WHITE PAPER

ARTIFICIAL INTELLIGENCE AND

ROBOTICS PROCESS AUTOMATION IN
CARDS
#IncentiveAutomation; #PrescriptiveAnalytics;
#AutomotiveBlockChain; #Mobility

Abstract
Nientibus et harum la aliquos que dunt harunte nat qui assimin
ctincti nimoloratur? Quis sin enim expello rescitis aliberiosam,
sumendu cienimil es ab in pelibus antiunt, eatur sit volorec tetur,
occus asi suntiss imporer eperis dolupta que quid quatis mo volorit
quas maio. Im acest, eos si beat.
Ur? Nonseque reribus. Itatium re, nissi nullupietur audis sit adis con
non corrum fugias eosae nones nonsenimus Itate esto moluptatur
autatis sinctota dolent labo. Sum autem reriossum eos acerestectur
rem que et haribus vel etur
 Introduction
Artificial Intelligence (AI) and RPA (Robotics
Process Automation) is the buzz word in all
industries today. You cannot get through
a day in the todays tech world without
hearing these terms. Financial services have
also taken towards it in a slow, nevertheless
steady fashion.
Father of Artificial Intelligence, John
McCarthy, defines it as “The science
and engineering of making intelligent
machines, especially intelligent computer
programs”. AI is used to define all systems
that can handle tasks performed in general
by humans which require skills like voice
recognition, perception of vision, touch,
speech, problem solving, learning etc.
This is usually used together with Machine
Learning (ML) where the machines are
not programmed to perform a specific
mundane task, but programmed to learn
from experience and build the learning
into the system for handling similar tasks
in future. AI systems provide best results
used on larger samples and they are
capable of analyzing huge amounts of
data and are usually used along with big
data technologies. Though AI concepts
have been in existence for quite some time
now, they have gathered momentum and
practicality only after the advent of big
data technologies as AI technologies are
considered only as good as the data and
the data processing technologies it runs on.
Robotics Process Automation (RPA) aka
digital labor is really a step before AI, where
the systems or (Ro)bots as they are usually,
called perform repetitive rule based tasks
without any intelligence (or continuous
learning) as in AI. It is usually simpler than
AI and can serve most common automation
requirements.
For instance, a call center decision
automation can be handled by RPA
bots, which can use decision trees to
answer simpler queries and pass on the
complicated queries to operators. AI may
be required when the call center wants the
system to learn from daily queries and build
the handling capacity within the system for
future calls.
RPA and AI used appropriately reduces cost,
minimizes human error, increases speed
and elevates customer experience.

External Document © 2018 Infosys Limited

Artificial Intelligence and Robotics Process Information in Finance
AI and RPA are not just sci-fi words, but Some typical examples are: to buy and use off the shelf, with little
becoming a reality and an integral way customization. An example is Infosys
• Humanoids like Nao (Bank of Tokyo),
forward in most industries. Financial Automation for Artificial Intelligence
Pepper (Muzuho Bank) which greet
services industry is usually known to take (Nia) and Robotics Process Automations
customers and assist them on their
a cautious approach to any change, as the (AssistEdge) which offers various AI/ RPA
requests
industry is governed by tight regulations products
and compliance is a top priority. But as • Chatbots like Erica (Bank of America),
• AMEX offers an application to
working smarter than working harder is Eno (Capital One), COIN (JP Morgan)
provide customized restaurant
getting more important, firms have to that provide instant responses to some
recommendations for customers
adopt automations into the core of their frequently asked questions or progress
which uses AI over huge number of
operations to stay in competition. The legal queries or contracts quicker
transactions to train the model
very same issue of the industry being • Robotics Process Automation systems
regulated, offers scope for RPA since the The following charts depict the adoption
which provide end to end service for
various compliance that are mandated trend of AI technologies in Financial
Customer Service operations, account
can be achieved better using RPA. Off late services and the areas where banks feel
opening, fraud detection, finance
banks that have waded into AI/RPA and are automations can be used. This clearly
management etc.
reaping the benefits. shows that these technologies are being
• Most Fin Tech service providers are investigated and embraced slowly in the
offering AI and RPA products, for banks financial sector.

