A NASSCOM-CMR REPORT

e ll i g e n c e
fi c i a l I n t
Arti , F i n a n c i a l
B a n k i n g e c to r
fo r u ra n c e S
c e s & i n s
Serv i

What CIOs and IT decision makers from

India’s top BFSI companies feel about
leveraging AI.
Artificial Intelligence for BFSI Segment 2
Table of Contents
I Message
II Preface
III Acknowledgements
Executive Summary 07
Key Takeaways 08
Section 1: Perceptions and Plans for AI 09
Section 2: Sector Specific Analysis:
• Banks 18
• Financial Services 20
• Insurance 22
Section 3: Global and Indian Use Cases of AI in Action 24
Section 4: Issues Hindering AI Deployment 29
Section 5: Deployment Challenges and Best Practices 32
Section 6: Regulatory Updates 36
Section 7: Recommendations on the Way Forward 40
Section 8: Annexures:
• I-Open Source and Commercial Tool-kits 43
• II-Glossary of Common AI Terms 46
• III-References 48

Artificial Intelligence for BFSI Segment 3

 Message

Artificial Intelligence for BFSI Segment 4I

 Foreword

Foreword
There’s an entire suite of cutting-edge technology which has brought on the Fourth
Industrial Revolution, and unsurprisingly enough, AI has seen proliferation like no
other to stake its claim at the vanguard. That, BFSI would be impacted too in no
small measure, was a certainty, though we cannot look at AI applications as one
whole for the BFSI sector. For instance, the Insurance sector will see a different kind
of adoption than Banks or Financial Services. The report goes on to explain how.
The variation is interesting. The survey reveals that the Insurance sector is far
more serious about using AI for customer support – a whopping 85%, at that. In
comparison, Banks and Financial Services show a healthy 71% & 74% respectively.
The clear objectives for AI adoption are: personalized customer experience, having
error-free back-end processes coupled with faster turn-around-time. In this, it may
be a tad worrying that security and compliance appeared to be the last priority,
even though it garnered 49% of the votes. The question remains – how long can
we not have AI to search for needles in a digital haystack in case of the other 51%?
Rising NPAs and frauds in the Banking sector don’t give us that assurance, frankly.

Wherever, the number of positive respondents fared closer to 50 % than say 85, one might reasonably ask - so what are really the
concerns? One, it is the lack of maturity of standards. This creates silos of options which makes vendor identification a major task.
Two, Machine Learning depends on huge amount of data – over 54% of the respondents have found it challenging to train their deep
learning systems with lesser amount of data. Interestingly, cost or even convincing the top management aren’t major concerns, as
is appropriate talent availability. Business leaders do not need too much of convincing anymore about what AI can do for them, but
getting the right people to do the job is still a veritable ask.

TRAI estimates that by 2020, India will have 730 million internet users and most will access it through mobile phones. Increased digi-
tal channels present both an opportunity and a threat. A massive rise in customer base is most obvious but at the same time all these
users will come in with heightened levels of expectation. Unless met, they will drop off the radar very quickly. The e-com and social
media apps have already set the bar incredibly high so users will expect nothing less in terms of usability from banking & insurance
apps as well.

That brings us to the impact, AI will have on jobs. Like every new wave, AI too will result in inevitable job cuts but that’s temporary.
Soon enough, new opportunities will be created which will require a higher application of human skills. And, humans can be more
gainfully employed. Monotonous and repeatable jobs are obviously at high risk but it is unlikely that AI systems can be left completely
unattended without any human monitoring even if they were to achieve very high degree of perfection of replicating humans. Unde-
niably, Robo-advisory services is the future but there’ll always be a section of society which will prefer humans to robots – particularly
the elderly citizens – when it comes to seeking financial advice. In reality, in a data-driven environment like the one we have entered
into, traditional IT will simply not be able to deal effectively with the volume, velocity and variety of data being generated. Particularly
variety. Inferences will have to be drawn from multiple data sources & sets which will be in disparate formats – that’s where AI steps in
seamlessly without much competition.

The peculiar approach to BFSI is that decision making often goes beyond the traditional consideration of speed, functionality, RoI
etc. For instance, in insurance underwriting or personal financial advice, there’s a sizeable element of subjectivity. Incorporating this
in AI systems may be tougher than we think. That’s because when the answers aren’t straightforward, eyebrows may be raised often
leading to an ethical dilemma. Towards addressing this, IEEE has been working on Ethically Aligned Design which necessitates that
transparency, accountability and algorithmic bias et al, will have to be considered right at the time of design and not merely as an
afterthought.

There you have a rather comprehensive report, waiting to be read. Granted that all answers may not be forthcoming but it aims to
cover as much ground as possible without being too technical. A preface or a foreword is expected to whet the reader’s appetite to
turn the pages. I hope yours is suitably whetted, and you are ready to find out at length, the Impact of Artificial Intelligence on
BFSI.

Happy reading!
Debjani Ghosh, President
NASSCOM

Artificial Intelligence for BFSI Segment II5

 Acknowledgement

Acknowledgement

CMR would like to thank the following in providing their precious time and inputs in making this report
possible:

Prasanto K Roy
VP & Head, Internet, Mobile & E-commerce Council, NASSCOM

Rakesh Kumar
Director – Research, NASSCOM

Neha Rawal
Deputy Manager, NASSCOM

Thomas George
Senior VP and Head, CMR

Anil Chopra
VP-Research and Consulting, CMR

Artificial Intelligence for BFSI Segment III

6
 Executive Overview

Executive Summary

Artificial Intelligence can have a profound impact on just about every industry, and
the BFSI segment is no exception. According to our survey of IT decision makers
from this segment, 75% of the respondents felt that there’s a strong need for AI in
BFSI.

In fact, there’s a high level of agreement that AI will have a huge impact on various
aspects of BFSI operations. Whether it’s banking, financial services, or insurance,
over 50% of respondents everywhere agree that AI will dramatically improve op-
erational efficiency and fundamentally transform core financial processes.

Using AI to enhance customer support and offer a more proactive and personal
experience seems to be the primary business objective across the board, followed
by management of back-end business processes to reduce human error.

Whether AI will replace manpower and the humble IVR based menu system is still a contentious issue, and most respondents only
partially agree that this will happen. Other reports by prominent advisory firms say that AI will create more job opportunities than
losses.

While the BFSI sector is extremely positive about using AI, there are some concerns. Over 54% of the respondents feel that AI based
solutions lack common standards to differentiate between different vendor offerings. With so many AI technologies to choose from
and no common framework to evaluate various offerings, it becomes a hindrance to their deployment plans. It also becomes difficult
to determine how an AI system could be plugged into their IT infrastructure and the outcomes it would deliver.

Another key issue many respondents faced is lack of sufficient use cases of AI in BFSI. This report delves into ample global and Indian
use cases of AI along with some of the best practices that could be followed while deploying.

Though it’s a very new technology, some respondents claimed that they were already using AI based applications extensively, and
over 64% are either seriously evaluating its usage or begun to introduce AI based products.

This report analyzes how AI is shaping up in the BFSI space. It covers the business objectives, challenges, and future plans of IT deci-
sion makers around AI. The eco-system of AI based startups is growing in India and this report covers some of them.

AI technology is here to stay and this report can act as a good reference point for decision makers and implementors from the BFSI
space to start their journey.

Anil Chopra, VP Research & Consulting

CMR

Artificial Intelligence for BFSI Segment 7

 Key Takeaways

Key Takeaways
Choosing Between AI Offerings is a Challenge due to Inadequate
Standards:
A majority of Indian CIOs and IT Decision makers we surveyed find it difficult to distinguish between AI offerings from different
vendors. Some are not confident of the outcome they would deliver, while others don’t understand how to plug it into their existing
setup. There’s no standard framework available to guide them with their selection process.

There’s a Need to ‘Indianize’ AI Products:

Some CIOs feel that the Indian regulatory framework doesn’t allow capturing personal information of individuals. As a result, it be-
come challenging to truly understand customer needs. AI tools for data capturing need to factor this in and help the BFSI segment to
truly understand customer behavior.

Proactive and Personal Customer Support are Top Business Objectives:

The clear business objective of IT Decision Makers of Indian BFSI companies is to use AI and offer a more proactive and personal
customer experience at a lower cost.

Training the Deep Learning system an issue while deploying AI:

Data management is a key issue while deploying AI, whether it’s to train the deep learning system to work efficiently with less data or
structuring the data as per the needs of an AI system.

Artificial Intelligence for BFSI Segment 8

Section 1:

Perceptions
and Plans for AI

Artificial Intelligence for BFSI Segment 9

 Perceptions and Plans for AI

Artificial Intelligence for BFSI Segment 10

 Perceptions and Plans for AI

There’s a Strong Need for AI in BFSI “Bank of America’s

There’s strong agreement among IT decision makers of BFSI companies that AI is badly needed
Erica pre-empts
in this segment to dramatically improve operational efficiency and fundamentally transform this customer needs
sector’s core financial processes. and offers relevant
However, this can only happen if there’s mass adoption of AI. One issue that’s possibly hamper- advice, e.g. informs
ing this is a lack of common standards in AI based solutions, according to a good 54% of the of low balance
respondents. Once these are in place, then AI can do wonders. In fact, 43% of the respondents
strongly feel that AI will replace a bulk of the tele-callers for routine customer interactions. This
by checking
may sound like a very strong statement and only time will tell whether this will happen or not. withdrawal history
However, going by the fact that the highest number of deployment cases for AI are chatbots (ex- and recommends
plained later), and that too for customer interactions (according to 42% of the respondents),
there’s a high chance that this will happen. solutions.”
See Section 3 for details.

Artificial Intelligence for BFSI Segment 11

 Perceptions and Plans for AI

Survey Methodology
The survey to gauge the impact of AI on India’s BFSI was conducted purely online. It
started with an invitation being sent to key IT decision makers from the BFSI compa-
nies in India. These included Indian public and private sector as well as multi-national
BFSI companies operating out of India. We received 83 responses, comprising of over
30+ responses each from banks and financial services companies, 13 from the insur-
ance sector, and 6 from consulting and other organizations working closely with the
segment.

The results have been analyzed overall as well as for specific BFSI sectors to under-
stand how the usage of various AI technologies differs between them. In case of deeper
analysis for each sector, we’ve only taken the responses from that sector.

The purpose of doing this survey is to understand the importance and plans that vari-
ous BFSI companies have around various AI technologies, and to provide a compre-
hensive report that not only provides the survey findings but solutions to key issues
and questions that the IT decision maker community wants answers to.

Apart from the survey findings, the rest of the report is based on secondary research.

