Microsoft Word - 20200410 .docx
China Intelligent Finance Development Report 2019
Editor-in-chief: Xiao Gang
Editors: Shao Yu, Shi Jinjian, Luo Rongya and Zhang Jiajia
Authors (in alphabetical order):
Zhiyi, Fan Lixin, Fan Minmin, Guo Weimin, He
Yafeng, Huang Binghua, Jiang Bo, Li Hongyu, Li
Jinlong, Li Ming, Li Xiaolin, Li Xiuquan, Liu Bo, Liu
Gang, Liu Haitao, Liu Shuoling, Liu Tieyan, Liu
Tingshan, Liu Weiqing, Lu Songhua, Meng Dan,
Meng Kaixiang, Qiu Han, Qu Bensheng, Shi Rong,
Tan Zetao, Tian Hui, Wang An, Wang Min, Wang
Ruyi, Wang Siyao, Wang Tiandu, Wu Haishan, Xie
Jun, Yang Qiang, Yang Tao, Yao Jiangtao, Yin
Youping, Yu Quanjie, Yuan Weibin, Yue Jianping,
Zhang Junfang, Zhang Weina, Zhou Youchi, Zhu
Qing, Zhu Taihui and Zhu Xiuye
Note: The group members participated in the research in their
personal
capacity, hence the views expressed in the report do not represent
the
official views of the organizations they are affiliated with.
ii. Definition and Significance of Intelligent Finance
iii. Technology Challenges Facing Intelligent Finance
iv. Intelligent Finance Development Trends
II. Applications
ii. AI in Middle-Office Business Scenarios of Financial
Institutions
III. Topics in Focus
Finance
iv. Financial Data and Financial Clouds
IV. Regulation
ii. Regulatory Challenges Facing Intelligent Finance and
iv. Inclusive Finance and Consumer Protection
V. Foreign Market Overview
ii. Foreign Intelligent Financial Application
iii. Deficiencies in the Development of Intelligent Finance
Abroad
iv. Suggestions for China
iii. Major Intelligent Finance Events in 2018 and 2019
Chinese President Xi Jinping has emphasized that “artificial
intelligence (AI) is a key driving force in the new round of
technological revolution and industrial transformation.
Accelerating
the development of new-generation artificial intelligence is a
strategic
issue, crucial for China to seize the opportunities in the new
round of
technological and industrial revolution”. A new round of
technological revolution and industrial transformation involving
AI,
big data, quantum information and bio-tech are gathering
strength,
and creating a large number of new industries, new business
formats,
and new models that bring rapid and earthshaking changes to
the
world’s development and the production activities and lives
of
humans.
The State Council, China’s Cabinet, issued the Plan for
Developing
New Generation Artificial Intelligence in 2017, which called
for
developing intelligent finance, building big data systems for
the
financial sector and enhancing finance-related multimedia
data
processing and comprehension capabilities. The plan also
stressed
introducing innovative intelligent financial products and
services,
developing new formats of financial business, encouraging the
financial industry to use technologies and devices such as
intelligent
customer service and intelligent monitoring, and developing
Technological advances and innovations always go side by side
with
financial development. They have historically led the changes in
the
financial industry, such as the automated teller machine (ATM) in
the
1960s, electronic payment in the 1980s, online payment and
mobile
banking in the 1990s, Internet finance since 2000, and fintech
since
2008. The integration of finance and technology creates new
business
models, applications, processes and products, gives birth to new
types
of customer relationships and partnerships, and thus has a
profound
influence on financial institutions, financial markets and
financial
services. The development of fintech has roughly gone through
three
stages: electronic, digital and intelligent. In the electronic
era,
financial institutions adopted information technology to
achieve
electronic business and automation; in the digital era,
financial
institutions innovated financial products and processes, and
transformed service approaches; in the intelligent era,
financial
institutions use machine to simulate the physical labor and
in
particular the mental labor of humans through the application
of
artificial intelligence technology, and to perform decision-making
and
control for financial services. It should be pointed out that
while
algorithms to copy human logic and reasoning, and replaces
human
brain with machine to quickly process massive data, thus
powerfully
extending human brainpower.
Intelligent finance is a new business format based on deep
integration
of artificial intelligence technology with the financial industry,
and it
transforms the financial pattern by using machine to replace
and
outperform part of humans’ management experiences and
capabilities.
As an advanced form of fintech development and an upgrade and
transformation of digitalization, intelligent finance defines the
future
development trend and the core competitiveness of the
financial
industry.
Primarily, there are two reasons why we have separated
intelligent
finance from fintech and prepared an independent development
report
on it.
First, developing artificial intelligence technologies has been
elevated
to an important strategy of the Chinese government, and it will
be
implemented in three steps: step one, by 2020, China’s
overall
artificial intelligence technologies and applications should be on
par
by 2025, China should achieve major breakthroughs in basic
artificial
intelligence theories and reach a world-leading level in some of
the
artificial intelligence technologies and applications, with
artificial
intelligence becoming a primary driver for the country’s
industrial
upgrade and economic transformation; and ultimately, by 2030,
China
becomes a world leader in artificial intelligence theories,
technologies
and applications, and records notable achievements in
intelligent
economy and intelligent society. Countries are nowadays
competing
fiercely with one another in new-generation artificial
intelligence
technologies. According to the Centre for Artificial Intelligence
and
Robotics at the United Nations Interregional Crime and
Justice
Research Institute (UNICRI), 30 member states of the United
Nations
have formulated national strategies for the development of
artificial
intelligence technology. The competition in technology will
eventually evolve to industrial competition using such
technology.
PwC predicts that global market for artificial intelligence could
be
worth up to USD 16 trillion by 2030.
Finance has a natural connection with AI, and is one of the
most
important domains for artificial intelligence technology
application.
Developing intelligent finance will help China capture the
opportunities of artificial intelligence advancement and gain
the
technology high ground. In particular, the unique nature of
the
financial industry will pose new requirements and challenges
to
intelligence technology breakthroughs and upgrades in China,
and
enhance the efficiency of technology application.
Second, comprehensively using a variety of fintech technologies,
e.g.
big data, cloud computing and blockchain, artificial
intelligence
technologies provide unlimited possibilities for the future
development of the financial industry. They present an advanced
and
upgraded form of existing fintech applications, and will usher
in
disruptive changes in the development of the financial
industry.
Focused research on intelligent finance is conducive to tracking
global
development of applications for integrating artificial
intelligence
technologies into the financial industry, strengthening the
adaptability,
competitiveness and inclusiveness of the financial industry,
greatly
improving the ability and efficiency of financial institutions to
identify,
prevent and control risks, promoting China’s structural reform
in
finance on the supply side, enhancing the ability of the financial
sector
to serve the real economy and people’s lives, defending the
bottom
line of no systemic risks, speeding up the modernization of
China’s
financial system, boosting the international competitiveness
of
China’s sector, and galvanizing the shift of China from a
financial
giant to a financial power.
6
and fintech by the fundamental change it brings to the efficiency
of
the financial sector. By replacing or even outperforming the
behavior
and intelligence of humans, intelligent finance can cater to all
types
of financial needs more efficiently, and drive change and
leapfrog
development in China’s financial industry.
The report consists of six sections and one annex. Section I
(Technologies) introduces the advances of artificial
intelligence
technologies and the technical issues related to artificial
intelligence
in finance. Section II (Applications) talks about the
practical
applications, and some typical cases, of artificial intelligence
in
China’s financial sector along the three process units—front
office,
middle office and back office—of the business operations of
financial
institutions and the corresponding scenarios. Section III (Topics
in
Focus) analyzes and discusses a series of hot topics and thorny
issues
in recent years’ development of intelligent finance, e.g.
standards
system, governance principles and ethical issues, construction
of
financial data cloud, and sharing technology. Section IV
(Regulation)
highlights the risks brought about by intelligent finance and
new
challenges and requirements posed by intelligent finance to
financial
regulation, and how to strengthen the protection of data privacy
and
consumer/investor rights. Section V (artificial intelligence
Foreign
7
the world and explores how China can draw on foreign experience
in
advancing intelligent finance. Section VI (Policy Proposals)
synthesizes the policy proposals made previously in each topic
and
section as reference for relevant authorities in
policy-making.