Adoption Trend for Artificial Intelligence

45%

40%

35%

30%

25%

20%

15%

10%

5%

0%
Not Started Investigating Engage Collaboration Build Solution Built a centre
Startups in group of excellence
initiatives

AI & RPA Usage possibility

13% Customer service

30%
Back office / operations
14%
Financial Advisors

Fraud detection
19%
24% Risk Management

The data is based on a survey conducted by Delloite and Efma (April 2017)

External Document © 2018 Infosys Limited

 Artificial Intelligence in Cards
Following are some case studies of AI existing data. These robots can take over
usage in the financial service industry; the credit scoring functions. One such
cards domain in particular. product is GiniMachine which offers
a bot that can predict credit behavior
I. Credit scoring based on existing customer data.
Credit scoring in an important exercise for
the credit card business, as this determines II. Customer Acquisition
the risk that the issuer undertakes when Sales is one sector which is heavily person
issuing a line of credit. The card product dependent and the success of any product
and the upper limit of credit that can be depends on the skills of the marketing
offered to a customer is decided based on team. AI has now advanced into this field;
the credit score. This forms a basis to the systems are built either to assist a sales
profit or loss of a product. Banks usually person make the right sale (as in the case of
rely on third party data providers like Rainbird Technology used by MasterCard)
Experian, Equifax, FICO to get credit scores or provide an end to end system where
as it is a specialized function and requires the customer can choose and purchase a
storage and analysis of huge data. product without staff assistance (as with
the comparison site - Masii)
Let’s look at some common problems
associated with Credit Scoring which can • Rainbird & MasterCard: MasterCard
be solved via AI: has entered into an engagement with
Rainbird technology to use an Artificial
• Unavailability of sufficient data for credit
Intelligence System in its sales team.
scoring; Based on a survey conducted by
The system will use Machine Learning to
ID Analytics, nearly 20% of US customers
capture subject matter expertise available
do not have enough data to generate
with a few individuals in the sales team
a credit score. This problem is getting
of MasterCard. This can later be used by
bigger with millennials entering the
the entire sales team who can look up
credit card market as buyers and form a
the system, when they speak to clients to
major chunk of the customer base. Credit
determine the best credit card product
scoring bureaus simply do not return a
that will suit their needs. The technology
score for this group. This problem can
acts a look up for sales staff and the
be handled by AI, where a set of simple
system will get better as more people use
rules can be coined based on age, gender,
it via self-learning
income etc. to decide a risk score. Any
number of rules can be created which • Massi: This is a comparison site where
can be difficult to track via human skills customers can choose between various
without automation insurance providers or credit cards and
make a purchase on the site directly
• The traditional score methods are
rather than contact agents or sellers. The
based on hard and fast rules which lacks
site designed by a Thai based startup that
in capturing the activities of certain
uses AI as base to provide the right filters
individual entities or provide optimum
and options for customers to narrow
ways to segment scoring models.
down their search and find the right
Segmentation is required because
product
different type of card products and
different customers require different • Machine Learning is also used for
scoring models. FICO is now using AI targeted advertising to identify the right
techniques like collaborative profiles products and campaigns. American
to perform the segmentation based on Express uses this to its advantage and
similarity rather than hard rules. This will data says that up to 40% of customer
capture subtle patterns into the risk score acquisition for AMEX is via online
platforms
• Another use of AI is to train bots using
machine learning techniques over