How will manpower issues be tackled is a different issue and not covered in this report, but the “CapitalOne’s
good news is that a good 39% of the respondents strongly feel that AI can bring highly profitable
growth, and 52% of them are from the banking segment.
ENO voicebot
can be enabled on
In fact, days are also numbered for the humble IVR system as over a third of the respondents feel
Amazon’s Alexa,
that AI will replace it in the next two years, out of which the highest percentage of respondents
are from financial services segment. so customers can
interact with it and
Tele-callers might have become targets to be replaced by AI, but not manpower that’s handling
other low-value back-office tasks. The good news is that less than a third of the respondents be-
ask for account
lieve that AI will replace them. This is understandable because according to various global case balance, pay credit
studies (published later in this report), AI is deployed to aid employees in their day to day func- card bills, etc.”
tioning and not replace them. There is a one-off case where manpower was reduced. See Section 3 for details.
Tele-callers might have become targets to be replaced by AI, but not manpower that’s handling other low-value back-office tasks.
The good news is that less than a third of the respondents believe that AI will replace them. This is understandable because accord-
ing to various global case studies (published later in this report), AI is deployed to aid employees in their day to day functioning and
not replace them. There is a one-off case where manpower was reduced.

ITDMs seem to be putting their money where their mouth is when it comes to AI deployment. A good 36% of the respondents are
seriously evaluating how it can be used, while another 28% have begun to introduce AI based products and services. There’s a
nominal 6% who’ve already put AI based technologies to extensive use, while another 8% have completed all their research and
are now planning their pilots and initial roll-outs.

Interestingly, there’s still about a fifth of the respondents who’re sitting on the fence and have no plans of using AI yet. Possibly
they’re still waiting and watching for AI to pick up more steam.

Interestingly, a similar pattern exists when we drill into individual sub-segments of BFSI. The highest percentage of respondents
are seriously evaluating AI, followed by those who’ve begun to introduce. The third highest comprises of those who’ve no plans
for now, while the last two are those who’ve completed their research and are either ready for a roll-out or are already using AI
extensively.

All said and done, the BFSI segment is seriously evaluating the use of AI based applications across their products or services.

Artificial Intelligence for BFSI Segment 12

 Perceptions and Plans for AI

Proactive and Personal Customer Support are Top Business Objectives

The clear objective that stands out in this segment for using AI is to offer a more “Citigroup’s AI based
proactive and personal customer experience at a lower cost, followed by man-
agement of back-end business processes to reduce human error and improve
system uses statistical
the turn-around-time for manual processes. models to pass regulator
stress tests. These tests
53% of the respondents aim to use AI for their marketing functions, where
they’d like to track consumer behavior so as to offer customized products. This study a bank or financial
is in a way a subset of the first priority of offering superior customer experience. institute’s ability to do
Using AI for security and compliance appeared to be the last priority for the
ITDMs, but it’s still important nevertheless, as it garnered 49% of the votes.
scenario analysis.”
See Section 3 for details.

Artificial Intelligence for BFSI Segment 13

 Perceptions and Plans for AI

AI for Fraud Detection and Prevention Top Deployment Priority

Given the pace at which online fraud and malware attacks is increasing, it’s becoming increasingly difficult for traditional security sys-
tems to provide the required level of protection. Moreover, companies can’t indefinitely keep hiring more people into their security
teams to fight new types of fraud, nor can they continue manually checking false positives. It simply becomes too costly. That’s where
AI comes in, which can use complex machine learning algorithms for fraud detection. These systems can be supervised to recognize
fraudulent activities and learn continuously. This way, it will automatically recognize such activities in the future.

The BFSI segment has therefore become more serious about using AI for fraud prevention. A good 48% of the respondents said that
they plan to deploy AI for fraud detection and prevention over the next six months to one year.
This is followed by none other than voicebots for customer support, which correlates to our earlier point where the respondents
strongly agree that AI will replace IVRS system in the next two years.

Chatbots is an area of maximum interest in this segment. The highest percentage of respondents have already deployed it fully, and
an equal number are running pilots, while another 46% plan to deploy it in the next six months to one year.

Robotic Process Automation is also garnering of a lot of interest among respondents in this segment with the highest percentage of
them already running pilots, and about a third of them planning to deploy it over the next six months to one year.

Artificial Intelligence for BFSI Segment 14

 Perceptions and Plans for AI

Different Flavors of AI
Chatbots or Voicebots:
self-learning programs for intelligent conversations with humans over chat or
audio; Available 24x7 and very easy to use but require long time for training.

Robo Advisors for Financial Products:

Online platforms that use algorithms to offer financial advice, re-invest dividends,
automatic portfolio creation/re-balancing, etc; Require minimal to zero human
CRM, intervention.

Marketing and Personalized Financial Services:

Robo-advisors to monitor customer goals and suggest stocks or bonds to buy/sell;
Customer Gives personalized attention to customers irrespective of their risk appetite.

support Smart Wallets:

Intelligence added to mobile wallets for smart services like chat, booking of bus
tickets, cab, events, movies, utility bill payments, etc.

Emotion AI:
A branch of AI to enable machines to detect human emotions with advanced
facial and voice recognition technologies.

Fraud Detection and Prevention:

Minimize need to add continuous manpower to detect and block security attacks;
These platforms use machine learning to automate the process.

Compliance Monitoring:
Security
Use AI to examine lengthy documents and flag potential issues in seconds, which and
would otherwise take many hours.

Intelligent QRC with RegTech:

Compliance
A new segment of AI startups that specialize in helping companies remain
compliant, e.g. ensure no document is missed out while filing something, do risk
mitigation by monitoring customer behavior from empirical data.

Robotic Process Automation:

The use of software robots to take over high volume, back-office processes and
repetitive tasks to save time, enhance efficiency, and increases accuracy.

Insurance Under-Writing:
Using AI for faster and more accurate risk assessment and pricing to measure the
risk exposure and determine premium to be paid by a customer.

Algorithmic Trading:
Back-End AI for high-frequency trading where inputs are taken from multiple financial
BPM markets to make investment decisions in milliseconds. Reports suggest that over
70% of trading worldwide today is being managed by algorithms.

Investment Research:
AI to guide investors on stock picking decisions. It can help cover more companies
in exchanges all over the world, do their research and portfolio management.

HR:
AI to save hiring managers’ time in various recruitment processes e.g. engage with
new recruits, shortlist resumes from social media sites, pre-screen candidates
over chat, determine candidate drop out chances, etc.

Artificial Intelligence for BFSI Segment 15

 Perceptions and Plans for AI

AI’s Impact on Manpower

Will AI lead to massive job cuts?

This is possibly one of the biggest concerns that most people have with AI. There’s no single ‘Yes’ or ‘No’ answer to whether it will
cause massive job cuts. The way to look at the situation is that some jobs always become redundant every time a new automation
technology emerges. The same will happen with AI. On the positive side, while AI might make some jobs redundant, it will also create
new job opportunities over a period of time. Organizations need to gauge the extent of this impact on their manpower during this
interim period.

AI would affect monotonous/repeatable jobs like answering common, repeatable customer queries According to
or scanning a document for certain types of information. Since the outcome of such tasks is known,
AI could take over from humans after some training, but not entirely. In some global chatbot
Gartner, by 2020,
deployments for instance, companies initially hired an entire helpdesk to monitor and intervene in AI will eliminate
all chatbot conversations so that it could be trained. Even after the training, the AI system can’t be
1.8 million jobs,
left to work on its own. So, while a part of the workforce would become redundant, some of them
would still be required to watch over the AI system. but also create 2.3
million jobs.
The other thing to understand is that AI systems are meant to assist humans in reducing their time
and effort for certain tasks, and not replace them. In fact, we’ve entered a data-driven economy,
where massive data is being generated from multiple digital channels. It’s humanly impossible to gather, store, monitor, and analyze
it all. AI and complex algorithms are the need of the hour to process this data for humans. In fact, according to a Gartner survey, AI will
create more jobs than lost. The research firm says that by 2020, AI will eliminate 1.8 million jobs, but it will also create 2.3 million jobs.

The bottom line is that just like any new technology, AI might initially make some jobs redundant, but this will be a temporary phase
until new job opportunities emerge. In fact, initially, there might be a spurt in the number of jobs, as observed in countries where AI
deployment is moving more rapidly. Once the system becomes good enough to handle certain tasks, then the trainers could be re-
deployed.

Artificial Intelligence for BFSI Segment 16

Section 2:

Sector Specific
Analysis:
• Banks
• Financial Services
• Insurance

Artificial Intelligence for BFSI Segment 17

 Sector Specific Analysis: Banks

AI in Banks

Top Business Objectives of using AI:

71%
Customer Support
68%
Back-end BPM
58%
Security &
55%
Marketing
Compliance

The primary business objective of banks for using AI is to offer a more proactive/ “Citigroup’s AI based
personal customer experience, followed by automating back-end processes to system uses statistical
reduce human error and improve the TAT (turnaround time) for manual pro-
cesses. The third most important business objective for banks is to monitor all models to pass regulator
processes and data for regulatory compliance and security. The top fourth busi- stress tests. These tests
ness objective with AI is to use it to use it for marketing wherein they’d like to
track consumer behavior so that customized products can be offered to them.
study a bank or financial
institute’s ability to do
scenario analysis.”
See Section 3 for details.

Artificial Intelligence for BFSI Segment 18

 Sector Specific Analysis: Banks

Voicebots Top the Deployment List in Banks in Next 6 Months to One Year

The pattern of deployments in banks specifically is a little different. For instance, voicebots are on top of the list of deployments over
the next six months to one year, followed by chatbots and then fraud detection and prevention services. Incidentally, the highest
percentage of pilots are also running for fraud detection and prevention, which probably indicates the urgency of this requirement.
Robotic Process Automation has the second highest percentage of pilots running.

Interestingly, a good 13% of the respondents said that they’ve already fully deployed AI for fraud detection and prevention, personal-
ized financial services, smart wallets, and robotic process automation. The largest number of deployments however, are of chatbots at
16%.

The Top 3
S. NO Issues Hindering Deployment Challenges while Deploying

1. Finding manpower with experience in a.  Training the deep learning system to work
cognitive systems efficiently with less data
b.  Getting right answers from the AI based
system

2. Identifying the right AI vendor Customers prefer interacting with humans

3. Immaturity of industry standards a. Addressing Security and privacy issues

around AI based systems b. Structuring data according to the require-
ments of AI based system

Artificial Intelligence for BFSI Segment 19

 Sector Specific Analysis: Financial Services

AI in Financial Services

Top Business Objectives of using AI:

74%
Customer Support
59%
Marketing
56%
Security &
50%
Back-end BPM
Compliance

Just like in banking, financial services have also put customer support as their “PayPal’s in-house,
top use case for AI. The second largest use case however is to use AI for market-
ing and so that they can track consumer behavior and offer customized product. open-source based fraud
The top third business objective remains the same as in banking, which is to detection engine has been
use AI for security and compliance. The last business objective is to use AI for
automating back end processes.
running for ten years, has
cut their false positive rate
by half”
See Section 3 for details.

