customer acquisition, and claims
AI Assessment of Claims and Risks
Take auto insurance as an example.
When an accident happens, it often
takes a long time to process a claim.
Customers wait onsite for investigators to arrive and assess the damage,
they wait as the claim is filed and
processed, and they wait for a final
decision. It is inconvenient for customers and costly to insurance companies. The process is also vulnerable
to fraudulent claims. To address such
problems, Ping An developed a system where the customer only needs
to take a few pictures of the damaged
car to file a claim from the accident
site. The claim is processed within
seconds and the customer given a
precise payment calculation. The system involves a series of key modules:
picture quality assessment, verification of insurance, car segmentation, identification of damage and
related parts, payment calculation,
and fraud detection. A number of AI
techniques, such as image processing, image segmentation, and object
recognition, were developed to support the functions. The system has
been running in production at Ping
An for over a year, successfully processing over 30,000 claims each day.
It not only improves claim processing
efficiency and thus customer experience, but also stops potential frauds
on the order of multi-billion RMB.
This system is now available to the insurance industry through the Ping An
Investment banks often need to
assess the potential value and risk of
targeted customers—individuals for
retail banking or enterprises for corporate. In today’s big data era, information comes from a broad range of
resources with complex relationships
between them. To make a precise assessment, it is crucial to organize and
analyze such complex information
in an efficient and effective manner.
This is exactly what knowledge graph
is designed for. Ping An has developed various knowledge graph techniques for retail and corporate businesses.
Take corporate risk assessment,
for example. There are over 70-mil-
lion registered enterprises, includ-
ing households, in China. Their in-
formation comes from three major
sources: commercial registration
and daily operation; public news an-
nouncements and social posts; and
business relationships including the
supply chain, investments, and le-
gal actions. To organize and analyze
such rich and dynamic data, Ping An
developed Euler Graph, an enterprise
knowledge graph. The graph covers
nearly all of China’s 70-million en-
terprises, using data from all three
sources. Millions of legal proceed-
ings are automatically interpreted
and over 40-million lawsuit relation-
ships have been extracted and incor-
porated into the graph. Signals on
enterprises are collected from over
300 news and social sites, totaling
hundreds of thousands of articles
daily, and updated every 10 minutes.
Information from these and other
sources grows quickly. Deep graph
analysis algorithms support business
decisions on risk assessment and
other matters. One advantage of Euler Graph is that business logic is directly integrated. For example, risks
are assigned using different business logic for investments, bonds,
or loans, and signals are extracted
from an analysis of social and news
data. Upstream and downstream relationships may also be encoded as
risk indicators. When a risk event occurs upstream, the incident passes
through the graph network and may
influence an assessment. Through effective analysis by Euler Graph, risks
such as defaults were successfully detected three to nine months ahead of
occurrence. Euler Graph is also used
for other applications, such as precision marketing and exploring investment opportunities.
Large-Scale Blockchain Architecture
Ping An OneConnect has identified
various shortcoming impeding the
wide scale adoption of blockchain.
Performance and scalability bottlenecks have hindered its potential in
building high volume financial transaction systems, and issues of data
privacy and confidentiality have limited its usage in public service areas
where few entities are willing to share
Ping An’s blockchain research and
cryptography team responded with
the FiMAX platform. The architecture
is designed to address all key problems hindering large scale blockchain adoption, with performance
matching traditional databases systems and privacy protection enabled
by advanced cryptology including
various Ping An designed zero knowledge proof algorithms.
FiMAX has not only earned praise
from Ping An’s business partners, it
has also gained recognition with its
selection for some of the largest international blockchain networks being built for banks and regulators.
For example, one cross border blockchain network to be launched later
this year will comprise over 10 international banks and over 100 nodes.
Ant Financial made a series of innovations that led to key technologies
behind mobile payment and microloan services in China. Ping An used
innovative techniques to improve
financial services for insurance, investment, and banking industries.
Much progress has been made, but
every problem solved opens the door
to further questions and considerations. How should we model the
transaction systems in a large-scale
dynamic network, and implement
intelligent inference and reasoning
for better financial services? How
can data be utilized and user privacy
protected at the same time yet better
than through current methods such
as differential privacy? How can causal inference be applied in a complex
system and when only observational
data are available? Answering these
questions will lead to tomorrow’s
Yuan (Alan) Qi is Vice President and Chief Data Scientist
at Ant Financial, Zhejiang.
Jing Xiao is Chief Scientist and Executive Member at Ping
An Insurance (Group) Company of China LTD, Shenzhen.
© 2018 ACM 0001-0782/18/11 $15.00