External Document © 2018 Infosys Limited

III. Fraud Management and location (from the usual location). For • Continually verify and update the
instance, if a customer in San Francisco behavioral attributes
False denials
performs a transaction in London and the
• Use aggregated behavioral data (from
According to data collected by Coalition location of the customers mobile is also
other users and service providers)
Against Insurance Fraud (CAIF), fraud London, then the risk score is reduced
to be used in tandem with personal
costs the Card industry dearly. Nearly since the geo location the mobile phones
behavioral pattern for fraud detection
2.4 billion dollars is lost every year by US emit is one of the best indicators of the
credit and debit Card issuers. The existing customer’s location. The data is not just Since Artificial Intelligence, Big Data
fraud protection systems works based used to authorize a particular transaction, and Machine learning go hand in hand,
on certain typical rules which are usually but is used by the system to make itself implementing these is no cheap affair
based on geography, the size of the smarter using self-improving algorithms which may pose a problem to smaller
transactions which will not work in today’s players. But off late, a lot of products
MASTERCARD
world. Various players are coming up with have been launched that leverage this
customized data driven solutions for fraud Fraud prevention though an important opportunity and offers a solution to small
management which are detailed below: aspect has a downside when it leads to too players who cannot afford such a set up on
much of false declines. These are scenarios their own.
VISA when a valid customer’s transaction under
VISA Advanced Authorization, the FICO is an industry leader as an analytics
credit limits gets rejected. This is seen
analytical system used by VISA to authorize solutions provider which helps various
as a greater risk than fraud. The amount
its transactions is built over AI, ML and organizations make better decision using
of false declines are 3 times the actual
Data analytics concepts. There are around predictive analytics and AI. Falcon is a
number of fraudulent transactions. False
150 million transactions processed by per fraud management solution offered by
denials are a loss to both the credit card
day over VISA and it becomes important FICO which can be implemented on site or
issuer and the merchant, as customer
to identify fraudulent transaction while can be an off the shelf product consumed
is not likely to return and is definitely
approving valid transactions that happen from cloud. Players can choose the
not going to be satisfied with the credit
less frequently. appropriate solution based on their size,
card issuer. MasterCard has developed a
requirement and budgets. Falcon reduces
comprehensive system for Fraud detection
• When a transaction happens on a VISA fraud and thereby protecting the clients
and approvals called Decision Intelligence.
card, information about the transaction from reputational damage and financial
One of the major focus of this systems is to
is collected and sent to the Authorization loss. It also reduces false positives (denials)
reduce the number of false denials.
System which uses models created to maintain customer experience. Key
to analyze the information on the • As detailed previously, traditional fraud features of Falcon are:
transaction in question, over past data detection systems use hard and fast
• Uses self-learning algorithms which not
rules and a limited set of parameters.
• The system uses up to 500 attributes only learn from transactional data, but
Decision Intelligence uses machine
to find possibilities of fraud. Some of also include new fraud patterns prevalent
learning algorithms over huge historic
the attributes used are location of the in the market
spend data to arrive at a score for each
transaction (is it from a store that the
transaction specific to the customer • Provides options for clients to customize
customer regularly shops?), the value of
rather than using generic rules. the rules engine
the transaction (is this transaction in line
with the customer spending pattern), etc. • The Decision Intelligence system’s • Individualized scores by leveraging Big
algorithm are self-evolving. Data to extract patterns from millions of
• Based on this a risk score is assigned to
the transaction which is shared with the transactions
• MasterCard has gone one step further
issuing bank which decided how best to to combine behavior analysis and • Analyses billions of transactional data
use this score biometrics to prevent identity theft and from issuers to continually build and
fraudulent transactions. NuDetect is update its fraud patterns
• It’s authorization system is so robust that
a solution offered by MasterCard that
it can handle 32,500 transactions analysis • Its Identity Resolution Engine uses social
uses a four factor model to identify
per second and about 6 billion per day media data for improved risk scores
imposters:
• VISA has also recently started using IBM is also offering a tool that creates
• Captures Enhanced Device ID for
mobile location confirmation to decide models based on historic transaction data
capturing the location
the authenticity of transactions. This for each card via machine learning. This
tracks the location of the mobile number • Captures various behavioral attributes model is then used to perform a predictive
and links it to the transaction location. like typing style, angle at which the analysis for each new transaction and
This helps especially in one off valid device is held, pressure, navigation assign a fraud score for it, based on which
transactions that happen in a different patterns etc. to identify the customer it can be flagged appropriately. The system

