AI and Big Data

Implications for
the Insurance
Industry in Canada

Emerging Issues
Research Report Series

Empowering insurance careers

insuranceinstitute.ca

i
 Emerging Issues Research Series

About the The Insurance Institute is the premier

source of professional education and
for-profit organization serving more than
40,000 members across Canada through 19
Insurance Institute career development for the country’s
property and casualty insurance industry.
volunteer-driven provincial Institutes and
chapters. For more information, please visit
Established in 1899, the Institute is a not- insuranceinstitute.ca.

About the CIP Society Since 1998, the CIP Society has represented
more than 18,000 graduates of the Insurance
the CIP and FCIP designations, continuous
professional development, professional
Institute’s Fellowship (FCIP) and Chartered ethics, mentoring, national leaderships
Insurance Professional (CIP) programs. As awards, and understanding of emerging issues
the professionals’ division of the Insurance in the industry. The CIP Society, on behalf
Institute of Canada, the Society’s mission is of its membership and for the benefit of the
to advance the education, experience, ethics industry, is proud to have contributed to the
and excellence of our members. The Society development of this research report. Please
provides a number of programs that promote visit insuranceinstitute.ca/cipsociety.

About the This research report represents the sixth in (2016), the sharing economy (2017), the
the Insurance Institute’s Emerging Issues changing workforce (2018), and climate

Institute’s research Research Series – providing relevant and risks (2020), as well as the demographics
insightful research reports on the issues of the P&C insurance industry in Canada
impacting the property & casualty insurance that the Institute has been conducting since
industry in Canada. The report joins reports 2007. For more information, please visit
on cyber risks (2015), automated vehicles insuranceinstitute.ca/research.

About the author of Paul Kovacs is a Senior Researcher with the

Insurance Institute and is Canada’s leading
the industry as the founder and Executive
Director of the Institute for Catastrophic Loss

this report authority on insurance and climate change.

For more than 20 years, Paul has been a
Reduction (ICLR) at Western University, and
the former President and Chief Executive
contributing author to the Intergovernmental Officer of PACICC. Paul has been a popular
Panel on Climate Change (IPCC), the world’s commentator on insurance, disaster safety
leading forum for the study of climate and economic policy, and has written more
issues. The Panel won the 2007 Nobel Peace than 200 publications and articles.
Prize. He is well known and respected in

ii
 Emerging Issues Research Series

Acknowledgements The Insurance Institute would like to thank the

following people who reviewed, commented
on and/or contributed to the report:

charles dugas, martha schrader,

associate partner, technology consultant, data analytics &
consulting, ey enterprise architecture
john juba, greg smith,
in-house lawyer and insurance chief operating officer, crawford &
industry executive company (canada) inc.
sven roehl, jonathan spinner,
executive vice president, canada, head avp, claims innovation, aviva canada
of innovation msg global, co-founder nikol kovacs & nancy kovacs,
cookhouse labs paul kovacs & associates
sachin rustagi, stephanie veltmann & emily debowski,
director, digital (innovation, bare brand
distribution & labs), gore
mutual insurance

Copyright ©2021 The Insurance Institute of Canada

All Rights Reserved

This research report is published in the public domain. When citing excerpts or the report in its entirety, please use the following
citation: AI and Big Data: Implications for the Insurance Industry in Canada, The Insurance Institute of Canada, 2021.

AI and Big Data: Implications for the Insurance Industry in Canada

insuranceinstitute.ca

iii
 Emerging Issues Research Series

A I and Big Data

Implications for
the Insurance
Industry in Canada

iv
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

Executive summary Digital technologies are reshaping the

world and will transform the way insurance
tools for brokers and claims professionals
working to support insurance customers.
products are designed, marketed, and Over time, the industry will be judged by how
distributed. One important element behind it manages the opportunities and pitfalls. In
AI and big data tools are expected the changes underway is the remarkable particular, how will the insurance industry
to reshape the insurance industry explosion in the ability to collect, analyze, address the needs of vulnerable consumers
and automate learning from very large that may be harmed by change? Industry
in Canada over the next 10 years. datasets. This report in the Emerging Issues actions will be judged relative to others in the
New approaches have the potential Research Series focuses on two factors driving insurance industry and to practices in other
to enhance service for consumers, change: artificial intelligence (AI) and big industries experiencing change. The industry
data analytics. Change began several years will need to tailor its approach to ensure that
improve the efficiency of business ago and has accelerated recently. AI and solutions are provided to address the needs
operations, and allow insurers to big data tools are expected to reshape the of insurance consumers, not just to reduce
help customers prevent losses. insurance industry in Canada over the next 10 industry expenses.
years. New approaches have the potential to
Insurance companies are on a spectrum
enhance service for consumers, improve the
today in terms of the application of AI and
efficiency of business operations, and allow
big data analytics. Some companies are on a
insurers to help customers prevent losses.
path toward widespread application of AI and
Change also brings risk—and large change, machine learning, where the performance of
like that associated with AI and big data models is measured, monitored, and adjusted
analytics, will bring large risk. Concern frequently. They are building teams of data
is expected to increase for consumers, scientists and other experts. An early focus
regulators, and others about explainability, for some involves the application of advanced
fairness, availability, and privacy. Greater analytics to assess the pros and cons of
capacity to target products and pricing will real-time, dynamic pricing in the Canadian
benefit many, but it will also disrupt long- insurance market. In contrast, others in
standing industry expectations for some. The the industry are beginning the journey by
speaking with vendors, implementing pilot
industry should anticipate and prepare for
modelling projects, and exploring options.
broad public interest in how these emerging