The report attempts to balance the comprehensiveness, relevance
and
continuity of the contents, presenting a general picture of
the
development of intelligent finance, while selectively homing in on
hot
topics and thorny issues, and laying ground for continued tracking
and
researching as more practices emerge in this field. For this
reason, the
report endeavors to meet the needs of different readers and aims
to
provide practical references for financial professionals,
artificial
intelligence technology researchers and developers,
specialists
interested in this area and officials of regulatory
authorities.
Due to the nascence of the topic as well as time and resource
constraints, there are inevitably limitations to our report.
Comments
and suggestions are most welcome. We will carry out more
in-depth
research in the future and look forward to your continuous
support.
China Finance 40 Forum Research Group on
China Intelligent Finance Development Report 2019
The Dartmouth College held a summer seminar on artificial
intelligence in 1956 which initiated artificial intelligence as a
research
discipline. The journey of artificial intelligence in the past
six-plus
decades can be roughly structured in the following three
phases.
Phase One (1956-73): inference and evidence based on symbolic
logic.
The main technology was logical calculation or heuristic
programming for solving algebra applicationproblems, proving
geometric theorems and realizing machine translation. However,
the
theory and technology at the time was unable to tackle more
complex
problems. In the early 1970s, artificial intelligence ran into
a
bottleneck, and governments began investing less money into
artificial intelligence projects.
Phase Two (1974-93): knowledge engineering based on artificial
rules.
The main technology was expert systems using a series of
artificial
rules to process knowledge and aid decision making. Related
applications were rapidly developed and put into use. However,
due
to insufficient data available for representing knowledge by
artificial
rules, difficulty to describe tacit knowledge of experts, as well
as the
high costs of updating and maintaining expert systems, the
technology
could not be deployed on a large scale.
renaissance of artificial intelligence occurred with the
breakthrough
in deep learning based on artificial neural networks and the
rapid
development of big data. A milestone event is Google’s
AlphaGo
computer defeating Go world champion, Lee Sedol, in 2016. Big
data-
based deep learning models and algorithms have found
extensive
applications and huge success in machine translation, intelligent
Q&A,
game and several other fields. What’s more, their industrial
applications were instantly recognized, ushering the development
of
artificial intelligence into a new, big data-drive chapter.
Advance in basic theories, support by growing information
environment and increasing industrial demands are jointly
guiding
artificial intelligence into a new phase of accelerated
breakthroughs
and wider applications, which shows the following characteristics
and
trends:
(1) “Big data plus deep learning” has become a mainstream
intelligent computing paradigm. The new round of advancements
in
artificial intelligence technologies benefits from three
technological
advances: a new generation of machine learning algorithmic
models
represented by deep learning; the application of
high-performance
parallel computing technologies such as GPU and cloud computing
in
intelligent computing; and the emergence of vast data in the
mobile
Internet era to support the high-speed development of AI.
10
infancy. The new progress in artificial intelligence is seen mainly
in
dedicated application fields. Right now, artificial
intelligence
technologies are shifting from being “unusable” to “usable”, and
have
to overcome many bottlenecks to achieve being “useful”.
Therefore,
the deep-level development of artificial intelligence urgently
requires
transformative technologies. In the next step, advanced
cognitive
mechanisms of the human brain may be referenced to seek a
breakthrough in deep learning methods and thus create more
powerful
knowledge representation, learning, memory and inference
models.
(3) New forms of artificial intelligence are emerging in
large
numbers. Driven by new theories and technologies such as
mobile
Internet, Internet of Things, big data, supercomputing and
brain
science, as well as strong demand for economic and social
development, new machine learning methods, e.g. deep learning,
deep
reinforcement learning, generative adversarial learning,
transfer
learning and incremental learning, continue to emerge, and
relevant
research proliferated and flourished. Artificial intelligence is
stepping
toward communicating and cooperating with humans, and it
boasts
vast application potential since human intelligence and
artificial
intelligence each have their own strengths and can complement
each
other. The combination of human and machine will be the main
direction of future development
(4) AI is beginning to display great economic and social
potential.
With the gradual maturity of technology, artificial intelligence
has
technologies, including language recognition and image
recognition,
have reached or even surpassed human-level performance in
recent
years, while such technologies as intelligent search and
recommendations and automatic translation have already entered
the
stage of commercialization. Furthermore, artificial intelligence
has
begun to assist humans in doing high-end work. AI-powered
industries are developing rapidly with the rise of deep
learning
technologies and the sophistication of related algorithms.
ii. Definition and Significance of Intelligent Finance
No uniform definition has been established for intelligent
finance.
Based on research, we defineintelligent finance as a new
business
format based on deep integration of artificial intelligence
technology
with the financial industry, and it transforms the financial models
by
using machine to replace and outperform part of humans’
management experiences and capabilities.
Intelligent finance is closely related to, but sharply different
from
digital transformation and fintech. Intelligent finance bases
its
development on the digital transformation of financial
institutions
which provides the very infrastructure. As an advanced form
of
fintech development and the upgrade and transformation of
digitalization, intelligent finance defines the future development
trend
and the core competitiveness of the financial industry.
(1) Increasing the efficiency and reducing the costs of
financial
institutions. Intelligent identification improves accuracy
and
efficiency, intelligent credit services shorten review time,
and
intelligent customer service robots reduce labor costs.
Precision
marketing reduces customer acquisition costs while improving
marketing efficiency and performance. Intelligent claim
processing eases the workloads of surveyors and loss
assessors.
Intelligent operations reduce costs and significantly raise
the
efficiency of business processes.
company has customized health insurance plans for children
and
the elderly, resulting in sales increased by over 200 times
and
more than five million coverage in the past four years.
Intelligent
credit assessment, which suits the characteristics of the
Internet,
provides more than 100 million customers with one-time micro
consumer loans. By lowering the threshold from RMB 1 million
to zero, intelligent investment advisors allows ordinary
investors
to access investment advisory services. Financial
institutions
introduce a wider range of better financial services for
consumers
through the innovation of intelligent financial products and
services.
Intelligent risk control techniques allow financial institutions
to
offer early warning on risks, guard against frauds, protect
the
safety of users’ funds, and substantially reduce the losses and
risks
for themselves and their customers. Regulators can greatly
enhance their ability to comprehensively and effectively
prevent
risks with the help of intelligent financial regulation and
supervision.
characteristics while deeply integrating with AI.
Technological
attempts and improvements have been made in practices in
recent
years, and the in-depth applications of intelligent finance
will
continue to lift artificial intelligence technologies to new
highs.
iii. Technology Challenges Facing Intelligent Finance
Artificial intelligence technologies have many subdivisions,
the
applications of which have been comparatively faster in other
industries but remain challenged in the financial sector.
(1) Search engine and personalized recommendation
technologies:
As financial services gradually switch from offline to
online,
Internet-based search engine solutions are also being
gradually
used in online financial services. Unlike Internet-based
scenarios,
search advertising and personalized recommendations for
financial services are subject to more complex rules. For
instance,
wealth management products should be recommended to
customers with matching risk tolerance.
(2) Image and video recognition technologies: These
technologies
have been widely used in face recognition, text recognition,
automated driving, emotion recognition, security and other
scenarios. But when it comes to finance, computer
vision-based
identification may be exposed to malicious attack, and
information extraction of financial documents cannot
guarantee
100% accuracy, thus failing to meet the strict requirements on
data
and documents in the financial industry.
(3) Natural language processing and understanding
technologies:
Such technologies as machine translation, reading
comprehension
and dialogue generation have been applied in many financial
business scenarios. However, the challenge lies in the failure
of
developing models with sufficient expert knowledge and the
lack
of adequate corpora.
updates and iterations. This requires corresponding knowledge
graphs to renew and enrich contents more quickly.
