Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
生命保険論集第 200 号
―1―
Guest Lecture
by Professor Dr. Gene Lai
ジン・ライ教授による講演会
(from17:00 to 19:00 at the large conference room on the 9th floor)
(於 日本交通協会9階 大会議室 17時00分から19時00分終了)
******************************
Technology, Big Data, and Insurance Industry (Gene Lai)
Thank you so much for inviting me. It is a great honor. Before I start my
presentation, I want to tell you that my parents speak Japanese. If they
wanted to say something that don’t want me to know, they would speak
Japanese. As a result, I learned some Japanese. And my favorite food
actually is not Chinese food, actually it’s Japanese food. And I use a
Japanese car. I had a two car both of them are Japanese car. My stereo
system was Sony. And even for the electronic shaver, it was made in Japan.
ジン・ライ教授による講演会
―2―
The shaver is still working well after I bought it 18 years. In other words,
I love Japanese product culture.
So today I want to talk about technology, big data and insurance industry.
Even I do a lot of academic research, I have never done big data academic
research. But I thought big data was an interesting area and spent some
time on it. I hope that you will find my speech helpful. Mainly I will
talk about the trend about the technology and big data and how we can use
in the insurance industry. Please note that my speech is not my
original contributions. I got the materials from various sources.
First, I would like to provide some background about technology. The
growth of internet connected devices and sensors, which are projected to
reach 50 billion by 2020. The growth in smartphones and tablets, coupled
with cloud computing, which provide constant access to the internet. In
addition, International Data Corporation (IDC) reports that the digital
universe will grow 300-fold between 2005 and 2020 to a total of 40 trillion
gigabytes. Yet, according to IDC, only one percent of this data is currently
being analyzed.
Advances in Artificial Intelligence techniques, such as machine learning,
natural language understanding and intelligent decision-making will allow
insurers to advance from using technology for transaction processing to
decision-making. For example, Tobias Preis, and two physicists published
a paper titled “Quantifying Trading Behavior in Financial Markets Using Google Trends.” They found that as the volume of searches for
words like “debt,” and “stocks” fell, the Dow Jones tended to rise.
Another great example is Gilt Groupe, which reached $500 million in sales
生命保険論集第 200 号
―3―
in just five years. Areport from McKinsey & Co. describes, Gilt Groupe
mines extensive customer data to personalize all e-communications.
By 2020, a number of biotechnologies will be available at the nanoscale,
providing the ability to embed devices and sensors unobtrusively within the
human body. The nanotechnology drug delivery market is expected to grow
at CAGR of 21.7% between 2009 and 2014, and reach almost $16bn by
2014. Such nanotechnologies have the potential to dramatically improve
health outcomes through enhanced monitoring and preventive control of
chronic disease.
Social media has drastically altered the landscape of personal and
professional communication as we know it. Companies across financial
services have adapted to involve social media as part of their core marketing
initiatives, and at an accelerated pace. At the midpoint of 2013, the majority
of financial institutions are on Facebook, Twitter and YouTube, as well as a
host of additional social media mediums.
Customer expectations: Customers are increasingly demanding simplicity,
transparency and speed in their transactions with businesses, including
insurance agents/advisers and carriers. The relentless march of online and
mobile technology is continuing to fuel this change in customer
expectations.
In a recent survey of US consumers, more than 32% of all respondents
and 50% of those aged 18 to 25 said they prefer to work directly with
insurance carriers.
More and more insurance will be ‘bought’ by customers as opposed to
being ‘sold’ by agents destroying the age old wisdom of ‘Insurance is sold
ジン・ライ教授による講演会
―4―
and not bought’.
Foster innovations in product/service design and delivery. Leading insurers
will get better at targeting customers and customising product and service
attributes to meet their specific needs.
Please also note that information security and risk is a top priority. Many
corporations and universities suffered from cyber attacked.
Next, I will like to talk about how to address the issues related to technology
changes. There are a lot of investments in mobile and interactive
technologies for multimedia content creation from insurance companies.
For example, Zurich Insurance, a multi-line insurer, is using a
MicroStrategy Mobile app that helps ensure transparency of operations and
compliance with financial regulations. We also need to discuss who is liable
for new kind of risks (e.g., driverless cars, software bug downloaded to
millions of cars wreaks havoc, or a hacker taps into the transportation
network)?