Change is inevitable. Digital interaction has become a norm in daily life

and will be increasingly extended throughout the insurance industry.
tools will be used to best serve the interests
As the insurance industry adapts to the digital
of all of society. The industry must establish
age there is scope to shift the practices and
the capacity to explain changes to individual
reputation of insurance to better support
consumers and verify fair treatment for
the risk management needs of society. The
stakeholders.
industry is criticized by some as too focused
Change is inevitable. Digital interaction on detecting and repairing losses. Change
has become a norm in daily life and will will be good for the insurance industry if AI
be increasingly extended throughout the and big data analytics are used to support
insurance industry. Machine learning, an industry focus on risk management
explainable AI, and big data analytics hold support for drivers, homeowners, and
great promise. The initial impact should be businesses, stressing the industry’s capacity
evident in improved efficiencies and better to predict and help consumers prevent loss.

v
 AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

It will also benefit brokers, adjusters, and paperwork. Importantly, change will benefit
other insurance professionals who are consumers who are offered personalized
provided with better tools to understand the coverage and better claims resolution. This
needs of consumers so they can spend more is the exciting potential if AI and big data
time using their experience and judgment to analytics are introduced in the Canadian
provide advice that will enhance the customer insurance industry with sufficient diligence,
experience and less time completing prudence, and care.

vi
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

Foreword The Insurance Institute is proud to publish

this sixth instalment in a series of reports on
issues that will impact the insurance industry
in Canada in the near and distant future.
emerging issues impacting the property and This report focuses on both the remarkable
casualty insurance industry in Canada. These opportunities to transform the industry
reports address important issues confronting through learning from enormous databases
the industry—cyber risks, automated vehicles, and the significant challenges expected as the
the sharing economy, the changing workforce, industry commits to ensure that the resulting
climate risks, and AI and big data. changes will benefit drivers, homeowners,
and business consumers.
The intention of this report is to provide
research of value to stakeholders. This We hope that this research report is not only
research report, and the series of reports in interesting and insightful, but also sets out
general, offers information and insights to the foundation for continued leadership by
enable insurance organizations to broaden the insurance industry as society works to
their understanding of how emerging risks address the challenges resulting from the
will impact the delivery of insurance products transition to a digital economy.
and services in Canada over the next five
to ten years. Sincerely,

This report provides a broad perspective peter hohman, fcip, mba, icd.d
on what is known about machine learning, president & ceo,
AI, and big data analytics and the potential insurance institute of canada

vii
 AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

Contents v Executive summary 20 Why are AI and big

data analytics expected
to transform the
vii Foreword insurance industry?

20 Insurance analytics
01 Introduction 23 Underwriting applications

24 Claims resolution
02 Six critical questions
27 Investment management
03 What is big data analytics,
and how will it support sound 28 Hiring and staff support
business decisions?
28 Marketing
04 Understanding big data
29 Lessons for the
04 Characteristics of big data insurance industry

05 Types of data 30 How can the insurance

industry best anticipate and
08 Sources of data
manage emerging risks?
09 Data storage
30 Explainability
10 Analytics and data science and transparency

11 Lessons for the 33 Fairness and bias

insurance industry
34 Availability
12 What is artificial intelligence
34 Privacy and security
and how will it support sound
business decisions? 35 Lessons for the
insurance industry
13 Understanding
artificial intelligence

15 Machine learning and

deep learning

15 AI will be applied
across society

18 Augmentation
versus automation

19 Lessons for the

insurance industry

viii
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

Contents 36 Six critical questions (cont.) 45 Why must insurers focus

on better outcomes
36 Why should insurers
for consumers?
expect increased attention
from regulators? 45 A proud history

37 Prudential regulation 46 Classification of risks

38 Market conduct regulation 48 Consumers’

views of fairness
39 Auto rate regulation
50 Lessons for the
40 International guidance for
insurance industry
big data analytics