In addition, the particularities of the financial industry
also
challenge the applications of artificial intelligence in
finance.
have a stronger demand for interpretability, posing a challenge
to
the extensive application of artificial intelligence technologies
in
finance.
(2) Uncertainty: Financial disciplines and participants are
changing
all the time, making it difficult to apply the rules contained
in
historical data or the experience summarized by experts. Such
constant changes contradict assumption that data is
independent
and the identically in AI, and thus requires innovation of
artificial
intelligence technologies.
(3) Privacy protection: Finance calls for strict privacy
protection, but
the data that artificial intelligence algorithms rely on are
often
very sensitive. Privacy protection has become a key challenge
for
the applications of artificial intelligence in finance.
(4) Biased algorithmic predictions: AI-based prediction
models and
their functions are inconsistent with the need of the
financial
sector, and may even be biased, affecting the fairness of
financial
services.
financial services scenarios involve continuous
decision-making,
the artificial intelligence technologies, needs more data and
requires simulators to generate sample data automatically. But
it
is difficult to automatically generate a large amount of data
by
simulating real-world financial operation scenarios based on
constant rules, and this limits the applications of
reinforcement
learning technology.
(6) Difficulties in learning games: The financial market is a
typical
second-order multi-agent ecosystem: each agent aligns its
strategies and behaviors to changes in the ecosystem, and the
agents can affect one another as well. In practice, the
decision-
making of each agent is not transparent, or their
decision-making
mechanisms are quite different, making it impossible to train
models in the traditional open way which is based on unified
rules.
To sum up, there is still a long way to go in realizing the
deep
integration of artificial intelligence and finance. For one, the
financial
industry should have more tolerance for artificial intelligence
and
continue to advance and improve novel artificial intelligence
technologies through applications. Second, the R&D of
artificial
intelligence technologies should take the particularities of
the
financial industry into full consideration, overcome difficulties
and
introduce new artificial intelligence methods and
technologies.
iv. Intelligent Finance Development Trends
Intelligent finance will reshape the operating mechanism and
17
logic of the financial industry. The first is to promote the
reallocation
of production factors and reduce transaction costs. The second is
to
rebuild the financial ecosystem by changing R&D models,
industrial
organization, division of labor, and interpersonal relationships.
The
third is to reduce information asymmetry, thus improving risk
identification, early warning, blocking and control capabilities.
The
fourth is to change the traditional financial logic from financial
data-
based to behavioral data-based. The fifth is to spur the
development
of reg-tech, use AI technologies to improve regulatory efficiency,
and
cut down compliance costs of financial institutions.
(Li Xiuquan & Zhang Junfang, Research Fellows at the
Chinese
Academy of Science and Technology for Development
Liu Weiqing, Senior Researcher at Microsoft Research Asia
Bian Jiang, Principal Researcher and Research Manager at
Microsoft
Research Asia
Asia)
1. Intelligent Identification
distinguish individuals by identifying their biological
characteristics,
including physiological characteristics (fingerprints, veins, face,
DNA,
palm prints, iris, retina, smell, etc.) and behavioral
characteristics
(keystroke, gait, voice, etc.). So far, typical intelligent
identification
technologies include fingerprint recognition and facial
recognition,
which have been put into massive applications.
Intelligent identification is (1) Inherent: biological
characteristics
exist in human bodies as inherent characteristic and attributes.
(2)
Unique: Each individual has unique biological characteristics.
(3)
Stable: relatively speaking, biological characteristics will not
change
with time and other conditions. (4) Universal: except for
certain
groups, everyone has these biological characteristics. (5)
Convenient:
biological characteristics do not need to memorize passwords
or
carry/use special tools, and will not be lost.
Intelligent identification technologies empower the financial
sector
mainly in the following three ways:
19
First, it reduces the costs associated with financial frauds and
raises
the efficiency of financial operations. The introduction of
intelligent
identification technology into opening accounts can slash
human
resource inputs in commercial banks and almost entirely eliminate
the
possibility of accounts being opened by identity thieves (vs.
0.05
percent in the past). Voiceprint recognition systems help
insurance
companies accurately identify policyholders. Trust companies
apply
intelligent identification technologies to on-site and remote
visual and
audio recordings and signing of transaction documents, which
can
accurately identify clients, meet compliance requirements and
save
labor costs.
Second, it extends the scope of financial institutions’ online
business
and optimizes customer experiences. Many financial institutions
have
realized the automatic review and online approval of small
personal
loans. With remote identification, they have simplified their
operation
procedures and addressed the pain point of cost and benefit
mismatch
of traditional risk control methods. For example, an
insurance
company has successfully used facial recognition in registration
and
authentication, logins, insurance application and claim
application,
with a biospy recognition rate of over 99%. The time from filing
to
reviewing an application for policy loans was shortened from
two
days to two minutes.
Third, it diversifies data dimensions of offline scenarios and
enhances
operational capacity of customers. Facial recognition makes
it
possible for financial institutions to target existing and
potential
profiles to enhance their ability to acquire and serve
customers.
Intelligent identification as a novel technology faces many
challenges
in applications, which mainly reside in the inadequate precision
of
detecting algorithms, insufficient terminal computing resources
and
the lack of unified standards for data collection.
2. Intelligent Marketing
Intelligent marketing, or precising marketing, uses
artificial
intelligence technologies to create multi-dimensional user
profiles
based on a rich set of characteristic data such as customers’
transactions, purchases and browsing histories, so as to tap
potential
demand of clients. Intelligent marketing connects financial
institutions with channels, personnel, products and customers, etc.
so
that their financial products and services can cover wider user
groups
and provide personalized and precise services to consumers.
Compared to traditional marketing methods, intelligent marketing
has
the following characteristics:
marketing methods, which is more efficient than traditional
marketing.
Moreover, it can reach ideal users and interact with them in a
more
natural, acceptable and convenient way.
method acceptable to users and facilitating their access to
financial
products. By accurately identifying target users from user groups
and
screening out their media and scenario preferences via
quantitative
analysis, intelligent marketing helps financial institutions make
the
best choices in advertising approaches, scenarios and timing,
and
improve both the cost-efficiency and effectiveness of
marketing.
Third, intelligent marketing predicts users’ demand and meets the
full
range of their needs. Intelligent marketing infers customers’
financial
service needs in different situations based on the behaviors of
similar
users, and arranges financial marketing in advance to gain leverage
in
the market.
improve their intelligent marketing models and methods based
on
feedbacks, and thus improve marketing effectiveness. A certain
bank
has created profiles for all of its private clients (more than 400
million)
through information integration, which enabled it to
automatically
recommend wealth management and fund products to hundreds of
millions of clients, substantially enhancing the efficiency and
success
rate of marketing over the traditional way.
A securities company uses machine learning algorithms to
accurately
locate potential customer groups, increasing the conversion rate
of
customers’ interest based on their individual preferences,
attracting up
to on average 300,000 daily visits to recommended news
columns.
In the future, financial institutions will continue to tap their
own
capabilities, seek cooperation and build service platforms
for
sustainable development in the field of intelligent
marketing,
particularly:
First, big data and artificial intelligence technologies will
define the
shared development direction of all parties in the financial
marketing
sector. All segments along the industrial chain of intelligent
marketing
for the financial industry are linked together through
artificial
intelligence technologies based on data generated by users.
Financial
institutions, third-party companies and marketing content
platforms
collect such data to create multi-dimensional profiles of
customers
and thus improve the reach efficiency of the financial
services.
Second, financial institutions will tap the value of their own data
with
the support of external technologies. Financial institutions
have
accumulated a vast amount of primary data on users which is
of
enormous marketing value. It has become a common practice for
financial institutions to go to third-party service providers
for
technological support owing to their weak technological base,
which
gives rise to the intelligent marketing’s model of cooperation
along
the financial industrial chain.
continue to optimize key technologies and actively build
intelligent
marketing platforms. Unlike financial institutions, third-party
service
providers of intelligent marketing do not have direct access
to
financial products or user data sources, so technology is central
to
their competitiveness in the industrial chain.