Artificial Intelligence for BFSI Segment 20

 Sector Specific Analysis: Financial Services

Chatbots of Maximum Interest in Financial Services Companies

Chatbots are the top priority over the next 6 months to one year amongst financial services organizations, followed by fraud detec-
tion and prevention and voicebots. Robo-advisors for financial products are another technology of interest among 47% of the
respondents from this segment.

The highest number of pilots in this segment are running for robotic process automation at 18%. Interestingly, only 29% of the re-
spondents said they’re likely to deploy AI based personalized financial services. Maybe the cost of deploying it is not feasible for this
segment.

The Top 3

S. NO Issues Hindering Deployment Challenges while Deploying

1. Immaturity of industry standards Getting right answers from the AI based

around AI based systems system
2. Identifying the right AI vendor Customers prefer interacting with humans

3. Finding manpower with experience in Addressing Security and privacy issues

cognitive systems

Artificial Intelligence for BFSI Segment 21

 Sector Specific Analysis: Insurance

AI in Insurance

Top Business Objectives of using AI:

85%
Customer Support
77%
Back-end BPM
38%
Marketing
15%
Security &
Compliance

Like banks and financial services, customer support came out as the top use for
AI in the insurance sector as well, and the second highest business priority was
“ICICI Lombard
the same as banks, i.e. automating the back-end processes. The third business leverages Robotic Process
priority for insurance was to use AI for their marketing efforts so that they could Automation to automate
understand their customers better and offer them more customized products.
The last business priority that came out from the survey was surprisingly, to use mundane tasks, clearing
AI for security and compliance. 100 transactions in 5
minutes as against 4 hours
earlier, increasing cross-
sale by 2%, to name a
few.”
See Section 3 for details.

Artificial Intelligence for BFSI Segment 22

 Sector Specific Analysis: Insurance

Fraud Detection and Prevention Top Deployment Priority in

Insurance Sector

Unlike banking and financial services sectors, the maximum emphasis in the future for this segment seems to be on using AI for fraud
detection and prevention, followed by, of course, insurance underwriting. This is contrary to the priority given by this sector for the
business objectives of using AI.

The next set of AI technologies of interest to this segment are robo advisors for financial products and voicebots. Considering the
number of insurance policy products that this sector has, such advisors can really help insurance companies automate the process of
offering the right insurance product to the right customer.

Chatbots have an interesting pattern here, with 31% of respondents each having fully deployed it, running pilots, and planning to do
so in the future. Chatbots are therefore the most popular AI technology in this segment.

The Top 3

S. NO Issues Hindering Deployment Challenges while Deploying

1. Identifying the right AI vendor Training the deep learning system to work ef-
ficiently with less data

2. Immaturity of industry standards Getting right answers from the AI based system
around AI based systems

3. Finding manpower with experience in Getting the AI system to interpret questions

cognitive systems correctly

Artificial Intelligence for BFSI Segment 23

Section 3:

Global and Indian

Use Cases of AI in
Action

Artificial Intelligence for BFSI Segment 24

 Global and Indian Use Cases of AI in Action

As customers increasingly prefer to interact online, BFSI companies are closing their branches because they’re becoming costlier to
maintain. This reduction also translates into reduced customer outreach, which is why Traditional online portals and mobile apps
can only address ‘the reach’ part of the problem. They lack the ability to deliver that personal human touch and attention that branch
executives would give to customers. AI based systems can fill in this gap and much more. Here are some global and Indian use cases
of AI for customer support, back-end process automation, security risk and compliance, and a few others.

Chatbots and Voicebots Notable Chatbot Startups

Bank of America’s Erica drills deep into customer data to pre-empt
their needs and suggest solutions. It’s voice and chat driven to manage
the customers’ money. Built by a team of 100 people, Erica is unlike
other chatbots that answer simple questions. It pre-empts a customer’s
needs to offer relevant advice. For instance, it analyzes customers’ Application: Customer Support
withdrawal history to gauge their spending patterns and inform them if Prime Industries Targeted: Bank and finan-
they’re running low balance. It even suggests solutions like transferring cial services Notable Clientele: Partnered with
funds, or sends notifications about where they can save money. Axis bank to launch chatbots
AI Specific Solutions used: Chatbots
Commonwealth Bank of Australia’s Bot performs 200+ banking tasks
for customers like activating cards, paying bills, sending bank state-
ments, etc. The bank plans to first roll it out to its 6.2 million NetBank
and CommBank app users, and then to the rest.
Application: Customer Support
Amazon’s Echo and Facebook’s Messenger platforms extend Prime Industries Targeted: All industries
their voicebot/chatbot technologies to companies to enhance their Notable Clientele: Built 100+ chatbots
customer engagement. BFSI companies can’t always create in-house AI AI Specific Solutions used: A domain classi-
powered software from both cost and time point of view. That’s why, as fier to figure out what user is talking about;
of writing this report, Facebook crossed 100,000 bots on its messenger Named Entity Recognition module to detect
platform, while Amazon’s Alexa voice platform crossed 15,000 “skills” relevant entities from a chat; Graph based
or voice-powered apps. These skills run on Echo products like Amazon chatflows to determine context of a conversa-
Echo speaker, and their number has been rising rapidly. tion; Deep learning layer to learn from previ-
ous conversations; and an ML orchtestration
MasterCard uses FB Messenger’s chatbot to allow its customers to layer to respond back in a conversation
review their purchase history, spending habits and account balance.

Artificial Intelligence for BFSI Segment 25

 Global and Indian Use Cases of AI in Action

CapitalOne, a diversified bank and fortune 500 company added

its voicebot ENO’s skill to Amazon’s Alexa. Customers simply enable
the skill on their Alexa app and start interacting with it, like asking for
account balance, paying credit card bills, etc. ENO is also available over
mobile based chat. Application: Chatbots for most all businesses
Prime Industries Targeted: All industries
Fullerton India Credit Company Ltd.’s ASHA is a loan acquisition Notable Clientele: HDFC Bank, ICICI Bank,
and fulfillment chatbot for Facebook Messenger that customers can ICICI Lombard General Insurance, HDFC Life,
chat with to apply, structure, submit documentation, and receive loan Club Mahindra, T-Mobile, Manipal Hospitals,
approvals, without downloading any other application. Elsevier, Edelweiss, SBI Card, Godrej Proper-
ties, Airtel, CAMS
HDFC Life Insurance launched an AI based chatbot that takes
customers through a 60 seconds quiz and then computes their Insur-
ance Quotient for the company’s various insurance products, like tax
planning, retirement, health, etc.
Application: Chatbots for Order Booking
HDFC Bank’s Electronic Virtual Assistant, or EVA has become In-
Prime Industries Targeted: All industries
dia’s largest banking chatbot. Deployed in March 2017, EVA successfully
Notable Clientele: Over 100K downloads
addressed over 2.7 million customer queries within six months of its
on Android App Store; Tied up with Infosys
deployment. Over 5.3 Lakh unique users who held 1.2 million conversa-
Finacle to offer chat-based commerce to
tions. EVA is available on all digital platforms of the bank, including the
banking customers
website, mobile site and a dedicated customer portal. Besides queries,
AI Specific Solutions used: Personal
Eva has also generated a lot of business leads for the bank.
Assistant
Emma, OCBC Bank Singapore’s AI-based Chatbot answers home
loan queries and generates leads. It has handled over 20,000 enquiries
since January 2017 and converted 10% of them into mortgage sales
prospects. Besides answering queries, it also mimics actions of a loans Application: Chatbots for Sales and Marketing
advisor by asking questions to determine an applicant’s loan eligibility. Prime Industries Targeted: All industries
Emma required three months of training to become ready to handle Notable Clientele: YES Bank, Manchester City,
queries. USL, Yamaha
AI Specific Solutions used: Human fallback,
Policybazaar’s Chatbot, PBee is an AI-powered chatbot to sell in-
natural language processing, live training,
surance online. As a result of the chatbot, the company’s chat volumes
real-time analytics, support for API call, google
increased 5x, and time to respond reduced by 25%. It handles over 60%
sheet, CRM integration, custom code
of all messages. The chatbot is powered using Google’s DialogFlow.
The company is planning to extend the chatbot to its health insurance
and investment customers as well.

“We’ve identified AI as a key ‘FutureNow’

focus for YES Bank. We were among Application: Chatbots for Sales and Marketing
the first banks to deploy AI-enabled Prime Industries Targeted: All industries
chatbots across multiple social Notable Clientele: YES Bank, Manchester City,
media and bank plaforms, and USL, Yamaha
have now also developed use- AI Specific Solutions used: Human fallback,
cases to help analyze our huge natural language processing, live training,
volume of data and generate real-time analytics, support for API call, google
insights, analytics and predictive sheet, CRM integration, custom code
solutions. We use AI-ML to get quicker
and more reliable credit scores, improve
our asset quality, and reduce NPAs. AI in RPA has helped us auto-
mate high volume repetitive mid- and back-office operations. We
Application: HR: Talent Acquisition and
are actively working with fintechs with expertise in NLP as part
workforce optimization
of our YES Fintech Accelerator, which is helping both internal use-
Prime Industries Targeted: Recruiters
cases as also client acquisition and servicing. We want to build a
Notable Clientele: HCL, Wipro, Homergize,
self-sufficient digital channel for customer interaction, including
Fourth Frontier
a robo-advisory for investments. What I look forward to seeing in
AI Specific Solutions used: Natural Language
India is a more players whose engineering teams we can access
Processing technologies with Data Science
and work with, and most importantly, a more mature ecosystem
approach and an Intelligence layer
for AI.”
--Anup Purohit, CIO, YES Bank