40 International guidance for

artificial intelligence 51 Recommendations for the
43 Regulatory arbitrage
insurance industry
43 Options for
self-regulation 55 Appendix: AI terminology
44 Lessons for the
insurance industry 56 Bibliography

ix
 Introduction Insurance is changing. Explainable AI and
big data analytics are driving forces that will
and explain why the new systems result
in changes in the treatment of consumers.
increasingly be applied across the insurance Regulators will closely monitor the impact
industry. There is excitement about the of introducing AI and big data analytics into
“Any sufficiently advanced potential to better serve the risk management the insurance industry. The new systems
technology is indistinguishable needs of consumers. The industry will have will be assessed by prudential regulators in
greater capacity to identify and manage risk terms of the risk to solvency, governance
from magic.” at a granular level. This will be evident in oversight, and reputational risk. The banking
coverage increasingly tailored to the specific industry is developing a capacity for third-
—arthur c. clarke needs of individual consumers, reduced party verification that decision support
prices for those with low risk, a better claims algorithms are statistically fair. This issue
response for those experiencing a loss, and will likely become increasingly important
the capacity to prevent losses. Work in the for the insurance industry. Market conduct
industry will become more satisfying as the regulators have been pressing the insurance

AI and big data analytics will accelerate innovation in Canada’s insurance

industry, an industry typically viewed as stable and unchanging.

automation of tedious tasks allows brokers, industry to demonstrate fair treatment of

agents, adjusters, and others to focus on more consumers in underwriting practices and
rewarding activities that serve consumers. claims management, a conversation that
Change is inevitable as insurers adopt a digital will increase with emerging AI and big data
mindset, and it will be positive for most. tools. It is the responsibility of regulators to
understand and, when required, challenge
The expectation of significant change will be
change in insurance practices to ensure fair
accompanied by significant risk. If the new
treatment of consumers.
tools demonstrate that some members of a risk
pool should pay less, then others will be asked AI and big data analytics will accelerate
to pay more, or they may find that coverage is innovation in Canada’s insurance industry,
no longer available. Many early AI algorithms an industry typically viewed as stable and
and big data tools provide outcomes unchanging. The pace of innovation will
that cannot be explained to others in the depend on the industry’s ability to explain the
industry, regulators, and, most importantly, changes to consumers, regulators, and other
to consumers. The industry should expect stakeholders. The process of digitization is
that some changes will be challenged as unfair, proving to be more complicated for some than
introducing regulatory and reputational expected. Nevertheless, change is inevitable,
risk. The industry will need to explain even if the timing is uncertain. Significant
the implications of the changes to each benefits for consumers are possible if the
consumer. Significant effort will be required introduction of AI and big data analytics is
throughout the industry to understand managed responsibly.

1
 Emerging Issues Research Series

Six critical
questions

2
 Six critical questions The digital revolution is the shift from
mechanical and analogue systems to a
around the world are the direct result of
the growing application of digital tools.
What is big data analytics, and society reliant almost exclusively on digital For many years, insurance companies
how will it support sound business systems. This transition began in the 1960s conducted rigorous actuarial analysis of
and accelerated with the growing popularity recent loss experience to anticipate future
decisions? of the Internet, personal computers, and claims. This was conducted quarterly or
smartphones. Mass production of integrated annually using established databases and
circuit chips was accompanied by the capacity analytical tools. Some other industries,
“Never interrupt someone doing to produce chips with higher transistor density however, began to assemble and assess
what you said couldn’t be done.” and lower manufacturing costs. Indeed, extremely large databases that could not be
since 1970, it has been possible to double assessed using commonly available tools.
the number of transistors on an integrated This included leading technology companies,
—amelia earhart
Remarkable and persistent improvement in the capacity to store and
process extremely large datasets ultimately has benefited all of society.
circuit every two years, an extraordinary rate banks, health care providers, meteorologists,
of growth sustained over 50 years.1 The shift and some government agencies. Use of cloud
to a digital world marked the beginning of storage and computing, supercomputers, and
the Information Age. Data, information, and new software facilitated approaches to the
ultimately knowledge emerged as critical analysis of large datasets—big data analytics.
business assets. In contrast, machine learning and AI do not
necessarily require large databases.
In a digital economy it became easier to
access large volumes of data, improve Remarkable and persistent improvement in
communications, and reconfigure production the capacity to store and process extremely
processes. The digital revolution set the large datasets ultimately has benefited all of
stage for the automation of industrial and society. These tools are now widely available,
manufacturing processes, robotics, the affordable, and usable across most industries.
Internet of Things, self-driving vehicles, cloud Leading insurers now employ teams of
computing, big data analytics, and artificial data scientists to assess datasets that were
intelligence. These changes are transforming unimaginable 10 years ago, and it is expected
society, a process that is described by that these tools will be in use across the
some as the Fourth Industrial Revolution.2 insurance industry over the next 10 years.
Rising productivity and living standards

1 This is known as Moore’s Law, named after Gordon Moore, who made this
observation in 1965.
2 The term was introduced at the 2016 World Economic Forum in Davos, Switzerland, by WEF
founder and chief executive Klaus Schwab.