The ability to measure and monitor everything that those insurance products
cover in real-time (including people) raises the profile for behavioral
sciences
Machine learning is here to solve workforce issues. The insurance
workforce is aging, and as has happened in many other industries,
automation is filling the gaps left by retirements. Automated underwriting is
already relatively popular, but soon artificial intelligence and machine
learning platforms could pop up in other areas as well
I would like to talk about Social Networks. In just 8 years since its launch,
for example, Facebook has attracted over 1 billion users. The experience
生命保険論集第 200 号
―5―
in in other industries will be discussed below. For example Yelp for food
or lodging and product reviews from Amazon. Amazon show you both
the positive and negative comments.
People want access to better health care or insurance product or services
using social media. People will look up the reputations of doctors, merits
of insurance products, and rankings of services of insurance companies.
The medical service and treatment model is evolving towards the
customization of healthcare. Consumers will eventually use personalized
medicine to create highly customized healthcare solutions that actively
change the body’s biochemistry in response to risks and conditions that are
unique to each person.
Sam Friedman, author of Mobile engagement: Insurers look to connect with
consumers on the go posits that the industry faces two basic challenges.
First, raising awareness and adoption and second, moving beyond the basics
like bill-paying to stickier, more frequent interactions.
Next, I would like to talk about big data. There are many definition of Big
Data. There are many definitions. In general, there are three
requirements: too large for common database, the data set must be too
large for common database management tools to handle, and the data must
come from multiple sources. Finally, the data must provide insights that
improve decision-making.
The computer industry had always operated under something called
Moore's Law, which presumed that computing speed would improve by a
factor of 10 every six years. Now, companies now can analyze data that
previously was too voluminous, or came from too many different sources.
ジン・ライ教授による講演会
―6―
Next, we like to address how big data is classified. First, by structure (or
lack thereof). Structured data is information stored in numerical formats
recognized by computers, such as spreadsheets and customer purchases.
Unstructured data, meanwhile, is the stuff that has no intrinsic numerical
framework—such as videos, photos, text and social media commentary. In
fact, some people believe “big data” applies only if part of the data is
unstructured.
Big data can be classified by source. Some people like to categorize big
data according to where the data came from. For example, there is mobile
data (such as information produced by smartphones), social media data
(such as tweets and website comments), and public data (like census
information).
Third, big data can be classified by analytics program. Liebowitz says a
third common way to categorize big data is by the type of analytics
programs needed to make sense of the data. These include such examples as
social media analytics, mobile analytics, video analytics, sensor-based
analytics, etc.
Insurers who are able to use real-time ‘big data’ and advanced
forward-looking simulation techniques will establish a significant
competitive advantage.
Next I would like to talk about IBM general solutions for big data. We
will start with big data exploration. Find, visualize, understand all big data
to improve decision making. Big data exploration addresses the challenge
that every large organization faces: information is stored in many different
systems and silos and people need access to that data to do their day-to-day
work and make important decisions. Second, extend existing customer
生命保険論集第 200 号
―7―
views by incorporating additional internal and external information sources.
Gain a full understanding of customers—what makes them tick, why they
buy, how they prefer to shop, why they switch, what they’ll buy next, and
what factors lead them to recommend a company to others. Third, we will
talk about security intelligence extension. Lower risk, detect fraud and
monitor cyber security in real time. Augment and enhance cyber security
and intelligence analysis platforms with big data technologies to process
and analyze new types (e.g. social media, emails, sensors, Telco) and
sources of under-leveraged data to significantly improve intelligence,
security and law enforcement insight. Please note that operations analysis
is very important. Analyze a variety of machine and operational data for
improved business results. The abundance and growth of machine data,
which can include anything from IT machines to sensors and meters and
GPS devices requires complex analysis and correlation across different
types of data sets. By using big data for operations analysis, organizations
can gain real-time visibility into operations, customer experience,
transactions and behavior.
Finally, we will talk about data warehouse modernization. Integrate big
data and data warehouse capabilities to increase operational efficiency.
Optimize your data warehouse to enable new types of analysis. Use big data
technologies to set up a staging area or landing zone for your new data
before determining what data should be moved to the data warehouse.
Offload infrequently accessed or aged data from warehouse and application
databases using information integration software and tools.
IBM Solutions for the Insurance Industry will be discussed next.
First, insurance companies need to create a customer-focused enterprise,
ジン・ライ教授による講演会
―8―
improve customer retention and satisfaction, accelerate offer acceptance,
learn customer attitudes, optimize cross-sell and up-sell. They can sell
from auto insurance to life insurance or from life insurance to long-term
care insurance and from low benefit to high benefit. As far as optimize
enterprise risk management, insurance company should prevent, predict,
identify, investigate, report and monitor attempts at insurance fraud. They
need to identify patterns and trends that can pinpoint fraudsters quickly and
improve fraud prevention in the future. Insurance companies can use big
data to prevent unexpected losses, civil and criminal penalties. Finally, big
data should help to accurately manage underwriting, reinsurance and
catastrophe bond pricing.