Artificial Intelligence for BFSI Segment 26

 Global and Indian Use Cases of AI in Action

AI in Back-End BPM Notable Data Analytics

Startups
JP Morgan Chase processes contracts faster with fewer mistakes
after creating a bot called COIN. Short for Contract Intelligence, the
bot uses machine learning techniques to help the bank’s lawyers
Application: Analytics for better customer
analyze around 12,000 contracts for commercial loan agreements per
engagement, performance reporting, BI
year. COIN has cut-down loan-servicing mistakes and already saved
reporting, and compliance reporting
the bank a whopping 360,000 man-hours of labor. What would take
Notable Clientele: HUL, HDFC Bank, Kotak
hours can now be done in minutes.
Mahindra Bank, dun&bradstreet, ICICI
Lombard, Motilal Oswal, Network18, HDFC
ICICI Lombard Leverages Robotic Process Automation to automate
Securities
mundane tasks that were hampering productivity, causing human
errors as their business grew and transaction volumes increased. The
results have been astonishing, with 100 transactions being cleared
in 5 minutes as against 4 hours earlier; Insurance policy digitization
time reduced for fire and group health insurance policy from 4 hours Prime Industries Targeted: All of BFSI
to 1 hour per policy; Time to respond to user management requests Application: Analytics: Data Analytics
has come down from 15 days to just 2 days. Similarly, time to send Notable Clientele:
out quotes reduced from 5-6 hours to less than 3 minutes, customer AI Specific Solutions used: Data stream-
satisfaction index improved from 97% to 99%, with an increase in ing and mapping; Genomics & psychology
cross-sale by 2% and a similar percentage of new agents and brokers pattern building; Cross industry intelligence;
getting onboarded. Prediction intelligence; Automated decision
making; Text & language intelligence;
ANZ Wealth ties up with University of Technology in Sydney to
explore AI based insurance under-writing and claims processing.
The collaboration aims to use big data, client behavior modelling, text
mining, natural language programming and social media predictive Prime Industries Targeted: Any industry
analysis for the job. Application: Personalized Analytics on
mobile devices based on department and
Japan’s Fukoku Mutual Life Insurance Company is planning to information preferences
implement IBM’s Watson and slash 30% of its staff from the payment AI Specific Solutions used: AI to understand
assessment department. While manpower reduction isn’t a good business context and learn user preferences
sign, several other companies took a queue and didn’t announce any to share right data to right user at the right
manpower adjustments for a similar type of automation. time

HDFC Life Insurance’s AI-based insurance Email Bot

India’s leading private sector life insurance company, HDFC Life,
launched an insurance Email Bot in 2017, named SPOK. The Email Bot
Prime Industries Targeted: All industries
can understand, categorize, prioritize and respond to customer emails
Application: Data Analytics
within milliseconds. Besides answering queries quickly, SPOK also
Notable Clientele: FIS, neilson, happiest
helps generate deeper insights into customer needs by identifying
minds, Alten Calsoft Labs, First Media, ESPN
patterns in their interactions. The email bot uses Artificial Intelligence
Cricinfo, Hortonworks, Cloudera, EA
and Natural Language Processing and mimics human cognitive abili-
AI Specific Solutions used: An Open Cogni-
ties in reading, comprehending, interpreting, and conversing.
tive platform that interfaces between data
analytics, deep learning, predictive analytics,
and IoT in real time.

Prime Industries Targeted: All of BFSI

Application: Intelligent Big Data Analytics
Infrastructure; improve decision making for
problems like churn, fraud and predictive
analysis
AI Specific Solutions used: Xurmo’s Machine
Learning algorithms can be used to create a
custom sentiment analyzer to predict senti-
ment of a news point of a particular company.

Artificial Intelligence for BFSI Segment 27

 Global and Indian Use Cases of AI in Action

Risk and Compliance Other Notable AI Startups

Deutsche Bank analyses conversations for risk and compliance
monitoring. It uses AI to go through audio and video recordings of
interactions between its bankers and clients and tries to identify
banking terms that are normally monitored by auditors. Prime Industries Targeted: Retail, Banking
Application: Other: Intelligence ATM surveil-
Citigroup uses an AI based system to pass the US federal govt.’s lance
stress test. It uses statistical models to pass regulator stress tests. Notable Clientele: Leading banks are evalu-
These tests study a bank or financial institute’s ability to do scenario ating for ATM monitoring for incidents like
analysis. infrastructure tampering, unexpected human
actions, assault, or other suspicious activity.
Danske Bank has used an AI Framework to increase its fraud
detection rate by 60%. The Danish bank was earlier creating rules
manually for fraud detection, due to which their false positive rate
was as high as 99.5%. Investigating such cases was therefore both
time consuming and costly. The bank created a framework and Prime Industries Targeted: All industries
advanced machine learning models that could detect fraud from Application: Other: Uptime and saving en-
millions of transactions. As a result, they were able to reduce the false ergy consumption in Bank ATM machines
positive rate by 50% and increased the detection of actual fraud cases Notable Clientele: large quick service res-
by 60%. taurant chains, bank ATMs and retail stores
having anywhere from 200 to over 1000
PayPal developed an in-house fraud detection engine using Open stores
Source tools, which works on a combination of human intelligence,
machine learning, and data mining. As a result, they’ve cut their false
positive rate by half. The company has been doing it for over 10 years
now.
Prime Industries Targeted: All of BFSI
Application: Process Automation
AI in Customer Service Notable Clientele: Working with one of the
world’s biggest consultance firms; New York
AmeriTrade, a US-based brokerage firm facilitates stock trading and London based banks, global insurance
through Twitter. Their software goes through tweets and provides companies, and leading manufacturing firms.
investment advice to customers, including news on companies, trad- AI Specific Solutions used: Personal
ing quotes, spikes in volume, etc. Non-AmeriTrade customers can also Assistant
request stock information, but only customers can initiate trades.

SmartBiz Loans’ AI tool offers loan advice to every SME customer

irrespective of their size, which was earlier not viable. San Francisco Prime Industries Targeted: Recruiters
based SmartBiz Loans’ AI-powered education tool advises SMEs Application: HR: Talent Acquisition and
about their loan readiness, gives them a loan ready score and even workforce optimization
advises on how to improve it. SMEs just upload their tax documents Notable Clientele: HCL, Wipro, Homergize,
to the platform, which then uses AI and ML to provide financial Fourth Frontier
advice. AI Specific Solutions used: Natural Lan-
guage Processing technologies with Data
IndiaFirst Life Insurance’s virtual assistant provides real-time cus- Science approach and an Intelligence layer
tomer resolution and support on insurance transactions. Called IRIS
(IndiaFirst Responds), the bot answers queries related to the com-
pany’s products, fetches premium statements and notices, answers
fund value inquiries, and provides updates on application status and
surrender value inquiry.

Artificial Intelligence for BFSI Segment 28

Section 4:

Issues Hindering
AI Deployment

Artificial Intelligence for BFSI Segment 29

 Issues Hindering AI Deployment

Top 3 Issues Hindering AI Deployment

Inadequate standards around AI based systems 66%

Identifying the right AI vendor 65%

Finding manpower with experience in cognitive systems 59%

High cost of investment 43%

Convincing top management to deploy AI 24%

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

Inadequate Standards to Differentiate AI Products is the Key Sore Point

Inadequate standards to differentiate between differnt AI products from different vendors emerged as the topmost issue hindering
their deployment. This is because even though there’s no dearth of AI based solutions out there, there’s no standard way of differen-
tiating between their capabilities. “If a particular AI tool claims to be the best in its class, there’s no real way to judge that claim”, said
Mridul Sharma, EVP and Head-Technology, IndusInd Bank. “To draw a parallel, if you want to buy a car, you would search through
various models, read through their specifications, features, etc. and then decide which one fits your needs. You wouldn’t buy an SUV
if you needed a hatchback.”, he added. The same case is there when choosing traditional software. You can choose from a standard,
professional, or enterprise edition by comparing their capabilities. Unfortunately, there are no common standards or checklist in AI
tools to compare against.

Chatbots for instance exist in many variants, and every vendor will claim theirs to be the best. However, some of them might simply
be keyword based, while others will offer NLP, etc. It’s also difficult to gauge whether they will scale up to the desired level or not.
Since there are no set standards in place, the only way to see if they really deliver is to implement and train them. This would waste a
lot of precious time and resources.

The other issue due to immaturity of standards is difficulty in integrating AI tools into an existing setup. Most AI tools don’t have the
required features to integrate with multiple channels and applications in an organization.

Lastly, there’s no regulatory framework to control the risk from AI tools. Today, they’re capable of doing anything, thanks to their deep
learning capability. However, there has to be a certain boundary around what they’re allowed or not allowed to do so that they don’t
pose any type of risk. Standards again play a key role here.

Nitant Desai, Chief Operations and Technology Officer, PNB Housing Finance has a different point of view around AI standards. He
says that “the issue is more around getting the right outcome from an AI based system, and not really industry standards”. “It’s too
early to talk about missing standards”, he added.

Identifying the Right AI Vendor is the Second Biggest Issue

While there are plenty of vendors offering AI based solutions, IT decision makers of this community find it difficult in choosing the
right one. Identifying relevant vendors can become an issue because of this, because lack of common standards creates silos of avail-
able options. That’s why, identifying vendors emerged as the second biggest hindrance in AI deployment.

Skillsets, the Third Biggest Hindrance

Being a new technology, there’s still a dearth of available skillsets in the market, with a majority of respondents facing an issue in

Artificial Intelligence for BFSI Segment 30

 Issues Hindering AI Deployment

finding the right set of people who have experience with cognitive systems. Interestingly, cost of investment and convincing top man-
agement to deploy AI seem to be of lesser importance. This appears to indicate that AI has gained acceptance as a technology that
can add immense business value. The wait is now for the right set of standards and a talent pool to deploy it properly.

A Note on Industry Standards

As AI systems gradually become a part of our routine lives, there’s a concern about their ethical functioning. Being autonomous sys-
tems, it’s important that their developers embed the highest standards of ethical practices so that they don’t cause any kind of harm
to humans, knowingly or unknowingly.

Take the automobile sector for instance. If an autonomous vehicle is about to run over a man running across the street, then what
should it do? Should it save that man by swerving and risk banging into a nearby tree or pole? Or should it crush the man and save
the passenger sitting inside? The decision it takes depends upon how it has been programmed by its developers.

A similar question of ethics arises for AI based software that handle back-end operations. In the BFSI segment, if algorithms for
insurance underwriting, personal financial advice, loans, etc. end up favoring a certain profile of people unknowingly, then its
trustworthiness is questionable. This is very unlike traditional ICT systems, which are designed keeping standard metrics in mind like
speed, functionality, RoI, etc. The design of AI based systems on the other hand, have to go beyond business and technology metrics
because of their autonomous and self-learning nature.

While there’s the BS 8611 standard to identify potential ethical harm arising from growing use of robots and autonomous systems,
there aren’t any standards for the back-end automation systems that are going to be used by the BFSI segment.
The good news is that work is underway to develop them.

IEEE’s Global Initiative on Ethics of Autonomous and Intelligent

Systems
IEEE has been working on a work called Ethically Aligned Design, which aims to prioritize ethical considerations while designing
autonomous and intelligent technologies. It also aims to create standards and associated certification programs.
The global initiative is a program to raises issues and provides recommendations that could be used by technologists while creating
AI systems. Several hundred renowned people from six continents participated in creating the Ethically Aligned Design document,
now in its second version. IEEE also has another goal of using the Ethically Aligned Design’s recommendations to frame their P7000
Suite of standards.