3
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

Six Critical
Big data is the present and the Understanding big data the world has increased tenfold over the
last seven years and is expected to increase
future foundation for business and The term “big data” began to be used in a further tenfold over the next seven years.4
society. the 1990s to describe the immense, often
This is a remarkable increase of 100-fold
complex stores of data that were too large to

Questions
over 14 years. The explosion in the ability
be processed using software tools commonly
to store data was accompanied by a similar
available at the time. In particular, early data
increase in capability to access and analyze
warehouses were used to store and manage
this information, known as big data analytics
large structured datasets generated by the
or data science.
companies, and software was developed to
process the data. These tools were initially The big data revolution is in full swing. The
introduced in a few industries, such as cloud has joined data warehouses to make
banking, health care, and the public sector. storage and accessibility even more efficient.
The important change with the transition to The sources of new data are seemingly
big data involved collecting data from beyond endless, as people put more of their lives
the enterprise. Also, new data warehouses and and transactions online, creating digital
software tools were extended to address both ecosystems with third parties who can
structured and unstructured data. provide access to the new data. Big data is
the present and the future foundation for
Since the 1980s, it has been estimated that
business and society.
the global capacity to store data doubled
every 40 months.3 The total data stored in

Characteristics of big data

The three “Vs” of big data set out in 2001 are volume, velocity and variety:5
• Volume relates to the size of the datasets. Billions or even trillions of
records are flowing from social media, smart devices, videos, business
transactions, industrial equipment, and other streams.
• Velocity refers to the rate at which new information is being
generated and the speed necessary for it to be processed for timely
insights. Increasingly this takes place in real time.
• Variety indicates the diversity of information sources that make up big
data. This includes alphanumeric data provided by customers, audio
and video files, emails, financial transaction data, and text documents.

3 Hilbert and Lopez, “The World’s Technological Capacity to Store, Communicate, and
Compute Information.”
4 Ibid.
5 Laney, “3D Data Management.”

4
 Six Critical Questions Emerging Issues Research Series

Four additional concepts that have emerged involve variability, veracity,

visualization and value:6
• Variability cautions that differences in perception can give a different
meaning to the same dataset, an issue that can only be solved
through an understanding of context.
• Veracity, reliability, accuracy, and quality of the information used in
big data analytics is essential to support sound decisions. Quality
information supports sound decisions.
• Visualization of data through charts and other media makes
information more accessible and understandable for stakeholders.
• Value provides a measure of the financial return from investing in
data management and analysis.

The evolving characteristics are an indication Types of data

of the changes in complexity of big data over
the last 20 years. Moreover, the issue of Data come from many sources. It is essential
veracity is critical for the insurance industry. that business decisions are based on reliable
Insurance is founded on trust. Management data. The insurance industry, for example, is
of these issues will affect the reputation using traditional and non-traditional sources
of the industry. Insurers require quality of acquired data about consumers. Initially,
data to make sound decisions, conduct consumers provided information directly.
risk assessments, and develop models. For Traditional sources for insurance include
example, data may be subject to adjudication demographic, medical, exposure (such
when used to resolve claims, so it is important as type of car, value of contents, type and
to document where data were acquired and features of dwellings), behavioural, loss, and
to track its veracity. hazard information. New and non-traditional
sources include the Internet of Things (IoT),
online media (web searches, online buying,
and social media), and personal preference
data (such as cell phone operating system
and email addresses).7

6 IT Chronicles, “What Is Big Data?”

7 IAIS, “Issues Paper on the Use of Big Data Analytics in Insurance.”

5
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

Type of data Examples Personal / non-personal Use Data source

Panel A: Traditional data
Demographic data Age, gender, civil and family Personal Risk Selection Policyholders
status, profession, address
Medical data Medical history, medical Personal Risk selection Policyholders
condition, condition of family
members, genetic testing
Exposure data Type of car, value of Personal / non-personal Risk selection Policyholders
building contents, type of
features of dwellings
Behavioral data Smoking, drinking behavior, Personal / non-personal Risk selection, Policyholders,
distance driven in a year, marketing industry statistics
deductible choice, life
insurance lapse rates

Loss data Claim reports from car Personal / non-personal Claims management Policyholders,
accidents, liability cases information
exchange
within industry
Population data Mortality rates, morbidity rates, Anonymised and aggregated Risk selection Government,
car accidents personal data industry
statistics, academia
Hazard data Frequency and severity of Non-personal Risk selection Government,
natural hazards industry
statistics, academia
Other Credit reference, claim Personal / non-personal Risk selection, Policyholders, credit
traditional data adjustment reports, information marketing, agents, partner
from the auto repair shops claims management adjusters or agencies
involved in the claim

Panel B: New and ‘non-traditional’ data

IoT data Driving behavior (telematics), Personal Risk selection, Data
physical activity and medical claims management collection devices
condition (wearables),
surveillance (smart home)
Online media data Web searches, online buying Personal Risk selection, Technology
behavior, social media activities marketing companies (Internet
providers, search
engine providers,
e-commerce
providers, social
media platforms)
Personal Mobile phone operating system, Personal Risk selection, Technology
preference data email addresses marketing, companies,
product development Policyholders

Source: Geneva Association, “Big Data and Insurance: Implications for Innovation, Competition and Privacy,” page 36.