Second, insurance can optimize multi-channel interaction to increase sales
channel productivity, minimize infrastructure costs, increase availability
of low cost self service labor options, and increase flexibility and
responsiveness to changing customer preferences.
Big data can be used to increase flexibility and streamline operations.
Specifically, insurance companies can Increase revenue from subrogation,
increase revenue per customer, lower payout of fraudulent claims, reduce
operating costs, increase response time to new compliance mandates.
Insurers already hold vast amounts of data, but now they can gather even
more from new sources such as GPS-enabled devices, social media postings
and TV footage.
Big Data has been used as Predictive Models. The key to unlocking this is
through data analytics, the process of examining large amounts of data of a
variety of types to uncover hidden patterns, unknown correlations and other
useful information. Such information can provide competitive advantage
生命保険論集第 200 号
―9―
and result in business benefits, such as more effective marketing
and increased revenue. For example, brokers are using advanced
technology to better analyze risks for clients to lower costs and manage
claims. Many easy to use software has been developed.
One survey shows that only 17 percent of insurers said they routinely use
analytics as part of an integrated, enterprise-wide approach, while 36
percent of insurers (compared with 20 percent of the total sample) have the
resources and ability to use analytics, they apply them in tactical—rather
than strategic—applications. Notably, 20 percent of insurers said they
“don’t know” how their organization was using data and analytics. In a
data-dependent business like insurance, that’s an eye-opener.
In its industrywide survey on the topic, Accenture finds that insurers
actually have an edge over other industries when it comes to big data, but
aren't taking advantage of it. While 36 percent of insurers (compared with
20 percent of the total sample) have the resources and ability to use
analytics, they apply them in tactical — rather than strategic —
applications.
It should be noted that insurers should use big data strategically and to serve
the entire enterprise, versus a single department or business line. “There is
huge potential in data analytics in insurers to better understand customers,”
says Costonis. “This includes the ability to make smart marketing decisions,
to introduce refined products and pricing, and finally to decrease losses.”
2013 Insurance Predictive Modeling Survey by ISO and Earnix. I
summarize the results below. Insurers already hold vast amounts of data,
but now they can gather even more from new sources such as GPS-enabled
ジン・ライ教授による講演会
―10―
devices, social media postings and TV footage. The key to unlocking this
is through data analytics, the process of examining large amounts of data
of a variety of types to uncover hidden patterns, unknown correlations
and other useful information. Such information can provide
competitive advantage and result in business benefits, such as more
effective marketing and increased revenue.
The survey shows that Larger companies make more use of predictive
modeling than smaller ones. 82% of the respondents use predictive
modeling in one or more line of business. Models are more commonly used
in personal lines than in commercial lines. Specifically, models are more
commonly used in personal lines than in commercial lines. The most
common use of predictive analytics is found in Personal Auto (49%),
followed by Homeowners (37%), Commercial Auto (32%) and Commercial
Property (30%). The benefits of the use of predictive modeling include
profitability (85%), risk reduction (55%), revenue growth (52%), and
operational efficiency (39%).
The most common use of predictive analytics is in pricing. Loss cost
modeling accounts of 75%. The use of predictive modeling in
underwriting includes risk selection (78%), additional underwriting
information (52%). Some examples are stated below. Auto underwriting
uses Telemetry-based packages. In other words, actual driving
information is fed back to insurers’ system to predict likelihood of an
accident or car stolen. To deal with privacy issue, insurer can use “Grade
A” or “Grade B” instead of tracking particular roads. Customers don’t mind
giving up some data if you’re transparent about what data you’re asking for,
and they are getting real value back for it.”
生命保険論集第 200 号
―11―
Another Example is related life and health insurance. Apple Watch and
Fitbit activity trackers are used by John Hancock to offer users discounts
on their premiums and a free Fitbit wearable monitor. It should be noted
that life style is controllable and can be used to for premium underwriting.
Believe genetic disadvantage is not controllable and should not be used for
premium.
The use of redictive modeling in marketing area includes target marketing,
customer retention (44%), cross-selling (32%), and agent performance
(28%).
Specifically, algorithms comb through the unstructured data in telephone
calls, emails and the information divulged on social media networks about
what policyholders do or do not like to create personalized marketing
strategies for each customer.
Insurers also track policyholders behavior while they logged into their
insurer’s web portal – such as length of time browsing FAQ and help
sections, as well as participation in user forums and message boards, can be
recorded and built into a customer’s unique profile.