The standards suite has various working groups to address different aspects of autonomous and intelligent systems, right from
addressing ethical concerns during system design to transparency of autonomous systems, data privacy process, algorithmic bias
considerations, etc.

The Ethically Aligned Designs program says that issues of transparency, accountability, and algorithmic bias must be addressed
during design and implementation of an AI system. It should basically follow five Principles--human rights, prioritizing well-being, ac-
countability, transparency, and mis-use as well as awareness about its mis-use. This is so that AI based systems don’t infringe on rec-
ognized human rights like freedom, dignity, cultural diversity, etc. Instead, they should prioritize metrics of well-being in their design
and use. The designers must also take responsibility and be accountable for their systems. Moreover, the systems must be designed
in a transparent manner so that even if an AI system ends up causing unintentional harm, then it should be traceable. The designers
must also minimize the risks of their mis-use. This is so that there’s some accountability on the designers of an AI system.

Unstated Standards
Considering their deep learning capabilities, AI based system could cause potential harm, though unintentionally. That’s why it’s
extremely important to hire experts who have the knowledge and experience in algorithms. Otherwise, it could lead to algorithm
bias, leading to wrong decisions being taken. As there are no set standards to prevent this from happening, it would increase the risk
profile of various processes. This is possibly the dilemma most BFSI companies are facing about putting AI systems to some serious
production use. Their use may not cause life and death situations like in unethical use of robots, but they can cause a lot of other
harm like financial losses, losing credibility and trust among customers, to name a few.
It’s therefore important for any company that’s implementing AI based systems to understand the ethical guidelines behind design-
ing and implementing AI based systems, so that it can draw up the right kinds of processes to prevent them from causing any kind of
harm.

Artificial Intelligence for BFSI Segment 31

Section 5:

Deployment
Challenges and
Best Practices

Artificial Intelligence for BFSI Segment 32

 Deployment Challenges and Best Practices

The Challenges
It’s worth mentioning the issues that companies who’ve ventured into AI are facing while deploying it. The top most issues are
related to handling the huge amount of data by the AI system. Over 54% of the respondents felt that training their deep learning
system to work efficiently with lesser data was the biggest challenge, followed by structuring it as per the requirements. Manag-
ing security and privacy concerns were next, followed by getting the AI system to interpret questions correctly and provide the
right answers. Finding similar use cases of AI is the next in line. In fact, the industry mentioned this as a key challenge in other
places in the survey as well. Keeping this in mind, we’ve dedicated a complete section in this report on some global as well as
Indian deployments of AI. The last challenge is that people still prefer to interact with humans. As AI deployments mature in
organizations, this would also change.

Best Practices
AI and cognitive technologies are becoming increasingly important for the BFSI segment. Each AI application will have its own
nuances, but there are certain common best practices to follow.

Identify a BU and Build a Strong Business Case for it: There are different solutions for different business functions, so organiza-
tions need to first determine which BUs need it the most. Next, they need to identify and prioritize what data is important for this
BU and how’s it being generated.

Prepare a Data Management Strategy: Data often resides in silos, so cleaning, extracting and centralizing it is imperative for the
success of an AI system. Not only that, but data must also be put in a structure that can be understood by the AI system.

Provision Enough Self-Learning Time: An AI system will not start delivering results immediately after being deployed. It needs
to be fed historical data and given enough time to learn. A chatbot for instance won’t be able to interpret conversations in the
beginning and would require a human helpdesk to support its learning. This would be time-consuming and depend upon the
volume of data.

Do Continuous Testing of AI Models Even After Roll-out: Just because an AI system delivers the right outcomes from historical
data doesn’t mean it will do so after going into production. It can’t be left to function on its own there because it may not behave
the same way with actual data. Continuous testing on your models is therefore essential to ensure accuracy of results.

Ensure AI Outcomes are Mathematically Correct, Practical and Useful: Just because an algorithm provides mathematically

Artificial Intelligence for BFSI Segment 33

 Deployment Challenges and Best Practices

AI Deployment Advice
What to Expect What to Ensure

More Accuracy and Predictability: AI’s primary goal Data Digitization: BFSI companies have a lot of paper-
is to increase accuracy and predictability of outcomes based records to be digitized. Before AI can be applied to
and reduce human errors in business processes, par- them, they must first be digitized and made searchable.
ticularly manpower intensive ones.
Ensure Centralized and Clean Data: AI systems are as
Availability and Scalability: AI based solutions can good as the data used to train them. Companies need to
work 24x7, 365 days a week, without taking leave or centralize their data from different silos and clean it before
getting tired. They’re scalable because they use deep applying AI on it.
learning technology to continuously improve them-
selves through self-learning. Validate Algorithm Outcomes: Algorithms used to
process data must be clean in order to deliver the right
Digital Fraud Detection and Prevention: With UPI outcome. Since they’re very complex, understanding their
transactions growing by 39% month-on-month and inner working can be a challenge. That’s why their out-
digital payments likely to grow to about USD 1 Trillion come must be continuously monitored and validates.
by 2023 (Credit Suisse report), it will lead to an equally
big spike in digital frauds. Here, a small error or lapse Remove Algorithm Bias: Outcomes from algorithms
could lead to huge losses. AI technologies can monitor should not be in favor of a particular outcome, which is
huge volumes of digital transactions to identify and why it’s essential to have the right data and do continuous
prevent digital frauds. monitoring of output.

Understand Customer Behavior: As customers go on- The Right API Standards for Data Sharing: In a digital
line, AI will be required to analyze customer behavior world, BFSI companies must partner with fintech and
and improve their experience. insurtech companies. This would involve data sharing, so
it’s critical to ensure that the right API standards are used
Maintain Regulatory Compliance: even small mis- to avoid data from getting compromised.
takes or negligence in security measures could lead to
significant financial losses. AI can help companies stay
on the right side of the law by analyzing data to detect
any regulatory deviations.

correct results doesn’t mean that they’re practical and useful. That’s why they will always need involvement of business manag-
ers to ensure that the outcomes are relevant.

Upgrade to a Software Defined Infrastructure: One thing common to all AI applications is their ability to predict. That’s why
the number and rate of predictions must be optimized to get the desired output. A deep learning system for instance, would have
to parse huge amounts of data, which would require massive amounts of parallel processing and storage capacity. Traditional IT
infrastructure may not be geared up to handle this. You’ll need infrastructure that’s flexible, agile and scalable to efficiently man-
age large and varying data coming from multiple sources. The extent of agility here depends upon the nature of the deployment.
A fraud management system will need All Flash storage arrays to minimize latency of response, whereas an AI system to extract a
particular type of target customer from various database sources for a marketing campaign will do with a hybrid storage.

Use Public and Private Cloud for AI Workloads: When running on-prem, AI workloads can tap the cognitive services of an exist-
ing public cloud services vendor like AWS, Google, or Microsoft via APIs. Or they could be run on a private cloud infrastructure on
a consumption-based IT model. This would be like having a public-cloud like infrastructure in your own data center, where you
only pay for what you use, just as in public cloud. An in-house developed AI infrastructure would be more aligned to your exact
requirements. The other option is to run your AI workloads completely on a public cloud.

Open Source or Commercial AI Frameworks: Numerous open source and commercial AI frameworks are available to use for
running pilots. You could run experiments and PoCs on the community versions of the tools and opt for external vendor support
when you’re ready to run pilots and do the final roll-outs. The challenge with using open source AI tools is that companies would
need to either hire additional skill-sets or train existing teams. If this is not possible, then companies could either work with an AI
based startup, which would already have everything in place, or opt for a commercial software.

Artificial Intelligence for BFSI Segment 34

 Deployment Challenges and Best Practices

Get the Right Skillsets: There are multiple skillsets required in an AI Team, some of which include:

• Data scientists: to develop your models and ensure that the algorithms function as per expectations.

• Infrastructure Management: An AI system requires a dynamic pool of hardware and software resources, which is only pos-
sible with a software defined infrastructure. You’ll need people who understand software defined infrastructure and know
how to use APIs to manage resource pools.

• Trainers: Another critical part of any AI team, trainers will be needed to train the AI system to use the dynamic infrastruc-
ture resource pool.

• DevOps: People with DevOps skills are needed to quickly setup a system and dismantle it as needed. For instance, it can
be helpful when you need to expand your current system to a public cloud for additional resources and bring it back when
done.

Artificial Intelligence for BFSI Segment 35

Section 6:

Regulatory Updates

Artificial Intelligence for BFSI Segment 36

 Regulatory Updates

The growing digital eco-system is changing consumer preferences, which as a result is impacting business priorities of BFSI com-
panies globally. Consumers increasingly prefer to access services online, as a result of which BFSI companies across the globe are
closing down their physical branches and reducing the number of ATMs. India’s BFSI sector is also facing some of this heat, with
some Indian banks also closing down their physical branches, and MNC banks reducing ATMs to curb costs.

AI can certainly play a key role here by helping Indian BFSI players tap the digital potential. Since it’s a heavily regulated sector,
amenable policies will go a long way toward enabling this. So far, there are hardly any policies or a roadmap in place to promote
the use of AI by the BFSI sector in India. Most developments and initiatives have been taken to support growth of the digital eco-
system. This is also a good move, since it would compel policy makers to speed up their plans to create AI related policies for this
segment.

Govt. Initiatives to Promote Growth of Digital Eco-system

As far as the digital eco-system goes, the govt. has taken many important steps to promote digital transactions, like demonetiza-
tion, Aadhaar-linked accounts, the JAM Trinity to enable financial inclusion, introduction of GST based return filing by digital
invoices, launch of payments banks, to name a few.

RBI even revised its very definition of a bank branch. Now, a bank branch is defined as any fixed service point or delivery unit
giving basic banking services, provide that it’s open 4-hours a day, 5 days a week. Similarly, IRDA is making it easier for policy ag-
gregators to sell their policies online. Flipkart also approached IRDA last year to give an approval to the e-commerce giant to sell
insurance policies online.

#AIforAll: NITI Ayog’s Proposed National Program for AI

The most popular development in AI by the govt. of India is NITI Ayog’s announcement to come out with an AI roadmap. In the
last budget INR 3,073 Cr were allocated to promote startups in AI, Machine Learning, as well as other new technologies like IoT,
blockchain etc. While the paper doesn’t specifically mention the BFSI sector, it does highlight the critical role that the govt.
should play to promote AI and ensure that India has its rightful place in this transformational era.