6
 Six Critical Questions Emerging Issues Research Series

There are a number of sources of data:

Acquired data
• “The automatic collection of data from sensors and readers in a
factory, laboratory, medical, or scientific environment.”
• “The gathering of source data for data entry into the computer.”8
• Acquired data can include both traditional and new, non-
traditional data.
Imputed data
• “The substitution of estimated values for missing or inconsistent
data items (fields). The substituted values are intended to create a
data record that does not fail edits.”9

Insurers can purchase data about crime, perhaps with some feedback from insurers
vehicles, and dwellings. Customers can summarizing findings. Consumers typically
upload a photo of a damaged vehicle. are asked to approve acquisition of data from
Telematics can provide information about third parties, but the details are often unclear
driving practices. Connected devices can and there may be no recourse to consumers
provide real-time information about water for verification. Special issues may arise when
damage or unauthorized entry. A remarkable data are derived from other sources, like
array of new data are available. credit scores or the aggregation of telematics
data by third parties.
Insurers have established processes to verify
the accuracy of data provided directly by Acquired data used to identify and approach
consumers. This includes assessing driving potential consumers is an approach used by
records with the police or a check of the all industries, including insurance. Businesses
property using a search of real estate records. are aware that most leads are incomplete,
Policyholders are aware of the information out of date, or otherwise unlikely to result

Special issues may arise when data are derived from other sources,
like credit scores or the aggregation of telematics data by third parties.
provided and have an opportunity to ensure in securing new customers. Nevertheless, this
that the information is complete and correct. approach remains important to market goods
There is an expectation of utmost good faith and services to potential new customers.
in the accuracy of the information provided Supporting data are sufficient to provide some
by customers and disclosure of the coverage value, and occasionally they are very helpful.
provided by insurers. The core foundation of Businesses that rely on approaching potential
insurance is trust. consumers are increasingly investing time
and effort in determining the value of the
Challenges may arise in the verification
data available to support this effort.
of acquired consumer information that
does not come directly from policyholders. Acquired consumer data are also used by the
Consumers can choose to directly share insurance industry to support resolution of
data from telematics and other devices, claims and underwriting decisions.

8 Both definitions are found at https://www.yourdictionary.com/data-acquisition.

9 Definition found at https://stats.oecd.org/glossary/detail.asp?ID=3406.

7
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

Insurers are responsible for ensuring that actual facts, but imputed data always has
these data are accurate, even if it was not a random factor. Consumer and regulator
acquired directly, given that it is used to expectations about the use of imputed data
support appropriate decisions. For example, in the insurance industry is unclear and will
analysis of policyholder postings to social evolve over time. Insurance is built on trust,
media may suggest that a claim should be so it is critical that the insurance industry
denied, and industry best practices to ensure understands how trust can be affected by
fair treatment of consumers are increasingly changes in the data used to support decisions.
documented. Regulation and court rulings will Techniques used to impute data will likely be
provide direction about acceptable practices. subject to a broader assessment of the fair

Insurance is built on trust, so it is critical that the insurance industry

understands how trust can be affected by changes in the data used to
support decisions.

Consumer and regulator expectations of the treatment of consumers. As insurers advance

insurance industry when using acquired data in their digital journey, the line between
for claims resolution and underwriting is marketing and underwriting will blend. If
much higher than data used for marketing. the approaches used cannot be explained to
As social media tracking becomes increasingly regulators, and perhaps directly to consumers,
automated, the responsibility to collect and then insurers should question the value of
verify the data used will increase. using imputed data.

Even with many sources to draw on, the data Sources of data
are often incomplete, and businesses will
apply analytical tools to fill gaps whenever Swiss Re reports that insurance is a low touch
possible. It may be necessary to fill in gaps industry. A study in the United States found
by estimating values based on similar cases that customers interact with their insurer, on
or providing an estimate that is consistent average, only 2.7 times each year. Insurers
with the person’s existing data. The aim investing in big data analytics are likely to
is to predict as closely as possible to the acquire information from many sources:

Owned touchpoints: The insurer invests and controls these consumer

touchpoints. They include agents, websites, and apps. The goal is long-
term relationship building; the benefits are cost efficiency and control.
Paid touchpoints: The insurer pays for the digital touchpoint to reach the
consumer, such as through paid searches and sponsorship. Paid touchpoints
can include various channels, such as bancassurance and agents.
Earned touchpoints: Here the consumer becomes the touchpoint.
This is when satisfied consumers become social media influencers and
spread awareness.
Social touchpoints: An insurer interacts with third-party channels but uses
its own profile, such as on Facebook or Twitter. Social media is forecast to
be the fastest-growing digital advertising channel over the next five years.