External Document © 2018 Infosys Limited

 learns with every transaction to better the a database of 68 billion transactions that AI techniques and traditional rule base
model, apart from the frequent updates to CyberSource processes annually, via a tie- static models in its fraud management
include new fraud patterns detected in the up with VISA. solution. It uses several ML techniques
market. instead of just one including regression
• One of its products, Internet Fraud Screen
analysis, Neural networks and decision
IBM also offers a tool called IBM Watson in (IFS) developed in collaboration with VISA
trees to increase the effectiveness of
collaboration with MasterCard to offer data to leverage its huge database is used by
prediction. This is because no one model
analytics solution, that can be leveraged merchants to prevent online frauds. IFS
will work in all situations and right model
by small and medium businesses. This will uses AI and neural networks (on VISA’s
should be used on the right set of data
be provided to merchants as a part of their database) to predict customer behaviors
(more the data better the results). The
MasterCard acceptance package and will and flag fraudulent transactions while
fraud management system also offers
give them access to aggregated data on reducing the number of false denial at
an UI where the business user can set up
revenue sections, market share and buyer the same time.
rules and also monitor rules that were
data. This will also be used by MasterCard
• A common issue with machine learning is triggered to arrive at a decision hence
to collate customer behavior trend.
that, it is usually a black box and generally offering more transparency. This is a
CyberSource, an ecommerce payment one cannot explain the result or output. key differentiator CyberSource offers as
management company provides payment One way to combat this is to use Machine against its competitors
gateway and fraud management services Learning with traditional rules based
for merchants. It offers various fraud approach. CyberSource uses Real Time
management products that are built over Fusion approach to combine self-learning

Robotics Process Automation for Cards

Following are some case studies of RPA relevant data which is then used to accounts, collect past dues status and
usage in the financial service industry; identify the credit worthiness of the other required details from various
cards domain in particular. individual before the account is opened. systems to processes the collections
Synechron is one such digital solution cases quickly and efficiently based on
I. Account Opening provider which offers end to end pre-defined rules. The RPA system also
Not all account opening processed are account opening via RPA has the capability to block the card or
completely digitalized. Some accounts forward the case to an employee for
• Synechron also uses OCR to perform
are still opened in a manual mode where manual action
KYC which is a part of the on boarding
a customer walks into a branch and fills
process, by scanning various III. Call Centre
up paper forms or contacts an agent. The
documented for the relevant fields
manual process of document verification Software bots are now used in answering
and updating the bank’s other internal
and application processing is mundane call center queries which were manually
systems
and rule based. handled previously. Some examples are
II. Back Office incident management, billing queries,
• Axis bank recently embraced RPA in
product enquiry.
its account opening process where RPA can be used to handle laborious
they used RPA to verify the supporting processes in the back office which are • ICICI bank has liaised with OpenSpan
documents. The solution uses optical usually mundane, and error prone. Some to implement its Robotics Automation
character recognition (OCR) to read the examples are ledger management, order Platform. Bots have reduced the
data from the supporting documents management, collection and invoicing. customer response time by 60% with a
(age proof, address proof etc.) and map 100% increase in accuracy. For instance,
• Swiss Post offers a RPA solution to
it with the entry on the application form ATM query resolution was a cost heavy
one of its clients to handle the list of
to validate it based on predefined rules operation for the bank with 200 million
accounts whose collection status is
transactions per year, leading to 0.1
• OCR is also used by various RPA past overdue. What used to be manual
million queries. The query resolution
products to make the entire account process following laborious rules to
time was bought down from 12 hours to
opening process digital. It does so by debit the amount past due from the
4 hours per query following the usage
allowing customers to fill forms online, account was automated. The system
of bots in the area
attach documents. These documents automatically screens various customer
are scanned using OCR to obtain