3. Intelligent Customer Service
process of large-scale knowledge base.
The traditional manual customer service system has such
shortcomings as high operating costs, high training costs, and
high
wastage of resources due to the answering of repetitive questions
by
customer service representatives. Intelligent customer service, on
the
other hand, can optimize and refine the intelligent knowledge
base
through self-learning, and help customers identify and solve
problems
in the shortest time possible, thereby raising the efficiency
and
effectiveness of financial institutions’ customer services.
As of the end of August 2019, a commercial bank’s online
robotic
text-only consulting service received 70 million messages, of
which
91% were handled by robots, equivalent to the workload of more
than
of risky POS transactions and account management fee
notification,
with core technical indicators exceeding 90% and the
single-channel
efficiency increasing by five times. In addition, intelligent
voice
navigation has replaced interaction using push buttons on the
menu
bar which was in use for nearly a decade with human-machine
voice
interaction. Presently, more than 16,000 people make inquiries
each
day, and the accuracy of navigation is up to 90%.
A securities company has been exploring how to apply
intelligent
customer service since 2017. In 2018, its intelligent customer
service
provided service for about 1.05 million times, around 41.2% of all
its
customer service orders, saving labor costs by about RMB 2.94
million. In 2019, its intelligent customer service provided service
for
0.93 million times, about 46.6% of all its customer service
orders,
expanding the intelligent service coverage by about 5.4% and
saving
labor costs by an estimated RMB 2.6 million. The intelligent
customer
service has been quite effective in reducing costs.
Fund companies can use intelligent customer service to
provide
investors with automated answering service and business
processing
services online. The intelligent customer service of some
fund
companies has been able to handle more than 90% of
business-related
questions, helping them cut customer service operating costs.
Artificial intelligence technologies have played a positive role in
the
business scenarios of insurance companies, e.g. insurance
renewal
Outbound phone calls by intelligent customer service of some
insurance companies have a success rate close to that of
human
representatives, while its work efficiency can be 1.2 times that
of
humans, saving 80% of labor costs for business. Some other
insurance
companies’ intelligent customer service can answer basic questions
on
a 24/7 basis, replace human representatives in 70% of the
scenarios,
save 80% of labor costs, boast an accuracy rate of over 90%, and
has
served customers for over 400 million times.
Meanwhile, trust companies use intelligent customer service
to
provide customers with automated consulting services in order
to
complete work more accurately and timely. It is estimated
that
existing intelligent customer service systems can answer more
than
85% of the common questions raised by customers, and enjoy a
more
substantial edge when answering frequently asked questions,
thereby
alleviating operating pressure and reasonably controlling
costs.
4. Intelligent Investment Advisors
Intelligent investment advisors, or robo-advisors,
automatically
generate customized asset allocation advice for users on the basis
of
their risk appetite, financial position and expected returns, and
keep
track while seeking dynamic re-balancing of portfolio, through the
use
of artificial intelligence algorithms and financial models such
as
modern portfolio theories.
Originated in the U.S., intelligent investment advisory
products
emerged in China from 2015. As per the projections of Statista,
one
of the leading statistics portals in the world, assets managed by
robo-
advisors in China will reach RMB 346.66 billion by 2019 and
expand
by 103.1% to RMB 737.05 billion by 2022.
Intelligent investment advisors are superior to traditional
manual
service in the following four respects:
First, providing a wide range of efficient, convenient
investment
consulting services. Thanks to the Internet and mobile phone
apps,
intelligent investment advisors are available any time to respond
to
customer queries and offer smart, dedicated and
around-the-clock
wealth management services.
Second, boasting low investment threshold, low fee rates and
high
transparency. Targeting middle class and low-net-worth
customers,
the intelligent investment advisory platforms have a capital
requirement of less than RMB 100,000. Intelligent investment
advisors fully disclose information on the range of financial
products
for selection and the detailed fees charged, and provide
customers
with real-time access to diagnostic reports of their
accounts.
Third, avoiding emotional investment behaviors and realize
objective
and diversified investment. The intelligent investment
advisory
platforms operate based on the internal algorithm strategy
modules
27
and propose the optimal solutions on how to allocate different
assets
in a portfolio.
customized scenarios. The intelligent investment advisory
platforms
can supply users with personalized risk assessment and manage
wealth for customers in a tailor-made way, leveraging the big data
and
cloud computing platforms behind it.
Intelligent investment advisory products fall roughly into three
types:
the first refers to start-up providers of intelligent investment
advisory
services which focus on rendering intelligent investment
advisory
services for institutions and/or individuals; the second refers
to
traditional Internet financial companies which derive their
advantages
from an abundance of long-tail customers and render updated
online
investment advisory services specific for wealth management
product
investments of fund companies; and the third refers to
traditional
financial institutions with naturally strong research or
sales
capabilities, which can integrate resources, customer bases
and
technological platforms at conglomerate level and provide
global
markets with asset allocation services across categories.
The emergence of innovative intelligent investment advisory
products
at commercial banks substantially enriches the product mix,
while
differentiated asset allocation services and retail services that
put
offline retail services in the shade provide customers with
more
choices. In the wealth management field, intelligent
investment
28
advisory products of a certain bank now manages more than RMB
12.9 billion worth of assets for 200,000 customers.
An intelligent investment advisory products launched by a
securities
company in 2016 has to date reported accumulative sales of more
than
RMB 36 billion, provided wealth management advice and
investment
recommendations for over 783,000 customers, recorded 538,000
active users each month, and boosted the capital by RMB 7.67
billion
for the company.
services in the future. One is to provide relatively
standardized,
simplified, and easy-to-understand investment products to meet
the
homogeneous needs of investors. This model is suitable for
Internet
financial companies and start-ups. The other is to leverage a
large
number of offline investment advisors and a wide distribution
of
business outlets and blend online with offline to meet the
personalized
needs of investors. This model is suitable for traditional
financial
institutions.
using artificial intelligence technologies, such as image
recognition,
biometrics and emotion recognition. The methods of early risk
warning and risk management are gradually evolving from
applications of machine learning and deep learning make risk
identification more accurate and effective.
At insurance companies, intelligent claim products
comprehensively
sort out and optimize the end-to-end process of auto insurance
claims,
covering all the steps in claim settlement from reporting and
dispatching, investigating and assessing damages, auditing
and
verifying claims, to settling claims, so that the auto and
property
damages, as well as personal injuries are assessed accurately
and
efficiently. Intelligent claim products help insurance companies
solve
the problems of frauds and low efficiency, and provide
policyholders
with premier service experience. With the help of image
recognition
technology, some insurance company can intelligently identify
the
images of damaged vehicles to automatically tell the vehicle
models,
damaged exterior parts, and distinguish between 23 different levels
of
vehicle damages. It matches the results of image recognition with
the
back-end database for automated pricing, completing loss
assessment
in seconds. Currently, applicable cases report an accuracy rate of
loss
assessment at more than 90%. After the insurance company put
the
intelligent online claims platform into application, case
processing
efficiency greatly improved, cutting back 30% manpower on
reviewing, and a total of over 25 million auto insurance claims
were
processed.
Institutions
distills the patterns hidden in massive macro-economic and
financial
market information with algorithms and independently
optimizes
models to predict the future trends of investment targets or
provide
early risk warning to improve investment decisions, inform of
and
control risks on a real-time basis.
Intelligent investment has three advantages over traditional
investment models:
on computer’s quick processing of large amounts of
information,
intelligent investment research can help analysts gather
industry
information, perform due diligence and raise work efficiency,
in
addition to assisting researchers in risk identifying and
prewarning.