8
 Six Critical Questions Emerging Issues Research Series

Changing sources of data introduces up and running at all times. The cost of using
issues about ownership. Insurers must cloud storage is low and can be controlled,
determine if and how information because the facilities often charge only
collected from social media, for example, for the space used. Cloud hosts may also
can be used to resolve a disputed claim. provide processing and analytics. Each
Who owns images of a property surveyed company must establish a balance between
by a drone—the homeowner or the insurer? on-site storage and the use of the cloud.
Consumers, regulators, and other This includes consideration of the need to
stakeholders will be interested in the sources protect confidential information provided
of data used by insurance companies, as by consumers.
well as how this information will be used,
Big data, because of its vastness and
verified, and protected. Policyholders will
complexity, requires the right tools to process
be divided in their views about the use of
and analyze it. Natural language processing
some data sources, and insurers will need
helps to draw value from unstructured text,
to win consumer confidence supporting the
like emails, chat logs, and other textual data.
practices they adopt.

Degree of difficulty incorporating new data sources into processes

New data sources

Electronic health records
Now easier to obtain (EHRs), third party data bases,
digitally-traditional data photos, images, voice, data
Easily obtained-traditional Tax filings, property from wearables & Iot devices,
data registrations, births & deaths, behaviourial insights, reviews
Data via application, crime records, government data
FNOL, medical info, motor
vehicle records

Predictive power Established Partially established Exploratory phase

Integration challenges Addressed Being addressed Less known

Acquisition cost Low Stabalising, was high Still expensive

Regulatory barriers Minimal Moderate High

Source: Swiss Re Institute, sigma No 5/2020, page 17.

Data storage Chatbots on an insurer’s website can ease

Big data only became possible as data storage the agent’s workload by eliciting information
capacity expanded in relation to its physical from consumers for use in formulating
size. A business would purchase a few servers policies or settling claims, leaving brokers,
on-site, along with staff to run them, and agents, and adjusters free to focus on tasks
make them accessible to the appropriate that are best served by human empathy and
staff. Cloud storage, by contrast, stores data understanding.
on a large number of servers connected
The big data needs of the insurance industry
digitally but not necessarily all in one place.
require a variety of approaches to managing
Cloud storage is provided by a hosting large data banks appropriate for the needs of
company that services a number of businesses. its consumers. Finding the right fit is essential
The host is responsible for maintaining the to optimizing a company’s use of data.
storage facilities and for ensuring that it is The skills and investments here cannot

9
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

The big data needs of the insurance be understated. Typically, this is not an deeper to identify correlations and causal
incremental build, but requires a new set of links between pieces of data quickly and
industry require a variety of
people and practices. efficiently. Analytics provides the business
approaches to managing large with a view of where it can improve, what
data banks appropriate for the trends to be aware of, how to better serve
Analytics and data science their customers, and what decisions they
needs of its consumers.
Analysis traditionally focused on might make to stay ahead of the competition.
understanding what happened in the past. Data science is a term that has recently
Analytics seeks to understand why something become popular, but it does not yet have a
happened to anticipate what will happen in consistent definition. Advanced statistical
the future. Actuarial science, for example, analysis, or management science, was
uses analytics to predict the risk of loss initially applied by the military with logistics
in the future using predictive models and optimization. Some describe data science
other tools. Data science is an emerging as an applied field of statistics, typically
profession largely focused on working with with a focus on the analysis of big datasets.
big data analytics. Others argue that data science is a new,
Internal and acquired datasets are, by nature, interdisciplinary field using concepts from
unstructured. The data must be put into a mathematics, statistics, and computer
manageable form to ensure quality and science. The objective of data science
be given a coherent structure to support professionals is to extract insights from large
analysis. When the data are ready, they can be sets of structured and unstructured data. Data

Analytics provides the business with a view of where it can improve,

what trends to be aware of, how to better serve their customers, and
what decisions they might make to stay ahead of the competition.

analyzed for insights that can help a business science includes analytic approaches like data
understand its consumers, visualize its mining and machine learning.
direction, look for trends, and make decisions Expert input is required to prevent spurious
for its future. and suspect correlation. The OECD observes
Data processing organizes information that “only through expert analysis can
into usable, visual formats, like lists, charts, causation be well established.” Building
10

graphs, and other formats useful to support a model is easy—understanding how the
analysis and decisions. Data analytics digs model can support better decisions is where
expertise is required.

The Organisation for Economic Co-operation and Development (OECD)

identifies four types of data analytics:11
1. Descriptive analytics: Analysis of what happened using past data
2. Diagnostic analytics: Assessing data to hypothesize why
something happened
3. Predictive analytics: What is likely to happen in the near future
4. Prescriptive analytics: Suggesting a course of action
going forward
10 OECD, “The Impact of Big Data and Artificial Intelligence (AI) in the Insurance Sector.”
11 Ibid.

