer_for_Review
Rule 3 (Re-require KYC): If (signal.IDR <= 0) and (signal.MQR <= 0) and (signal.Group_30Day_txn_volume > 1000.00) and (signal.name_address_match = false) then Request_Address_Remedy
decisioning
Manual Review
43© 2014, WePay Inc.QCon New York, 2014
decisioning
Metrics & Feedback
Metrics & Feedback
45© 2014, WePay Inc.QCon New York, 2014
0"30$days$ 30"60$days$
60"90$days$
90"120$days$
120"150$days$
150"180$days$
Net$Loss$Matura,on$Curves$
2013"10$
2013"11$
2013"12$
2014"01$
2014"02$
2014"03$
0"100$
100"250$
250"500$
500"1000$
1000"2500$
2500"5000$
5000"10000$
>10000$
Net$Loss$by$Txn$Size$
Aug$13' Sep$13' Oct$13' Nov$13' Dec$13' Jan$14' Feb$14' Mar$14'
Net$Loss$Forecast$
21.0%&
33.9%&
35.6%&
6.2%&
1.5%&1.1%& 0.8%&
Virtual_Terminal&
Invoice_Payment&
API_Checkout&
Store_Cart&
DonaDon&
Ticket&
SubscripDon_Payment&
metrics
Summary1. Bottoms-up economy is growing and is the high-growth frontier for
electronic payments
46© 2014, WePay Inc.QCon New York, 2014
2. Big data needs to be pieced together bit by bit using flexible infrastructure
3. Machine learning, rules, and manual review when combined properly gives enables correct decisions on big data