Called #AIforAll, the program whitepaper aims to lay the ground-work to evolve a national AI strategy. The objective of this strat-
egy should to leverage AI for economic growth, social development, as well as a creating an AI ‘garage’ for emerging and develop-
ing economies. Such a national AI program will enable the govt. to play a catalyst’s role in supporting partnerships, providing
access to infrastructure, and fostering innovation through research and creating a demand by asking for solutions to address
various govt. needs.

To enable all this, the paper proposes a three-pronged approach:

1. Undertake exploratory PoC AI projects

2. Craft a National Strategy
3. Collaborate with experts and stakeholders

Five sectors have been targeted in the #AIforAll program—healthcare, agriculture, education, smart cities and infrastructure,
smart mobility and transportation.

The paper elaborates on five key barriers that must be addressed to enable #AIforAll. These include a lack of expertise in research
and application of AI, absence of enabling data ecosystems, high resource cost and low awareness for adoption of AI, privacy and
security, and absence of a collaborative approach to adopt and apply AI.

To tackle the first barrier of research in AI, the paper suggests a two-tier structure. First is to setup Centre of Research Excellence
(CORE) that would develop a better understanding of existing core research and push the technology frontiers via knowledge
creation. The paper also proposes to setup a consortium of Ethics Council at each CORE to address ethical considerations while
developing AI.

Second is to has International Centers of Transformational AI (ICTAI). These could have the mandate to develop and deploy
application-based research. Private sector collaboration will be a key aspect of ICTAIs.

The research efforts are to be complemented by an umbrella organization that would guide the research teams by analyzing
socio-economic indicators, studying global advancements, and encouraging international collaboration. They’ll develop a su-
pranational agency to channelize research to solve potential problems with AI, and to develop a common computing and other

Artificial Intelligence for BFSI Segment 37

 Regulatory Updates

related infrastructure for AI.

Re-skilling India’s existing workforce is critical to develop future talent in accordance with the changing needs of the job market.
For this, the paper proposes that the govt. works with private sector and educational institutes to offer certifications and pro-
mote job creation in new areas, like data annotation.

The paper also suggests adoption of a marketplace model to address barriers to AI development and deployment. Here again,
it’s suggested to create a common National AI Marketplace for data collection and aggregation, data annotation, and deployable
models.

Data security is a key challenge in AI, and to ensure that data is not mis-used, individuals are not identified, and data selection is
not biased, the paper suggests establishing data protection frameworks and sectorial regulatory frameworks as well as promot-
ing adoption of international standards.

Owing to the unique nature of AI solutions, the paper also proposes to setup an Intellectual Property framework via IP facilita-
tion centers. These can help bridge the gap between practitioners and AI developers, with training of IP granting authorities,
judiciary, and tribunals.

RBI to Setup Data Science Lab

Governments around the world are embracing data science to help them craft the
right policies, gain deeper insights into their country’s financial systems, get more
visibility into NPAs, to name a few. Considering all this and the state of growing
NPAS in India, RBI recently announced its decision to setup a Data Sciences lab
by the end of Dec this year. The lab will help the central bank harness big data to
aid its various key responsibilities like inflation management, currency, debt and
reserves management, etc.

Artificial Intelligence for BFSI Segment 38

 Regulatory Updates

NASSCOM’s AI Initiatives
NASSCOM has a number of initiatives to promote AI across India. These include:

1. FutureSkills Platform: This initiative is being built as an industry utility to accelerate India on the journey to build sustain-
able talent and collaborate with global partners to become the worldwide hub and partner of choice for talent in the field
of emerging technologies. Skilling for AI and Data Sciences has bene identified as the key technology and is going live first
on this platform. The platform creates both learning and deep skilling pathways that will enable employees to upgrade their
current skills as well as keep abreast with all the latest trends and shifts in these segments.

2. CoE- AI and Data Sciences: NASSCOM has partnered with Govt of Karnataka and Telangana to set up a Centre of Excel-
lence on AI and Data Sciences that aims to build a co-creation platform with government, industry, startups and academia
to catalyse AI solutions, evangelise AI adoption, leverage open datasets for startup innovation and create POCs the theme of
# AIforIndia.

3. Evangelisation: NASSCOM in collaboration with its members has been working on creating awareness of AI technologies
amongst the technology professionals in the country. Through focused workshops, webinars and events, more than 10,000
professionals have engaged with NASSCOM on the different technology aspects of AI and how they can adopt for their work.

4. AI Startups: Through the 10,000 Startups initiative and the Product Council, NASSCOM runs a DeepTech Startup Mentoring
Program, wherein pre-curated startups are supported through an extensive mentorship program on technology, organiza-
tional building, go to market and connect with enterprise for business and funding agencies. The first batch of 20 startups
has a strong

5. Showcase and Awards: NASSCOM has announced an annual AI Game changer Award to showcase the best application
being built on AI. More than 300+ applications were received that would be shortlisted through a jury process and then
published as cae studies.

6. Research Reports: During the year, NASSCOM will publish a series of research reports that track the business opportunity
in AI, use cases of AI

Artificial Intelligence for BFSI Segment 39

Section 7:

Recommendations
on the Way
Forward

Artificial Intelligence for BFSI Segment 40

 Recommendations on the Way Forward

There’s a Need for Standards to Differentiate Between AI Offerings: While just about every BFSI company is planning to em-
bark on their AI journey, they must tread with caution. Being a relatively new technology, there no common standards to evaluate
solutions from different vendors, nor is there a common check-list on how organizations can move forward on their AI journey.
Different stakeholders like vendors, third party advisories, consultants, etc. should get together to create a framework, evalua-
tion criteria, and CSFs (Critical Success Factors) to help choose between various AI offerings.

An Agile, Scalable, and Flexible IT Infrastructure is Essential for AI: Since AI systems have to process massive amounts of
data in real time, traditional IT infrastructure may not be able to handle it and might become a bottleneck. So putting a software
defined infrastructure for AI is highly recommended.

Open Source Tools and Cloud Services are a Good Way to begin Pilots: Many CIOs and IT decision makers we spoke to were
not comfortable with the idea of using open source tools to build AI systems for pilots. Open Source AI tools are the easiest way to
become comfortable with the technology, and its usage should be encouraged.

Continuous Testing a Must in AI Systems: AI pilots can’t be considered to be successful if they’ve only been tested successfully
with historical data. Due to their complex nature, the algorithms used in AI systems can deliver different results when applied to
real data. It’s therefore essential for companies to ensure that AI pilots are tested thoroughly with real data before moving them
to production. Even after that, they need to be continuously tested for accuracy of outcomes and to eliminate algorithm bias.

AI Team Should Have a Balanced Skillset: AI systems have to deliver outcomes that are both practical and mathematically cor-
rect. Otherwise it would be a waste of time, effort, and cost. This is only possible if business professionals are also a part of the
team to ensure that the outcomes have business value. Data Scientists must ensure that algorithms behave correctly and are not
biased in any way.

Artificial Intelligence for BFSI Segment 41

Section 8:

Annexures:
I. Open Source and
Commercial Tool-kits
II. Glossary of Common
AI Terms

Artificial Intelligence for BFSI Segment 42

 Annexure I-Open Source and Commercial Tool-kits

Annexure I
Open Source and Commercial Tool-kits
Open Source Toolkits
Name Description Common Uses
TensorFlow A machine learning framework by Google, popular for Voice/sound/Image/Video recognition;
(https://www.tensorflow.org/) cloud-based applications; It’s also supported by AWS, Sentiment Analysis; Threat/Fraud De-
Azure, and Google Cloud platform. tection; Language Detection like Google
Translate; Time series data analysis for
recommendation.

Caffe A Deep learning framework for research experiments

(http://caffe.berkeleyvision.org/) and industry deployments. Has high speed, which en-
ables it to process over 60M images per day on a single
Nvidia K40 GPU.

Apache Mahout An Apache machine learning framework with program-

(https://mahout.apache.org/) ming environment and framework for building scalable
algorithms. It has a wide variety of premade algos for
Scala+Apache Spark, H2O, Apache Flink and Samsara. It
has a vector maths experimentation environment with
R-Like syntax.

Accord.Net Frame- A .NET Machine Learning signal processing framework, Can be used for production grade com-
mostly written in C#, that contains libraries to process puter vision, computer audition, signal
work video signals and image streams. There’s a comprehen- processing, and stastics applications.
(http://accord-framework.net/) sive set of sample applications on its website to get you
quickly started. Has algorithms for face detection, to
stitch images together or track moving objects. Works
on Visual Studio.NET, and on Linux through NuGet and
VirtualBox.

NuPIC Short for Numenta’s Platform for Intelligent Comput- HTM for Stocks to track anamolies in
(https://github.com/numenta) ing, it’s a machine learning framework that’s based on stock market; Rogue Behavior Detection
Heirarchical Temporary Memory, which analyzes data to find anamolies in human behavior
streams in real time. It helps find anamolies in a time and Geospatial tracking to find anamo-
series of your own data and compare them to other lies in objects moving through space.
anamoly detection techniques.

Oryx 2 A framework that specializes in real-time, large scale Real-time performance prediction, large
(http://oryx.io/) machine learning. It’s used to build custom applications scale machine learning.
and also has packaged applications for collaborative
filtering, regression, clustering, and classification.

H2O.ai An open source machine learning platform for enterpris- Claims processing, credit scoring,
(https://www.h2o.ai) es that makes ML accessible and allows business users customer churn, fraud detection, ICU
to extract insights from data without needing expertise monitoring, operational intelligence
in deploying or tuning machine learning models. Its and predictive maintenance.
users include 222 of Fortune 500; 8 of top 10 banks; 7
of top 10 insurance; 4 of top 10 healthcare companies;
Capital One, Comcast, Kaiser Permanente, Macy’s, Neu-
star, Nielsen, Progressive Insurance, PwC, Walgreens
and Zurich North America are some users.

PredictionIO A ML server, part of an Apache project that leverages big Used by data scientists to quickly create
(https://predictionio.apache.org) data tools like Hadoop, HBase, and Spark. It’s used by prediction engines; build and deploy
oganizations to implement lambda architecture. an engine as a web service; respond to
dynamic queries in real-time.

Artificial Intelligence for BFSI Segment 43

 Annexure I-Open Source and Commercial Tool-kits

Apache A machine learning platform for Big Data analysis, Big Data analysis
(http://systemml.apache.org) particularly for high-order mathematical calculations.
It will go through your code line by line and determine
whether it should be run on the system or an Apache
Spark Cluster, thereby saving the time and trouble so
that data scientists can focus on their own work. It sup-
ports R language and Python syntax. Runs on Apache
Spark, and also supports Apache Hadoop, Jupyter,
Apache Zeppelin, and other platforms. It’s being used
by automotives, airport traffic, and social banking
industries.