10
 Six Critical Questions Emerging Issues Research Series

Lessons for the insurance industry better decisions about claims, pricing, and
operations—the foundation for sustaining
Some industries have been using big data
healthy earnings.
analytics for many years. Big data analytics
and data science are the future of global Early commitment to big data analytics varies
interactions and transactions. With constant considerably across the Canadian insurance
improvement in the algorithms, the growing industry. Some companies, large and small,
dependence on smart technology, and the are recognized leaders within Canada and on
ever-increasing number of companies the world stage. Others are just beginning
providing big data analytics services, these the journey and have yet to establish a clear
tools will become entrenched in every plan. These emerging tools will be important
aspect of life for business, government, across the industry, so they must be assessed
and individuals. by insurance companies, brokers, agents,
and adjusters. Remarkable change evident
Analytic approaches used in the Canadian
in other industries appears inevitable for the
insurance industry will naturally expand to
insurance industry, but the pace of change
embrace the new tools offered by big data
is uncertain. Each organization will need
analytics. Some insurers in Canada have
to determine the specific elements that are
already gone all in on big data analytics.
appropriate for their needs. The insurance
Analytics will grow and improve, along
industry will be transformed by this transition
with the tools to drive its use, including
over the next 10 years.
data science. Big data analytics will support

11
Six critical questions Artificial Intelligence (AI) is being applied
in organizations across the financial sector.
systems provide benefits that include faster
execution of trades to secure the best prices,
What is artificial intelligence It involves the use of computer algorithms more frequent checks to ensure the portfolio
and how will it support sound to simulate and augment human intelligence. complies with established investment
Modern statistical methods are applied guidance, and reduced risk of errors due to
business decisions? to analyze datasets in an effort to find the psychological or emotional causes.
optimal solutions to specific questions. The
With some exceptions, the insurance industry
objective of AI is, ultimately, that it will be
has been late to adopt machine learning and
“Beware of false knowledge; it is able to “think for itself,” imitating human
AI. Nevertheless, new systems are expected
more dangerous than ignorance.” reasoning. Narrow AI is used to solve specific
to be developed over the next 10 years across
issues. General AI works to resolve multiple the insurance industry to support companies,
problems at the same time. That goal is in the brokers, and adjusters. Some insurance
— george bernard shaw future. At present, AI is being used to train companies in Canada, large and small, have
machines to support decision makers through been vocal about actions underway that
the assessment of large volumes of data and demonstrate their commitment to a digital
look for patterns through the study of human future. The Centre for Study of Insurance
decision making. Operations is a leader in describing how the
These systems seek to reduce the risks insurance industry and independent brokers
associated with human error in the entry can best use AI to better serve customers.
13

and analysis of critical consumer information. There is an expectation that chatbots and
The banking industry and some insurance other new tools will be developed to better
companies are developing decision support support communications with consumers
models to increase cross-selling of products when a claim is filed or a customer is

With some exceptions, the insurance industry has been late to adopt
machine learning and AI.

to existing customers. This can include the exploring coverage alternatives. Most early
development of digital campaigns and virtual development appears to address consumer-
marketing teams that support the efforts of facing activities. The number and capacity
relationship managers to target new offerings of chatbots and other AI-based tools will
and promptly anticipate customer needs.12 increase. AI is also expected to improve
Fraud detection is a further benefit, as internal industry operating systems, including
systems are trained to identify suspicious workflow optimization and stronger support
behaviour across very large datasets, for communication between brokers and
supporting early and aggressive corrective insurance and reinsurance companies, and
action. This will be used in the real-time between adjusters and companies. Moreover,
identification of organized criminal activity, AI tools and models will be developed by data
money laundering, credit card theft, and scientists to improve insurance underwriting
other financial fraud. Investment managers and develop coverage for previously
are using algorithmic trading programs. These uninsurable risks.