Second, intelligent investment can reduce costs. Although the
development costs of intelligent investment-related platforms
or
models are high, relevant replication, promotion and operation
costs
are extremely low. Intelligent transactions can assist traders in
drafting
Third, intelligent investment can promote rational trading. It
is
impossible for analysts and traders to perform rationally all the
time
during transactions due to emotional factors, which may lead
to
transaction errors and investment losses. Machines, on the other
hand,
can avoid irrational behaviors and respond to the market with
pure
rationality.
has developed a “Crow Bond Default Prediction System” for
bond
default events. This system analyzed more than 4,500 bond issuers
to
forecast the probability of potential default, with an accuracy of
more
than 90% for the test set, and 100% for issuers who defaulted on
their
first-time bond issuance in 2019 as of August 2019. This
system
reduces the pressure on credit researchers while expanding the
scope
of investment, and thereby improves the company’s overall
investment research capabilities.
The index-enhanced products developed by a fund company based
on
AI-powered quantitative trading models have achieved good
returns
in testing on the CSI 300 and CSI 500 indexes. From March to
May
2019, its CSI 300-enhanced strategy ranked first in the industry,
with
excess earnings exceeding 4%; CSI 500-enhanced strategy
ranked
second in the industry, with excess earnings surpassing 6%.
By making a variety of credit decision-making rules and
combining
these rules in different ways, intelligent credit assessment
forms
differentiated credit access, limit and pricing strategies for
customers
in different scenarios and different credit life cycles.
The core feature of intelligent credit assessment is its
automated
information processing and credit decision-making process.
Online
intelligent credit assessment mainly collects transaction and
payment
data, external credit reference data and third-party data online.
It has
relatively high accuracy given the extensive coverage of data
across
multiple dimensions and a high degree of automation.
Commercial banks classify the borrowers and their assets into
different groups and serve them in differentiated ways through
data
screening, modeling, and prediction scoring. At the loan
approval
stage, commercial banks assess the risks of services to be
provided
and adopt corresponding risk prevention measures based on the
detailed information filed by customers and by such means as
fraud
prediction, credit scoring, price modeling and limit
management.
Internet banks base their intelligent credit assessment on the use
of
vast data. For example, an Internet bank usually bases its
online
intelligent credit assessment on operation data, financial data
and
100,000 frequently used indicators to judge the authenticity of
key
credit granting indicators with an accuracy of usually above 90%,
and
assess credit limits based on the true conditions of operations so
that
credit risk assessment is more accurate.
With intelligent financial technologies in place, financial
asset
management companies can effectively integrate and process
past
non-performing asset data and feed them into valuation and
pricing
models for bad assets as material and basis. These models are
then
continuously optimized to raise accuracy in trials and errors
through
the introduction and validation of new data and the testing
and
feedback of new practices.
3. Intelligent Risk Control
online financial risk control modeling. After the model accuracy
is
improved through massive calculations and verification exercises,
the
models are finally applied into financial business processes
including
anti-fraud control, customer identification, pre-loan approval,
credit
pricing and post-loan monitoring, so as to enhance the risk
control
capabilities of the financial industry. Intelligent risk control
provides
a control model based on online business for risk control in
the
financial industry, which covers the entire process of fraud
prevention,
customer identification and authentication, credit approval
and
pricing analysis, post-loan management and overdue
collection.
A commercial bank’s intelligent risk control platform “Libra
System”
uses advanced technologies such as big data analysis and
machine
learning to intercept and identify suspected fraud transactions in
30
milliseconds, reducing card frauds to 0.7 ppm, and thus
effectively
safeguarding customers’ money.
risk control technologies. Take ex-ante risk control system as
an
example. Through a flexible combination of nine types of
monitoring
indicators, it can centrally identify and monitor a total of 24
abnormal
trading rules in 11 categories. The system has identified
altogether 14
false orderings during opening call auction, three false filings
during
intra-day trading and four day-trading transactions in three
months
after being launched.
With the help of such technologies as image and video
recognition
and detection, and video tracking, a certain insurance company
can
provide loss assessment results within seconds, with an accuracy
rate
of more than 98%. It can also effectively identify
photoshopped
pictures and duplicate claims, reduce the risk of fraud, and cut
the
workload of loss assessors by 50%.
Customer rating, debt rating and early risk warning are involved in
the
risk control of financial asset management companies. An
asset
management company is building and maintaining a large-scale
inter-
35
bank default loss database, the first of its kind in China. The
database
now covers nearly 250,000 loan defaults of more than 100,000
delinquent customers across the country. Using the data, an
intelligent
risk control application has been put in place.
4. Intelligent Compliance Management
and management approach for financial institutions to improve
their
compliance capabilities, reduce compliance costs, and meet
regulatory requirements through the automation of data and
processes.
The intelligent knowledge engine developed by commercial banks
for
intelligent compliance and knowledge management can process,
manage, transfer and learn textual knowledge. Based on
natural
language processing and knowledge representation and reasoning,
the
engine pools an array of advanced technologies, such as
Q&A
matching, graph-based reasoning and semantic text retrieval,
processes the banks’ texts such as product manuals, policies
and
regulations in depth, so as to provide accurate intelligent Q&A
and
document query functions, judge and answer difficult questions
in
business processes, and perform advance review on the compliance
of
business processes. So far, the knowledge engine has been
widely
used in the review of bank’s process and operational compliance.
Its
application in a certain bank’s review of foreign exchange
operations
36
and increase the efficiency of single business review by 78%,
greatly
improving the overall operation efficiency.
Securities companies use intelligent semantic analysis technology
to
check financial documents in the areas of investment banking,
compliance management and research. Through text analysis and
semantic analysis, words, sentences, paragraphs, data, formulas
and
other information are automatically extracted from the documents
to
build a financial knowledge graph. Continuous optimization
and
training based on intelligent technologies such as deep learning
and
machine learning endow the computer with certain judgment
capabilities and make it possible to intelligently check and modify
the
documents, thereby easing the workload of manual review,
improving
the document quality and reducing operating costs. So far,
the
intelligent financial document review system has been able to
identify
semantic errors, check context consistency, examine data
articulation,
and validate financial indicators’ formulas. From July 2018 to
October
2019, the intelligent financial document review system of a
securities
company checked nearly 1,900 investment banking business
documents, inspected close to 550,000 data points, and helped
confirm the correctness and consistency of nearly 500,000
data
calculations.
contract texts, thereby reducing employee workload and the
probability of operational risk, and ensuring the compliance
across
business units. In addition, artificial intelligence may be
leveraged to
assist in automatic drafting, review and performance management
of
contracts based on historical data and industry rules.
iii. AI in Back-Office Business Scenarios of Financial
Institutions
1. Intelligent Operations Management
In terms of operations management, financial institutions can
further
unleash the internal vitality of their data assets, increase the
cost
efficiency of operations and push traditional operating models to
go
intelligent.
A commercial bank has added a mobile channel by opening a
virtual
business hall to provide customers with remote video teller
services,
which has greatly improved user experience and business
efficiency.
As of the end of 2018, 1,288 million customers had been
served
through the intelligent channel, with 99.56% of the services
being
provided by intelligent robots. Another commercial bank has
adopted
technologies such as big data analysis and machine learning to
predict
ATM transaction volume in order to optimize the amount and
timing
of cash refill, saving an estimated over RMB 40 million of
operating
costs in 2019.
up previously scattered and automated business operations via
identification. As a result, its operation departments
shortened
average account opening time by 44.85%; and increased the
amount
of daily tasks handled per person by 3.63 times than that
during
decentralized operations in the past. At the same time, the
securities
company has developed robotic process automation (RPA) system
to
simulate a series of routine computer operations, including
mouse
click, keyboard input, copy and paste. This non-intrusive
mode
integrates data and operations to automate business without
changing
the original IT architecture.
Based on strong foundation in the industry, a certain
insurance
company has established an indicator database and statement
template
library across the full business life cycle, covering
marketing,
underwriting, claim settlement, collection and payment, finance,
risk
monitoring, performance management and customer relationship
through reorganizing the analysis indicators of various
business
scenarios of insurance agencies. The system provides data
indicators
that fit the company’s business processes, saving 70% of the time
on
business analysis that is usually manpower-intensive and
time-
consuming.
operations management. The first is intelligent settlement
management, which automatically checks settlement results.