Scikit-Learn Based on NumPy, Scipy, and Matplotlibe open source

(http://scikit-learn.org/stable) projects, this is a Python based ML tool with algorithms
for classification, regression, clustering, dimensionality
reduction, model selection, and pre-processing. Spotify,
Evernote, OKCupid, and change.org are some of its
users

Turi Create A tool that was released by Apple in Dec 2017 that helps Recommender systems, image clas-
(https://github.com/apple/turi- simplify the development of custom machine learning sification, object detection, activity
create) models. It’s meant for non-experts so that they can do classification, text classification.
it without having to write much code. The software is
very easy to use and focuses on the task rather than the
algorithms. It has built-in streaming visualizations to
explore data, and supports text, audio, images, video,
and sensor data. It has algorithms for classification, re-
gression, graph analytics, clustering, nearest neighbors,
topic models and more. The tool-kit runs on x86, 64-bit
architecture; MacOS 10.12+, Linux, or Windows 10.

Microsoft Cognitive A commercial grade tool-kit that trains deep learning

algorithms to learn like the human brain. It supports
Toolkit C++, Python, and BrainScript. It integrates with Azure,
(https://www.microsoft.com/en- evaluates and trains deep learning algorithms, and can
us/cognitive-toolkit) scale in a range of environments from CPUs to GPUs to
multiple machines. It’s being used in Skype, Cortana,
Bing, Xbox use it to develp commercial grade AI.

Distributed Machine A toolkit designed to use in big data applications for

running and training AI systems faster. It comprises of
Learning Toolkit DMTK framework, LightLDA topic model algorithm,
(http://www.dmtk.io/index.html) LightGBM, and Distributed (Multi-Sense) Word Embed-
ding algorithm.

Shogun An ML toolbox that supports many languages like Rapid prototyping of data pipelines
(http://www.shogun-toolbox.org) Python, Octave, R, Java/Scala, etc. It supports multiple
platforms (Linux/Unix, MacOS, Windows), and works
in cloud from a browser and contains multiple data
representations, algorithm classes, and general purpose
tools.

DSSTNE Created by Amazon, DSSTNE or “Deep Scalable Sparse Recommendation Models; Neutral
(https://github.com/amzn/ama- Tensor Network Engine” is an AI tool to create your own Network Modeling
zon-dsstne) recommendation engine. Amazon has used it to gener-
ate recommendations for their products. It’s useful for
retailers and online organizations.

Artificial Intelligence for BFSI Segment 44

 Annexure I-Open Source and Commercial Tool-kits

Commercial AI Frameworks, Tools, and more

Name Description
AWS SageMaker and Amazon Web Services’ SageMaker is a fully managed machine learning service for build-
ing, training and deploying models. It supports any AI framework or algorithm, and comes
DeepLens pre-configured for TensorFlow and Apache MXNet frameworks. It also includes ten popular
(https://aws.amazon.com/sagemaker/) machine learning algorithms by default.

AWS DeepLens AWS DeepLens is perhaps the world’s first deep learning-enabled video camera for develop-
ers. It includes tutorials and projects to gain hands-on experience with computer vision.
(https://aws.amazon.com/deeplens/)

Azure Machine Learn- Microsoft Azure’s cloud-based artificial intelligence tool, the Azure Machine Learning Studio
features an intuitive drag-and-drop interface that makes it easy to get started with predictive
ing Studio analytics, even if you don’t know how to write code. For more advanced users, it supports
(https://studio.azureml.net/) Python and R. The studio’s webpage provides ample links to tutorials and videos to help you
learn more about machine learning.

Google’s Cloud AI Ser- Google’s Cloud AI provides a plethora of machine learning services with pre-trained models
as well as a service to allow you to create your own ML models tailored to your specific
vices requirements. Some of these services include language translation, natual language process-
(https://cloud.google.com/ml-engine/) ing, text to speech recognition and vice versa, language translation, image and video analysis,
a development suite to build conversational interfaces, and a job discovery service.

Kaggle A Google-owned website that offers learning resources and discussion boards on AI. Its real
draw are its competitions where organizations can post data science and machine learning
(https://www.kaggle.com/) challenges, and teams can compete to come up with best solutions. Some of these competi-
tions also offer handsome cash prizes.

Rocco An AI powered tool to help organizations in their social media marketing efforts. This can be a
great tool for any company, including the BFSI segment to engage with customers or prospec-
(http://www.rocco.ai/) tive customers on social media. The tool uses AI to learn more about your business and based
on that suggest what content are your followers likely to engage with. This reduces time
and effort in creating and re-creating content strategies and social media posts. It will then
provide insights on which of your social media campaigns are working and provide details on
how your different pieces of content are performing.

Vidora “A real-time, machine learning platform that helps optimize marketing and product automa-
tion. It provides intelligent and proactive AI-based churn prediction to subscription-based
(https://www.vidora.com/)
service companies and help identify customers likely to churn. It can also help identify free
subscription holders that might be ready for an upsell.

For financial services companies, it can predict which customers could be potential candi-
dates to be upsold for various products like credit cards, savings accounts, loans, insurance
packages, investment portfolios, etc. It can also help retain customers by predicting which
customers are likely to churn and sending them personalized promotional offers. It can even
manage risk by predicting customers making late payments, likely defaulters, likely to make
an insurance claim, etc.“

Albert An artificial intelligence platform that can autonomously drive digital marketing platform,
right from building them to optimizing and finally measuring results. This approach requires
(https://albert.ai/)
little human input while AI analyzes, manages, and optimizes paid ad campaigns. The tool
works across multiple channels including paid search, social media, programmatic display,
and emails. BFSI companies struggling with managing and continuously optimizing their B2C
digital marketing campaigns can use this to manage everything autonomously 24x7. Some
of its users include Natori, Gallery Furniture, RedBalloon, Evisu, Harley Davidson, Cosabella,
Dole, Anastasia Dates

Artificial Intelligence for BFSI Segment 45

 Annexure I-Open Source and Commercial Tool-kits

Annexure II
Glossary of Common AI Terms
Artificial Intelligence: The science of creating machines or applications that have cognitive capabilities to mimic human behav-
ior. It deals with creating machines and software with self-learning capabilities that are able to recognize their environment and
solve problems, must like humans.

AI tools, platforms, frameworks, and libraries: These terms tend to cause a lot of confusion among the uninitiated, so they
merit some attention. AI tools are software programs to do specific machine learning tasks. Platforms are software that provide
everything you need to run an AI project, i.e. data analysis, preparation, modeling, etc. Libraries on the other hand perform
discrete functions, like those mentioned in a platform. They can be used to complete parts of a machine learning project. An AI
framework on the other hand could be a tool or a library that helps you build machine learning models without having to deal
with the underlying algorithms. Some well-known AI frameworks include Apache Mahout, Amazon Machine Learning, Caffe2,
H2O, Microsoft Cognitive Toolkit, Scikit-Learn, TensorFlow, and Theano. We’ve given more details about what some of these
frameworks do elsewhere in this report.

Algorithm: A procedure or formula designed to solve a specific, recurring problem. Search engines use algorithms to take key-
words as inputs, search for them in databases and display the most relevant results. Similarly, encryption algorithms are used
to protect sensitive data to prevent its mis-use even if it gets stolen. E-Commerce sites use algorithms to recommend products
matching your search as you browse through products. Similarly, maps use algorithms to calculate the fastest route to a particu-
lar destination. Algorithms are the foundation for any AI setup.

Bayesian Network: A statistical model that uses graphs to depict variables and their dependencies. For instance, a Bayesian
network could represent probable relationships between say a disease and its symptoms. The network could be used to compute
the probability of various diseases. It’s also used to compute different types of risks in banks be it operational or credit risk, etc.

Chatbots: A computer program that can hold a conversation with an individual over text-based chat. There’s a whole variety of
chatbots available. Some will scan your input keywords to respond while others will use more sophisticated natural language
processing algorithms to infer a meaning and react accordingly. For instance, one chatbot could scan your keywords ‘bank bal-
ance’ and tell you how much it is, whereas another could check the balance and let you know if it’s too low or give you some
investment options if it’s high.

Data Science: A field that uses processes, algorithms, and systems to derive insights and knowledge from both structured and
unstructured data in various forms. It’s a growing field and in order to truly leverage benefits from this field, data scientists must
possess multiple skills, like analytics, machine learning, statistical and data mining.

Deep Learning: It’s a subset of the broader field of machine learning and involves learning data representations through multiple
layers. Each layer processes data to a certain extent for the next layer, until the final output. The depth of a deep learning system
depends upon the number of layers that process the data. Deep learning can be used for a variety of applications like speech and
image recognition, visual art processing, natural language processing, recommendation systems, CRM, etc.

Machine Learning: This is a name given to the science of enabling computers to self-learn with the help of statistical techniques.
It involves creating algorithms that can work on data to provide an outcome for a specific task. These algorithms are designed
such that they can self-learn from the outcome to continuously improve their performance.
Neural Networks: This is a set of hardware and software configured to function like neurons in a human brain. They use a variety
of deep learning technologies to solve complex problems, like speech-to-text recognition, oil exploration, weather prediction,
facial recognition, etc.

NLP (Natural Language Processing): The field of computer science that aims to recognize human language to enable humans to
have textual or audio-based conversations with a computer. NLP would comprise of various components of a system that allow
machine to human conversation in a natural language. For instance, one typical use of NLP is chatbots.

NLU (Natural Language Understanding): A subset of NLP, natural language understanding deals with understanding the human
language input. The NLU part of the system will take human language input and convert it into a format that the computer can
understand.

NLG (Natural Language Generation): The other subset of NLP that deals with computers generating language from the input

Artificial Intelligence for BFSI Segment 46

 Annexure I-Open Source and Commercial Tool-kits

they’ve received. So, if NLP is the overall system, NLU aims to decipher human language into a format that the computer under-
stands, and NLG aims to process a response that humans could understand.

Robotic Process Automation: the use of computer programs with artificial intelligence to do high-volume repeatable tasks that
were otherwise done by humans. What’s more, RPA software would do it more quickly, accurately and tirelessly. RPA have been
known to reduce the time taken to do a particular job from days to minutes.

Robo-advisor: A class of software that can provide investment, financial, or insurance advice to customers based on inputs they
receive. They do this with moderate to minimum human intervention, making them a good option to put on self-service portals.

Rule based system: A system used in AI that uses a set of if-then statements to analyze data and provide answers to problems.
Such systems derive their answers from a set of data and rules.

Voicebot: It’s the same as chatbots, except that it engages in a voice-based conversation. If used effectively, voicebots could take
over the old and tedious IVR based menu systems.