External Document © 2018 Infosys Limited

Conclusion
Automation technologies detailed above
have surely moved away from academia to
being implemented in business real-time
and benefits being realized. Clearly any
financial institution not using these will
lose out in the race in the near future.
These technologies are not to replace staff,
but to augment their functions, relieve
them of mundane tasks and enable them
to concentrate on other innovative aspects.
Of course there will be disbursements of
units, re-skilling of staff doing the manual
work. The organizations will have to adopt
strategies not just to embrace automations,
but to ensure that transition is smooth for
the staff. Automations are enablers and are
not to replace people.
Data privacy and security is another big
concern especially in the financial service
industry. Unprecedented huge amounts
of customer data are being collected
and stored to be used by AI, RPA and ML
using various Big data technologies. It is
important that the data access and privacy
are controlled by rules and regulations as
applicable in the geography. Implementing
these regulations with the right technical
security controls should be part and parcel
of these automations.

About the authors

Ramya Alagappan
Ramya Alagappan is a Technology Lead with Cards and Payments Practice at Infosys. She has around 12 years of
experience with core consulting expertise in Credit Cards Issuance and Maintenance.

She can be reached at [email protected]

Kannathal Alagappan
Kannathal Alagappan is a Senior Consultant with the Infosys Financial Services Domain Consulting Group. She has over
10 years of experience in E2E project management & Consulting in Cards & Payments domain. Her core experience
includes Managing Risk & Compliance in Card Management and Payment Processes.

She can be reached at [email protected]

External Document © 2018 Infosys Limited

References

https://en.wikipedia.org
https://www.pwc.com
https://thefinancialbrand.com/52735/robots-artificial-intelligence-ai-banking
https://www.abe.ai
https://www2.deloitte.com/content/dam/Deloitte/nl/Documents/financial-services/deloitte-nl-fsi-ai-and-you.pdf
https://www.cnbc.com/2016/03/03/mastercard-using-artificial-intelligence-to-turn-staff-into-top-sellers.html
https://www.techinasia.com/masii-financial-comparison-site-thailand
https://www.hostmerchantservices.com/2016/12/mastercard-using-artificial-intelligence-to-attack-false-declines/
https://www.riskified.com/pdfs/Riskified-Javelin-Whitepaper-2015.pdf
https://www.infosys.com/
https://ginimachine.com/
https://www.research.ibm.com/foiling-financial-fraud.shtml
http://www.fico.com
https://usa.visa.com/visa-everywhere/security/outsmarting-fraudsters-with-advanced-analytics.html
https://www.cybersource.com/solutions/featured_solutions/optimizing_fraud_management/
http://phx.corporate-ir.net/phoenix.zhtml?c=94844&p=irol-newsArticle&ID=132938&highlight=
http://www.livemint.com/Technology/WCjwMFggazeG7fcQFsSTEN/How-Axis-Bank-is-shaping-its-digital-journey.html
https://www.swisspostsolutions.com/en/knowledge-center/intelligent-automation-in-banking
https://www.cybersource.com/content/dam/cybersource/NA_Machine_Learning_Whitepaper.pdf
https://mapr.com/blog/machine-learning-american-express-benefits-and-requirements/
https://developer.mastercard.com/product/nudetect
http://www.zdnet.com/article/mastercard-to-leverage-ibm-watson-to-give-smbs-big-data-access/
https://cio.economictimes.indiatimes.com/news/strategy-and-management/icici-bank-deploys-software-robotics-to-
automate-business-processes/54179316
https://quizlet.com/40977/artificial-intelligent-flash-cards/
https://www.ventureradar.com/organisation/GiniMachine/b7232a1d-430f-42a2-9f7b-ef96f8aa5ef2

For more information, contact [email protected]

© 2018 Infosys Limited, Bengaluru, India. All Rights Reserved. Infosys believes the information in this document is accurate as of its publication date; such information is subject to change without notice. Infosys
acknowledges the proprietary rights of other companies to the trademarks, product names and such other intellectual property rights mentioned in this document. Except as expressly permitted, neither this
documentation nor any part of it may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, printing, photocopying, recording or otherwise, without the
prior permission of Infosys Limited and/ or any named intellectual property rights holders under this document.

Infosys.com | NYSE: INFY Stay Connected