Met Auto and Home: Use CLUE (Comprehen Loss Underwriting
Exchange). Armed MetLife is able to pre-fill most of the demographic
information of the prospect who indicates interest in getting a quote, with
that consumer asked only to make changes if information is incorrect. Then,
once submitted, the company accesses the data bureaus via API and comes
back with a quote in a short time.
“At a minimum, it takes the process from 20 minutes down to two,”
For homeowners insurance, there’s a lot more that has to be collected, and a
ジン・ライ教授による講演会
―12―
lot more nuance. Yet, customers still demand a shorter process, and an
Xcelerate-style process might not be that far off.
“I know there’s a company trying to stitch together all the municipal
databases [that are needed for homeowners insurance] out there.”
A quarter of consumers surveyed by LIMRA and the Life Insurance
Foundation for Education (LIFE) think they need more life insurance, and
more than twice that many have no plans at all to purchase life insurance in
the near term. While cost is cited as the largest barrier to purchase,
significant portions of respondents to a 2014 survey said they felt they
“wouldn’t qualify for coverage” (19 percent) or simply “hadn’t gotten
around to it” (30 percent). The overall results show there is a market out
there.
With a long application and full medical exam waiting for those customers
that do show up, the industry is working hard on making a better initial
impression by leveraging emerging technologies.
A good example is Lincoln Financial which has moved away from asking
producers to walk customers through a full application
Now, the advisor collects some key demographic data, then ships that data
off to a call center where that data is used to call APIs at major data bureaus
and are correlated with the medical information that would’ve had to
confirm the information anyway – similar to the MetLife approach for auto
insurance.
The Timetric report cites South Africa-based insurer AllLife as using big
data to underwrite entirely new risk that could not previously be covered
profitably. AllLife offers life and disability insurance at low premiums for
生命保険論集第 200 号
―13―
manageable diseases such as HIV and diabetes. Using Big Data analytics,
AllLife assesses policyholders' risk every three to six months, and clients
who do not adhere to strict medical protocols will have their benefits or
cover reduced. As a result, AllLife aims to insure 300,000 HIV patients by
2016.
The use of predictive Modeling in Claims includes claim loss forecasting
(44%) and fraud (50%).
Fraud is a big issue for insurers. The Insurance Information Institute
(III) estimates that 10 percent of property-casualty insurance industry
losses each year are attributable to fraud, to the tune of $32 billion.
The III reports that 95 percent of insurers say they use antifraud
technology, but about half say a lack of information technology
resources prevents them from fully implementing it.
At the beginning of 2016, Towers Watson reported that 26 percent of
insurance companies were using predictive analytics to address fraud
potential. In the next two years, that number is expected to jump to 70
percent—more than any other big data application.
Many are looking to text mining as a crucial analytical tool for decoding
enormous amounts of unstructured data. Text mining is a way to scan
large amounts of data for keywords. Text mining can interpret claims
adjusters’ handwritten notes and scan a claimant’s social media accounts for
suspicious activity in nearly real time. Text mining and other tools
represent a fundamental shift.
The focus of fighting fraud has been changed from claims-centric to
person-centric. In other words, insurers will shift from focusing on the
ジン・ライ教授による講演会
―14―
claim itself to the individual filing the claim. Claims models will pull
information about the would-be beneficiary from across claims, policies and
external data sources, including information from other insurers, medical
professionals, police, auto body shops and a host of other sources.
This shift raises issues regarding privacy and data quality. Individuals may
opt out of sharing information with their insurers through vehicle telematics
or social media, lessening the impact of insurer data analytics initiatives and
creating a competitive advantage for less scrupulous organizations willing
to harvest data against consumers’ wishes. Compounding the issue is the
fact that data collected is not always accurate or easily manipulated.
Concerns about privacy and access to personal information are not
exclusive to the insurance industry. Encouraging insurance consumers to
share data by spelling out how the increased data benefits all parties. The
benefits of additional data for reducing claims fraud, and the subsequent
reduction in coverage costs, is a good place to start.
Insurers also use big date to predict claim losses and catastrophic events.
The severity and frequency of catastrophic events, both natural and
man-made, have been increasing over the past 20 years. Between 1990 and
2009, hurricanes and tropical storms accounted for 45.2% of total
catastrophe losses and the rate and intensity of these storms is predicted to
increase with global climate change.