Artificial Intelligence for BFSI Segment 47

 References

Annexure III
References
• Active AI: http://active.ai/#why
• Wharton: Customer Engagement: Does Talking to Humans Still Matter?: https://whr.tn/2eg7MBH
• BusinessWorld: Turning AR VR Mainstream-Prashant Mehta- https://bit.ly/2JWj3KC
• Telegraph UK: Can AI boost your customer service offering?- https://bit.ly/2gGWhYv
• HBR: How Artificial Intelligence Will Redefine Management- https://bit.ly/2edtqpu
• Entrepreneur: Natural Language-Generation Startup Narrativa Uses AI To Create Content At Scale-https://bit.ly/2I0Dtgr
• Express Computer: Artificial intelligence will overtake human intelligence in 30 years, predicts SoftBank CEO Masayoshi Son- https://
bit.ly/2ljs2ro
• Bloomberg: The Humans Hiding Behind the Chatbots- https://bloom.bg/2JW8abT
• IBM: Is Your Chatbot Ready for Prime-Time?- https://ibm.co/2iT3DGR
• Sabre: Chatbots in travel: 4 things the industry must get right for success- https://bit.ly/2liLNzs
• VentureBeat: The 200 billion dollar chatbot disruption- https://bit.ly/1W1Fgq5
• Information Age: A CIO’s guide to chatbots: Everything you need to know- https://bit.ly/2n513CW
• CIO: 7 things marketing pros need to know about chatbots-https://bit.ly/2e8vZP8
• Botlist: An App Store For Bots- https://botlist.co/
• ET CIO: The rise of chatbots and Indian startups, IT News- https://bit.ly/2JYF8Z4
• Forbes: Could Chat Bots Replace Human Jobs?- https://bit.ly/2JPAmxX
• Softwebsolutions: 5 Jobs where bots will replace humans, Build a Chatbots- https://bit.ly/2JVi4Hj
• VentureBeat: How AI Startups and Tech Giants Collaborate to do good deeds-https://bit.ly/2nbcwj9
• VentureBeat: Chatbots will not replace 5 million jobs-https://bit.ly/2aUBUCz
• Vice: Why The First Wave of Chat Bots Are a Bit of a Mess- https://bit.ly/2JPgLgY
• PCMag: How to Find and Use a Facebook Chatbot- https://bit.ly/2leSAdl
• Techworld UK: 12 best uses of chatbots in business in the UK- https://bit.ly/2mdiJII
• AnalyticsVidhya: 5 AI applications in Banking to look out for in next 5 years- https://bit.ly/2pNhgwP
• Quant: The Growth And Future Of Algorithmic Trading- https://bit.ly/2EU775T
• Hot Hardware: JPMorgan Chase Uses COIN Machine Learning Program To Eliminate 360K Lawyer Hours A Year- https://bit.
ly/2qGbRod
• Bloomberg: JPMorgan Software Does in Seconds What Took Lawyers 360,000 Hours-https://bloom.bg/2mosyqu
• Global Data: Bank of America’s new chatbot is a higher form of intelligence-https://bit.ly/2LXXM0q
• Global Data: Bank of America, Wells Fargo, PNC add more digital banking services- https://bit.ly/2LXXM0q
• Tear Sheet: Inside the development of Erica, Bank of America’s AI-powered bot- https://bit.ly/2hgf2np
• BusinessWire: SmartBiz Loans® Launches AI-powered Digital Advisor Tool for Small Business Owners- https://bit.ly/2lhOdhq
• eConsultancy: Five things we learned from launching a Facebook Messenger chatbot- https://bit.ly/2LZp4U9
• Expanded Ramblings: 57 Interesting Facebook Messenger Statistics and Facts (December 2017)- https://bit.ly/2JVjhOR
• TechCrunch: Amazon’s Alexa passes 15,000 skills, up from 10,000 in February- https://tcrn.ch/2uDVUBi
• VentureBeat: Facebook Messenger hits 100,000 bots- https://bit.ly/2o0pOB0
• ETCIO: icici bank: ICICI Bank’s AI chatbot iPal empowers customers with information and financial services- https://bit.ly/2t8UyjG
• Forbes: How Artificial Intelligence Will Impact The Insurance Industry- https://bit.ly/2ti0mGX
• Mainichi: Insurance firm to replace human workers with AI system- https://bit.ly/2JPBEJj
• Metlife: Welcome to the Future: How AI is Transforming Insurance- https://bit.ly/2lf3zTZ
• MindTree: Everything You Need to Know About Artificial Intelligence in Insurance Industry- https://bit.ly/2CtNYep
• Samsung: Insurance Fraud Prevention Gets Help From Artificial Intelligence- https://bit.ly/2JJoqg6
• Logiq: The Impact of Artificial Intelligence on the Insurance Value Chain- https://bit.ly/2JUInNI
• Map a Research: 5 use cases for AI in banking (beyond those helpful chatbots)- https://bit.ly/2tgOpRP
• The Financial Brand: How Financial Institutions Are Turning AI Into ROI- https://bit.ly/2xrVTVt
• ICT Career: ANZ and UTS seek AI underwriting- https://bit.ly/2tlUzAm
• Deal Street Asia: DBS completes acquisition of ANZ’s wealth, retail units in 5 Asian markets- https://bit.ly/2MCpnFt
• ComputerWorld: ANZ Wealth exploring AI for insurance underwriting - https://bit.ly/2LVIvNy
• RGare: Artificial Intelligence and Underwriting- https://bit.ly/2JVQ06A
• Business News Daily: How AI is Radically Changing the Insurance Industry- https://bit.ly/2CkjAy3
• Search Inc: How AI is changing underwriting within the insurance industry- https://bit.ly/2MDuZiK
• Fintech: Deutsche Bank Deploys Artificial Intelligence to Help Meet Demands of Regulatory Compliance- http://bit.ly/2lgIDvN
• WSJ: Deutsche Bank Deploys Artificial Intelligence to Help Meet Demands of Regulatory Compliance- https://on.wsj.com/2HY1Yem
• Dow Jones: Deutsche Bank Enlists AI to Assist in Compliance Goals-http://bit.ly/2M2PJ2G
• Forbes: Citi Ventures Deploys Machine Learning And Artificial Intelligence With People- http://bit.ly/2yl2yDd

Artificial Intelligence for BFSI Segment 48

 References

• Emerald Insight: Stress test of banks in India across ownerships: a VAR approach- http://bit.ly/2tj6XRg
• RBI: Reserve Bank of India – Database- http://bit.ly/2K2n5RN
• FinExtra: How can banks take the pressure out of stress tests?-http://bit.ly/2t3RRAa
• Eli Next: Fraud Management: Detection and Prevention in Banking Industry- http://bit.ly/2tgQMnD
• American Banker: How PayPal Is Taking a Chance on AI to Fight Fraud- http://bit.ly/2M1NdJQ
• PR News Wire: Danske Bank and Teradata Implement Artificial Intelligence (AI) Engine that Monitors Fraud in Real-Time- https://prn.
to/2ynHNXE
• Chatbots Magazine: AI-Powered Chatbots to Deliver Better Banking Experiences- http://bit.ly/2JKOJDm
• Cognizant: Provider Benchmarking for Insurance Fraud Detection-https://cogniz.at/2JZK9kc
• Entrepreneur: Regtech in India: The need of the hour?- http://bit.ly/2MB9b7K
• Infosys: FINspeak: Artificial Intelligence and Regtech: The changing face of compliance in banking- https://infy.com/2lekTbE
• BNP Paribas: Emotional AI coming soon?- http://bit.ly/2JNXbC0
• Affectiva: Emotion AI Overview- http://bit.ly/2HY3QUq
• Gartner: Emotion AI Will Personalize Interactions- https://gtnr.it/2ylbMPW
• Machine Learnings: The Rise of Emotionally Intelligent AI – Machine Learnings- http://bit.ly/2t4diRC
• BSCI 21: 3 Ways Behavior Analysis & Artificial Intelligence Will Change The World- http://bit.ly/2JY13fA
• Comply Advantage: AML Screening & Monitoring- http://bit.ly/2JKZTId
• Deutsche Bank: Deutsche Bank rolls out new AI-based equities algorithmic platform in APAC-http://bit.ly/2JL07Pz
• Economic Times: artificial intelligence: How artificial intelligence is reshaping recruitment, and what it means for the future of jobs-
http://bit.ly/2LY0IKu
• Big Data Made Simple: AI for recruiting: 5 ways bots can help you with hiring- http://bit.ly/2t6gfAW
• HR Katha: How AI and robots are transforming banking jobs in India- http://bit.ly/2HVHVgB
• Top Tal: Beyond Chatbots: Artificial Intelligence in Finance and Banking- http://bit.ly/2JLcOu2
• Content Marketing Institute: 8 Ways Intelligent Marketers Use Artificial Intelligence- http://bit.ly/2JKa7sB
• The Financial Brand: The Five Fundamental Building Blocks for Artificial Intelligence in Banking- http://bit.ly/2JTkCFL
• Phrasee: The Ultimate Glossary of Artificial Intelligence Terms- http://bit.ly/2ylc6hx
• CISDAI: Recommender System for a Bank- http://bit.ly/2JKzgDr
• Wikipedia: Recommender system-https://en.wikipedia.org/wiki/Recommender_system
• Statsbot: Recommendation System Algorithms- http://bit.ly/2MA8207
• Wikipedia: Knowledge representation and reasoning- http://bit.ly/2t5fNCT
• TechTarget: AI components make tools more than the sum of their parts- http://bit.ly/2tjjlkx
• TechCrunch: The Evolving Landscape Of Recommendation System- https://tcrn.ch/2HYlI1t
• TCS: Drive Governance - Ensuring Customer Centricity in Banking: Employing Cognitive Computing- https://on.tcs.com/2MChlwy
• Consultancy UK: How Artificial Intelligence is transforming the banking industry- http://bit.ly/2LYrfaG
• ComputerWorld UK: Barclays on the four biggest risks AI poses to banking- http://bit.ly/2K2v3KK
• Gartner: Jobs: By 2020, Artificial Intelligence will create more jobs than it eliminates- http://bit.ly/2HZM7fg
• Elite Data Science: Open Source vs Commercial Machine Learning Software- http://bit.ly/2t7fpns
• ELETS Online: Artificial Intelligence — A Cutting-Edge Technology Revolutionising Banking Sector- http://bit.ly/2t6ScBP
• BSI Group: BS 8611:2016 Robots and robotic devices. Guide to the ethical design and application of robots and robotic systems-
http://bit.ly/2yh7tVM
• Analytics India Mag: RBI To Set Up Data Sciences Lab By December 2018- http://bit.ly/2lfb2lZ

Artificial Intelligence for BFSI Segment 49

Copyright © 2018 — CyberMedia Research | CMR. All Rights Reserved

Artificial Intelligence for BFSI Segment 50