The
second is intelligent disclosure, which oversees reporting and
generate
monitoring reports semi-automatically. The third is intelligent
failure
fourth is intelligent failure handling, which diverts flows from
heavily
loaded server in advance based on real-time performance of the
site,
and thereby reduces the probability of server failure or the
consequent
impact on users. The fifth is intelligent protection of network
security
using artificial intelligence technologies.
transaction documents in paper form. In trust business
contracting,
transaction texts confirmed by both parties are sometimes printed
by
the counterparties first and then sealed and signed. In this case,
optical
character recognition (OCR) technology can improve
recognition
accuracy, quickly locate differences through comparison, and
facilitate manual verification. In the meanwhile, trust companies
use
online banking transaction robots to perform unified management
and
authorization of corporate online banking accounts, automatic
collection of online banking transaction information, and
convenient
inquiry of account balance, transactions and receipt
information.
2. Intelligent Platform Building
Intelligent platform is a core engine for financial institutions
to
improve services, reshuffle processes and pursue transformation
and
upgrade in the intelligent era, and also a key direction of
innovative
artificial intelligence applications.
information system transformation project, some commercial
bank
has realized product integration, process linkage and
information
sharing in key business fields. It has built three platforms based
on
cloud computing, big data and artificial intelligence
technologies
respectively to constantly provide a diverse set of
multi-dimensional
intelligent services for upper-layer applications, and also put in
place
an intensive operating services system featuring acceptance
over
different channels, centralization at head office, and front-,
middle-
and back-office integration.
In building an intelligent auto service platform, an insurance
company
integrates and shares its offline partners’ service resources, and
has
built an all-inclusive service platform covering auto repair, auto
use
and auto maintenance. The quality testing rate of auto rescue
services
has increased from 40% to 100%, 70% of the manpower for
rescue
management has been saved, and 3.8% of the loss has been
mitigated.
3. Intelligent Situational Awareness in Security
While accelerating the front-office innovations, it is also a must
to
actively and continuously promote information security
supervision
at the middle and back office against the pressure from
continuous
dynamic adjustments of the information security system.
41
From the beginning of 2019, a securities company has launched
the
construction of its network security situational awareness
platform
with big data technology as the base and intelligent security
analysis
as the core to support the enhancement of three core
capabilities
concerning information security, i.e. “threat detection,
situational
awareness and security protection.” The first capability is to
gradually
realize the collection and unified storage of security
element
information across the network. The second capability is to build
a
smart security brain with new analytical technologies such as
machine
learning and artificial intelligence. The third capability is
to
progressively introduce threat information from multiple
external
sources, to promot the information-sharing mechanism of the
securities industry, and to set up a shared information center.
The
fourth capability is to cover all threat detection scenarios in
a
business-oriented manner and with data collection steps as
the
roadmap.
Currently, intelligent finance is mainly utilized at the front,
middle
and back office.
identities. The mainstream intelligent identification
technologies,
represented by fingerprint recognition and facial recognition,
have
already been in massive use, in particular in remote
verification,
payment by face, smart outlets and operational security.
2. Intelligent marketing reduces marketing costs and enhances
service
efficiency and effectiveness. Intelligent marketing is undergoing
a
transformation from a division of labor between human and
machine
to human-machine collaboration. In the future, it will further
increase
the efficiency and effectiveness of financial services
through
integrated human-machine cooperation across sectors.
3. Intelligent customer service saves customer service
resources and
enhances service efficiency. Intelligent customer service not
only
answers questions automatically but also connects with each
front-end
channel to provide unified, automated customer services.
Furthermore,
it remains committed to improving itself to render high-touch
and
more efficient services around the clock.
4. Intelligent investment advisors are already available on a
trial basis,
but there is still some distance to go before full-scale
promotion.
Intelligent investment advisors have been in practical use at home
and
abroad. However, owing to a lack of clear business model and
service
positioning, Chinese financial institutions should take more steps
to
5. Intelligent investment begins to make profit and boasts
huge
development potential. With the help of artificial
intelligence
technologies, some companies continuously optimize
algorithms,
strengthen computing power, make investment forecasts more
accurate, increase returns and mitigate tail risk. They have
recorded
substantial excess returns in firm offers via portfolio
optimization.
Intelligent investment boasts huge development potential in the
future.
6. Intelligent credit assessment enhances the capabilities of
providing
credit services for micro and small lending. With real-time
online
operation, automatic system judgment and short review cycle,
intelligent credit assessment is in a superior position to provide
more
efficient credit service for micro and small lending. It is already
being
widely used at some Internet banks.
7. Intelligent risk control transforms the risk control
business of
Financial institutions. Intelligent risk control provides a new
risk
control model based on online business for the financial industry,
but
currently only a small number of financial institutions are capable
of
operations. Intelligent operations management centralizes and
smartens up previously scattered and automated business
operations,
thereby improving the efficiency of business operations,
reducing
business handling errors, and saving management costs.
Intelligent
operations have become the prioritized scenario for financial
institutions to carry out intelligent finance business.
9. Intelligent platforms empower financial institutions to
improve
services, transform processes, and seek transformation and
upgrade.
The construction of intelligent platforms is central to the
“go-
intelligent” initiative of financial institutions. It continues to
provide
a wide array of multi-dimensional intelligent services for
upper-layer
applications and builds a complete service ecosystem.
In summary, intelligent finance as a whole is currently still in
the
primary stage of “shallow applications”, mainly dealing with
intelligent transformation of routine and repeated tasks. The
applications of Artificial intelligence technology are
currently
penetrating to the core from the periphery of financial business,
and
have enormous development potential.
(Guo Weimin, Chief Scientist of the Bank of China
Yang Tao, Senior Manager at Fintech Research Center of the
Bank
of China
Wang Siyao, Manager at Fintech Research Center of the Bank of
China
Li Jinlong, Director of AI Laboratory of the China Merchants
Bank
Lu Songhua, Deputy General Manager of IT Management
Department of Haitong Securities
Haitong Securities
Qiu Han, Co-General Manager of OneConnect
Bi Wei, CEO of OneConnect Insurance Division
Wang Min, Executive Deputy General Manager & Board
Secretary
of ZhongAn Online P&C Insurance
Tian Hui, Head of Development Planning Department of ZhongAn
Online P&C Insurance
ZhongAn Online P&C Insurance
Liu Shuoling, Deputy General Manager of Fintech Department of
E
Fund
Department of E Fund
China Asset Management
Liu Haitao, General Manager of IT Department of AVIC Trust
Wang An, Senior IT Manager at IT Department of AVIC Trust
Meng Kaixiang, General Manager of IT Department of China
Minsheng Trust
International Trust
Corporation
Yuan Weibin, Secretary General of Fintech Research Institute
of
Tongdun Technology
Tongdun Technology)
III. Topics in Focus
i. Establish a System of Standards for Intelligent Finance
1. Overview of Financial Standards at Home and Abroad
Relevant international organizations, countries and regions
are
actively carrying out studies on financial standards to provide
a
fundamental support for the development of cross-border
financial
services, prevention of financial risks and protection of
consumer
rights and interests. Specifically, the International Organization
for
Standardization (ISO), the Financial Stability Board (FSB),
the
European Union (EU) and other global organizationshave set up
or
are studying the establishment of multiple standards for the
classification, coding, description and trading of financial
products.
Over recent years, China has made a fair amount of progress
in
building a new financial standards system. This system consists
of
government-led national financial standards and industrial
standards,
financial group standards independently established by the
market,
and enterprise standards. As of the end of June 2019, China had
issued
65 national standards and 251 industrial standards in the
financial
sector.
intelligent finance, which is increasingly expanding. There is
a
pressing need to build a system of standards for intelligent
finance.