Historically, the insurance sector has been good at developing catastrophic
models that capture known high severity/low frequency events (e.g.
earthquakes, tsunamis, etc.). However, most of these models perform poorly
when it comes to unknown ‘Black Swan’ events. Over the next decade the
insurance sector could be overwhelmed with uncorrelated catastrophic
生命保険論集第 200 号
―15―
events reducing capacity and raising prices.
Alternatively, new sensing and monitoring technology, together with risk
transfer mechanisms, could cushion insurers and reinsurers against
abnormal losses.
Who Use Big Data
Big Data is predominantly a big company affair at this point.
Of the companies with over $1B GWP, 51% either currently use Big Data
or are evaluating / implementing Big Data initiatives, compared to 30% of
the companies with under $1B GWP.
It should be noted that life insurance just getting started with big data.
Data sources that are already being used or explored include transformation
of business model (12%), expansion of customer relationships (41%),
enhancement of customer value proposition (18)%), and improvement of
internal performance management (29%) in 2015.
The internal data source for life insurers is from administrative system,
claim data, agents, underwriting data, and ZIP code data. The external is
from medical records, prescription data, credit scores, and motor vehicle
data.
Costs of Using Big Data
Companies spend considerable time on data preparation and deployment
before and after the actual modeling. Google spent over $3 billion on big
data infrastructure in 2012 and Facebook spent $1 billion.
The future data source is predicted from email, Clickstream, Voice-to-text
ジン・ライ教授による講演会
―16―
logs, credit scores, websites and social media.
Chief Data Officer (CDO)
Capital One, the Federal Reserve, Google, New York City and the U.S.
Army all have at least one thing in common: they each employ a chief data
officer to oversee their big data programs.
The emergence of the Chief Data Officer closely mirrors a trend we saw in
the 2000s when companies began to appoint CROs.
Barriers and Challenges will be discussed next. Three challenges include:
lack of sufficient data, lack of data scientists, and conflicting priorities.
The top three barriers life insurers must overcome are infrastructure
limitations, financial constraints, and lake of knowledge and/or expertise.
I would like to talk a little about data scientists. They are mathematicians,
computer scientists, academically trained scientists (astrophysicists,
ecologists, biologists, etc.), hackers and software engineers who use the
scientific method.
They ask a question, develop a hypothesis, collect data (often through an
experiment), analyze the results, report their findings and then often
engineer them into applications.
Chief Data Officer (CDO) will be discussed next. A CDO oversees a team
that handles all data analysis — from social media, marketing and
advertising to pricing, customer service and operational processes. CDOs
and their teams, unbound by departmental divisions, can become a
fountainhead of intelligence and solutions that aim to make every
department more effective.
生命保険論集第 200 号
―17―
But because data scientists may operate across departments and frequently
challenge the status quo, they often butt heads with company culture and
co-workers. For example, Greg Linden, formerly a data scientist at
Amazon, wanted to make shopping recommendations based on
items already in a shopper’s cart. As Linden wrote in his blog in 2006, “we
had an opportunity to personalize impulse buys.”
In the absence of a CDO, someone like Linden has to risk his job just to test
a hypothesis. When a CDO makes it clear that such experimentation is
expected and enrolls fellow executives in the opportunities, the CDO can
free data scientists to investigate issues they might otherwise be afraid or
unable to investigate. Ultimately, a CDO allows data scientists to operate
free from the constraints of company politics and HiPPOs (Highly Paid
Person’s Opinions).
In the absence of a CDO, someone like Linden has to risk his job just to test
a hypothesis. When a CDO makes it clear that such experimentation is
expected and enrolls fellow executives in the opportunities, the CDO can
free data scientists to investigate issues they might otherwise be afraid or
unable to investigate. Ultimately, a CDO allows data scientists to operate
free from the constraints of company politics and HiPPOs (Highly Paid
Person’s Opinions).
Creating a culture of seeking opportunities, finding and attracting the people
to do so, and creating a business that's executing effectively.
Asian universities can educate skilled modelers. Companies go all the
way to attract talents.
For example, a few years ago, a company moved a research unit to a
ジン・ライ教授による講演会
―18―
building across the street from a major university and engaged academic
department heads in technology, statistics and other areas where they
wanted to get a better draw of talent. They started offering research grants
to graduate students and created quite a bit of affinity that nurtured a flow of
fresh talent.
The Big Data Institute at Temple University has five centers with individual
specializations that include big data usage in mobile analytics, social media,
health sciences, oncology research, statistics and biomedical informatics.
These centers have used big data to connect brain imaging to successful
advertisements, to use technology to create vast amounts of DNA for
clinical study, and, in the School Tourism Hospitality and Management, to
decrease dissatisfaction in the leisure industry, among other research
projects.
THANK YOU!