The goal of constructing such a system is to guide the
systemic
performance of intelligent financial services in accordance with
a
scientific classification system, and solve common technical
and
management issues occurred during the provision of
intelligent
financial services. For the fundamental purpose of addressing
existing
problems in traditional finance and following the principles
for
constructing a system of standards, the intelligent financial
standards
system unfolds at five levels: applications in banking,
securities,
insurance, etc, operations management, technologies and
resources,
information security and basic common standards (see Figure
1).
intelligent financial services, and focus on standards such as
the
maturity model of intelligent financial service and guidelines
for
evaluating the quality of intelligent financial services.
Operations
management standards govern the day-to-day operation and
management of intelligent financial services, and highlight
the
standards for data asset management, outsourcing management
and
business continuity management of financial enterprises.
Technical
standards focus on the core technologies used in providing
intelligent
financial services, including the intelligent financial
technology
service requirements, and the standards for data center
construction
Applications Intelligent customer service
Operations management
Data security
Network security
General requirements on intelligent financial services
Classification of intelligent financial services
Intelligent financial service quality evaluation
Existing standards Key standards to be set Future standards
and operation and for the integration of emerging technologies
such
as cloud computing, big data and artificial intelligence into
intelligent
financial services. Information security standards deal with
the
security protection and management of financial companies,
and
revolve mainly around firm’s internal data security, network
security,
information system security and service security. Business
application
standards regulate the specific services provided by
financial
companies, which mainly include intelligent identification,
intelligent
marketing, intelligent customer service, intelligent
investment
advisory service, intelligent investment, intelligent credit
assessment
and intelligent risk control.
The following principles should be considered in building a system
of
standards for intelligent finance: First, it should be demand-led
and
designed at the top level. Led by the development and
application
demands of the banking, securities and insurance sector, the
standardization efforts should be pushed forward in a coordinated
and
orderly manner under scientific planning and proper top-level
design
in order to carry out the standard research and implementation,
with
the goal of putting in place a system of standards and
conducting
51
breakthroughs are needed. Efforts should be made to classify and
sort
out the needs of intelligent financial standardization and carry
out
orderly planning factoring in both long-term objectives and
current
work; research on and setting of key standards for basic, key
and
advanced sectors should be promoted first. Third, there should
be
concerted efforts with companies playing the leading role. Efforts
of
all players should be pooled in to promote financial
standardization.
Banks, securities companies and insurance companies should play
a
leading role. Relevant standardization organizations such as
financial
industry alliances and IT service subcommittees, should
strengthen
their communication and cooperation.
2. Measures to Improve Intelligent Finance
Standardization
The first is to set up a joint working group of financial unions
and the
Information Technology Service Standards Sub-Association. IT
standardization experts should join fintech professionals in
establishing the standard-setting working group for the purpose
of
galvanizing the standardization of intelligent finance in China.
The
second is to provide proper top-level design for the system
of
standards. In view of the intelligent financial service needs and
pain
provide a basis for standard setting/revision planning and
standardization arrangement. The third is to accelerate the
implementation of existing standards in the financial
industry.
Intelligent financial service standards and existing mature
standards
should be coordinated in a bid to adopt common standards,
prevent
repetition of standards and speed up the standardization of
intelligent
finance. The fourth is to actively study and set key standards.
Priority
should be given to setting the key standards which are absent in
the
system of intelligent financial standards on the principles of
“setting
common standards and urgently-needed standards first ”.
ii. Governance Principles of and Ethical Issues in
Intelligent Finance While creating countless opportunities and
expectations, intelligent
finance also continues to challenge existing laws, ethics and
order.
The issues brought about by the applications of intelligent
finance
need to be resolved jointly by the government, market and society,
in
an effort to mitigate the risks of intelligent finance, maximize
the
productivity unleashed by artificial intelligence technologies,
and
enjoy the benefits of scientific and rational
decision-making.
Intelligent finance ushers the financial service system into
the
“machine-based” service era from the “human-based” era.
However,
investors could lose money due to poor data quality or
defective
algorithms. Intelligent finance relies on algorithms, but
procedural
errors such as “overfitting” in algorithms may trigger butterfly
effects
and cause systemic risks. Financial decisions rely on
intelligent
processing of big data, where personal investment information
or
sensitive company data could be leaked, highlighting the need
to
protect personal privacy and data security; intelligent finance
blurs the
boundaries between different business, and thus requires
collaborative
joint governance by all relevant parties.
In June 2019, the National Governance Committee for the New
Generation Artificial Intelligence issued Developing Responsible
AI:
Governance Principles for New Generation AI, emphasizing that
all
stakeholders concerned with artificial intelligence
development
should follow the eight principles: harmony and
user-friendliness,
fairness and justice, inclusion and sharing, respect for privacy,
safety
and controllability, shared responsibility, open and collaboration,
and
social, and environmental sustainability, and jointly build a
community with shared future for all mankind.
Intelligent finance needs to abide by the general governance
principles
for AI. Meanwhile, it should take the particularity of applications
in
finance into consideration, and insist that both innovative
applications
and risk prevention are given equal emphasis. For one, the
innovation
in artificial intelligence technologies and the financial
industry
models should be encouraged. For another, effective
regulatory
measures should be implemented. It is important to follow the
patterns
of financial development and prevent the use of artificial
intelligence
technologies in circumventing financial regulation or
exercising
regulatory arbitrage. Multi-level players should be encouraged to
take
part in governance, with government agencies setting standards,
high-
tech companies ensuring the security and controllability of
technologies, financial institutions increasing the transparency
of
product applications and consumers participating in
rule-making.
2. Ethical Issues in Intelligent Finance
55
Intelligent finance may infringe upon the rights and interests of
users.
For instance, the opacity of algorithms can possibly lead to
morally
wrong decisions or discrimination. The sources of big data can
cause
trouble, too. When data is incomplete, unrepresentative or biased,
the
decision-making results will be affected. Therefore,
financial
institutions are obliged to understand artificial intelligence
systems
and the potential negative impacts they may have on their
customers,
and to take responsibility for any discrimination caused by
algorithms.
Many institutions around the world have begun to explore
countermeasures to deal with the ethical issues in intelligent
finance.
Bank of America has set up a committee to study how to protect
user
privacy. Google recommends a human-centered design approach,
using multiple indicators for evaluation and monitoring, and
inspecting data extensively to identify possible sources of bias.
The
Department of Finance Canada issued a policy paper outlining
the
quality, transparency and public accountability of using AI.
Some
universities and research institutes are studying how to
pre-process
data to reduce bias.
To develop intelligent finance, there must be clear guidelines
and
safeguards to ensure the technologies are properly developed
and
Institutions practicing intelligent finance must ensure their
employees
responsible for processing data or developing, validating and
monitoring artificial intelligence models have valid qualifications
and
experience, understand the possible social and historical biases in
the
data, and know how to adequately correct them. Financial
institutions
also need to formulate internal policies and management
mechanisms
to ensure that algorithmic monitoring and risk mitigation
procedures
are adequate, transparent, regularly reviewed and updated.
iii. Innovation Platforms and Basic Environment of AI 1.
Open-source Frameworks for AI
Artificial intelligence development is picking up speed in
becoming
open-source, platform-based and ecological. Open source and
openness define the trend of artificial intelligence
development
throughout the world. Global competition in open-source
artificial
intelligence platforms mainly takes place among foreign tech
giants.
Although a few Chinese high-tech companies have rolled out
their
own artificial intelligence platforms, there is still a big gap
between
them and those of foreign tech giants.
Open innovation platforms play a crucial role in advancing
artificial
intelligence technologies. In August 2019, China’s Ministry
of
Science and Technology announced a plan to start developing
open
innovation platforms for next generation AI, a move aimed at
building
national open innovation platforms in autonomous driving, smart
city,
intelligent healthcare, intelligent voice recognition, intelligent
vision,
visual computing, marketing intelligence, basic software and
hardware, inclusive finance, video perception, intelligent supply
chain,
image perception, security brain, smart education and
intelligent
home. To date, the number of national open innovation platforms
for
next generation artificial intelligence has risen to 15.
By building open innovation platforms, leading players in
artificial
intelligence open industrialized technologies such as speech
recognition, image recognition and natural language processing
to
users through interface. They promote the deep integration of
artificial
intelligence with the real economy in an application
demand-based
manner through efficient combination of such resources as
talents,
technologies, data and industries, and in so doing drive the
development of micro-, small- and medium-sized enterprises as
well
sectors, lead the innovation of artificial intelligence
technologies and
the improvement of industrial ecosystem in China, and spur
the
utilization of AI-related research findings in industries.
3. Open Innovation Platforms for Intelligent Finance
Large financial institutions have launched attempts to build
open
innovation platforms for intelligent finance. A financial
holding
company has established a business model driven by “finance
plus
technology”. Leveraging multiple financial innovation platforms,
the
company applies artificial intelligence technologies in finance,
creates
AI-based “end-to-end” solutions, sorts out the whole process
of
business and solves the pain points of the entire industry.
Thousands
of financial institutions are using the services provided by
the
company.
The Ministry of Science and Technology has accelerated the
development of pilot zones for innovation-driven development
of
entrepreneurial environment favorable for artificial
intelligence
advances. Technology demonstrations, policy experimentation
and
social experiments will be carried out in the pilot zones to
accumulate
experiences for the healthy development of artificial
intelligence
which can be replicated elsewhere. So far, the construction of
national
pilot zones for innovation-driven development of new
generation
artificial intelligence in Beijing and Shanghai under the support
of the
Ministry of Science and Technology has made positive
progress.
iv. Financial Data and Financial Clouds 1. Types of
Financial Data
Data processed in intelligent finance can be either structured
or
unstructured based on their format.
Structured data: Traditional financial information systems
mainly
process structured data, e.g. transaction records, customer
databases,
price data and market data, which is typically processed in
relational
database. The main tasks involve optimization, classification
and
predict future market trends.
Unstructured data: As IT and big data technologies are applied
more
extensively, a large amount of raw data has been accumulated
during
financial operations, and data sources have gradually expanded
from
traditional structured data to electronic documents, images,
audios,
videos and web pages. Structured data that can be processed
in
relational databases accounts for about 20% of the total, while
the
other 80% is unstructured. Multimedia data (voice and image),
commonly in the forms of scanned financial statements, scanned
bills,
scanned ID documents, accident site photos for insurance claim,
face
images, customer service recordings and voice inputs, mainly
depends
on deep learning technologies.
2. Methods to Process Financial Data
Data of different scenarios has varied demands on technologies. So,
a
diverse set of technologies is often needed, mainly
including:
First, knowledge acquisition-related technologies, which can be
sub-
divided into two categories: One, data recognition technologies
that
are primarily composed of recognition of scanned documents
(mainly
based loss assessment and speech recognition. Currently, this
group
of technologies mainly use deep learning in combination with
traditional machine learning and neural network methods. Two,
data
understanding technologies that are primarily made up of
document
analysis, document review, clause analysis and understanding,
notice
analysis, research report analysis and public opinion analysis.
Data
understanding goes a step further than data recognition. For
example,
in the case of processing scanned financial statements,
recognition
technologies only restore the table structures and the characters
in
cells; but understanding technologies go beyond to include
accurately
restoring the tables and semantics of the characters and carrying
out
table merge, unit understanding, subject alignment, entity
disambiguation and the like. Therefore, there is a need to use a
mix of
related technologies, such as natural language processing,
knowledge
graphs, image processing, and expert systems.
The second group is knowledge modeling and analytical
techniques
in the financial sector, which can be sub-divided into four
categories:
First, entity profiling technologies, including customer
profiling,
enterprise profiling, banking product mapping, securities
product
subsequent matching, filtering, recommendation, querying, etc.
The
main processing methods are statistical machine learning,
entity
linking/alignment/disambiguation in natural language processing
and
knowledge graph query/fusion. Second, statistical analysis
techniques,
including transaction anomaly detection, connected account
analysis
and time series analysis. Statistical features of data are
summarized
and extracted to support applications in risk control and
investment
modeling, etc. The bottom layer is mainly based on machine
learning
models, e.g. classification, regression, clustering, and
dimensionality
reduction, to which deep learning methods are also applicable.
Third,
graph modeling technologies, including large-scale graph
analysis,
financial statement analysis, industrial chain modeling and
financial
knowledge modeling (such as articulation relationship). They are
used
to gather instance knowledge in different domains and aid
decision-
making. Its bottom layer is mainly based on knowledge graphs.
And
fourth, expert knowledge, including rule modeling and reasoning,
rule
conflict detecting, smart contracts and regulations modeling. They
are
used to gather the experiences of experts or the standardized
processes
accumulated by organizations, and present such experiences
and
with the support of natural language processing technologies.
The third group is data interaction technologies, which can be
sub-
divided into three categories: First, information retrieval
interaction
techniques, including legal queries, intelligent Q&A (such as
account
opening assistance, outbound phone call assistance and query
assistance) and voice-based customer service. They are mainly
based
on natural language processing technologies, with the support
of
knowledge graphs and speech processing technologies. Second,
responsive interaction techniques, including intelligent push
(market
warning, marketing warning and risk event warning), precision
marketing (graph-based matching) and constraint checking (e.g.
non-
compliance warning). They are mainly based on knowledge
graphs,
with the support of machine learning. And third, visual
interaction
techniques, including automated charts, automated reports,
visual
information disclosures and graph visualization. They are
mainly
based on knowledge graphs and natural language generation
technology.
64
volume of data, which is still inadequate for machine deep learning
as
some of the data is low-dimensional and small sample-sized,
unable
to meet the requirement of artificial intelligence technologies for
large
samples and high-dimensional data.
Second, data isolation. For data protection, financial institutions
are
not allowed to make all of their data accessible. Isolated data
islands
also exist among financial regulators and thusly hinder the
development of intelligent finance.
Third, poor data integration and governance. As per a sample
survey,
91% of banks are inadequate in data governance, only the other
9%
has achieved effective data cleaning that can support the needs
of
intelligent financial applications. Some securities companies
have
accumulated years’ worth of raw investor data. To use the data,
they
have to spend 70% of the project time in cleaning the data and
allocate
hundreds of employees to mark the data. They are trapped in a
dilemma of premising “go-intelligent” on “intensive
manpower”.
4. Financial Cloud is an Important Infrastructure for
the
Development of Intelligent Finance
of artificial intelligence technologies. As a new type of
infrastructure
and an enabling platform, financial cloud empowers financial
institutions in large scale to reduce costs and acquire agility
through
Internet platforms, lowering the threshold for adopting such
complex
technologies as distributed technology and AI, and providing
great
computing power and data storage resources for intelligent
financial
applications. Thus, it has become a consensus in the financial
industry
to support digitalized and intelligent business innovation
with
financial cloud.
By service deliverable, cloud computing for intelligent finance
offers
three forms of service: Infrastructure as a Service (IaaS),
Platform as
a Service (PaaS), and Software as a Service (SaaS). In an IaaS
model,
users remotely access computing, storage and network
resources
provided by cloud service providers over the high-speed
Internet
using software virtualization and automated deployment
technologies.
In a PaaS model, financial institutions utilize data processing,
model
training, service deployment and other service resources provided
by
platforms to create intelligent financial applications. In a SaaS
model,
financial institutions directly use intelligent financial
applications
provided by cloud platforms to serve users.
finance are divided into public, private and hybrid clouds.
Public
clouds allow financial institutions to have direct access to
cloud
services without the physical servers of cloud computing.
Private
clouds consist of cloud computing services used exclusively by
a
financial institution, for which the financial institution owns
the
physical server. Hybrid clouds combine the two models.
Financial cloud plays an important role in spurring the
development
of intelligent finance mainly in the following four respects:
First, financial cloud supports massive high-frequency data
processing in intelligent finance. Cloud computing slashes
the
application costs yet boosts the processing capacity of
intelligent
finance, and allows financial transactions to become fully
electronic