12 Microsoft, “Empowering the Digital Insurer.”

13 Smola, “How Artificial Intelligence Is Transforming the P&C Insurance Industry.”

12
 Six Critical Questions Emerging Issues Research Series

Importantly, AI looks to harness datasets to

Modern AI tools use computer algorithms to
develop tools that will better anticipate analyze and learn from datasets. These systems,
and serve the needs of consumers. These developed by software engineers, adjust and
tools also have the potential to automate adapt their learning as new data come in.
many tedious tasks, freeing workers to The algorithms do not need the data to be
focus on the more productive and satisfyingstructured. The systems are capable of acting
efforts of serving consumers on a personal on what they have learned to solve problems,
level. Responsible application of these newreach conclusions, predict behaviour, operate
technologies holds great promise to reduce machinery, and make recommendations. As
business costs, make work more satisfying, new data are added to the existing dataset,
and better serve customers across most the algorithms work to refine and improve
industries in Canada. their capacity to replicate or perhaps exceed
the capacity of human decision makers.
Examples of how AI works in everyday life
Understanding artificial are seen in our interactions with Siri, Alexa,
intelligence and Google. Automated vehicles depend
Artificial intelligence involves training on AI to observe and adjust in real time to
machines to make decisions and support changing road conditions. Presently, AI is
those who make decisions, in contrast to very good within specific areas. The aim
natural intelligence, where decisions are for the future is to continue to increase its
made by humans or animals. The objective ability to handle more complex situations
of AI is to supplement or emulate human with greater accuracy and learn to behave in
deduction, reasoning, and problem solving. a more human-like way. Complex issues can
Development of machines that can “think” be addressed by more complex architecture
has been an academic pursuit since the if there are data. A further challenge involves
1950s, but major breakthroughs required moving from narrow AI (addressing a single
the co-development of large databases and issue) to general AI (addressing many issues).
analytical capacity.

MI development timelines

1888 Francis Galton: 1982 John Hopfield: 1995 Corinna Cortes 2011 IBM Watson
correlation, standa rd, Recurrent neural & Vladimir Vapnik: wins Jeopa rdy
deviation, regression network Support vector
machines

1763 Thomas Bayes: 1950 Alan Turing: 1970 Seppo 1980 Vector 1995 Tin Kam Ho: 2012 And rew Ng &
underpinnings of Learning machine Linnainmaa: Autoregression (Sims) Random Forest Jeff Dean: Deep
Bayes theo rem Backpropogation algorithm learning

1900 Pearson: 1987 Cointegration 2016 Google’s

statistical hypothesis 1972 Haza rd models Alpha Go beats
testing human Go champion

1812 Pierre-Simon 1957 Frank 1972 Generalized 1997 Hoch reiter &
Laplace: Bayes Rosenblatt: linear models Schmidhube r: Long
theorem Perception short-term memory
(LSTM)

1944 Logit model 1970 ARI MA

Visual guidance and

First neural net and learning programs Speech recognition Deep learning Data mining, Big data
logic based learning

Pre 20th Century 1950s 1970s 1980s 1990s 2000 2015 2020

Pre AI era Expert Algorithmic Big data, better compute,

Genesis of AI and early hype AI winter AI winter AI winter
systems innovation algorithmic efficiency

Source: Swiss Re Institute, sigma No 5/2020, page 5.

13
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

Topic areas within artificial intelligence (non-exhaustive)

Natural Sensors / Audio /

Knowledge internet
language speech
representation of things
generation analytics
Natural Graph Simulation
language analysis modeling
processing

Machine Deep Q&A Image

Robotics analytics
learning systems

Social
Deep network Soft
learning analysis robotics

Virtual Machine
Visualization personal Recommender
systems translation
assistants

Source: PWC
Source: PwC, “AI in Insurance: Hype or reality?,” page 4.

The pioneers who developed deep learning are AI is currently used to compete at the highest
Yoshua Bengio at the University of Montreal, levels in strategic games, conduct complex
Geoffrey Hinton at the University of Toronto, military simulations, and operate automated
and Yann LeCun at New York University. In vehicles. Indeed, leading artificial intelligence
2018 they were awarded the Turing Award and systems have defeated the highest-rated
$1 million for the conceptual and engineering competitors in games that include chess, go,
breakthroughs. Jeff Dean of Google AI said, and Jeopardy. Chatbots or virtual assistants
“Deep neural networks are responsible for have developed an extraordinary capacity to
some of the greatest advances in modern understand human speech and participate in
computer science, helping make substantial a conversation, answering questions quickly
progress on long-standing problems in and efficiently. These natural language and
computer vision, speech recognition, and machine learning tools can be designed to
natural language understanding. At the heart provide customer service, marketing, and
of this progress are fundamental techniques sales support. The apparent intelligence
developed starting more than 30 years ago of these tools comes from access to large
by this year’s Turing Award winners, Yoshua datasets containing more information than
Bengio, Geoffrey Hinton, and Yann LeCun. can be retained by human decision makers.
By dramatically improving the ability of Some people prefer interacting with these
computers to make sense of the world, deep systems relative to their experience with
neural networks are changing not just the field some human representatives.
of computing, but nearly every field of science
and human endeavor.”14

14 ACM, “Fathers of the Deep Learning Revolution Receive ACM A.M. Turing Award.”

14
 Six Critical Questions Emerging Issues Research Series

Machine learning and deep learning If the sample of human decisions is

biased, the algorithms will repeat these
Machine learning is a subset of AI. One errors. Teaching artificial intelligence systems
element, supervised learning, focuses on to support decision making will always be
performing a particular task and searches for constrained by the quality of data and the
patterns that allow the machine to improve design of the initiative.
its performance when measured against a
database of human decisions. Unsupervised AI will be applied across society
learning models are not told what to look
for. Deep learning also supports speech Machine learning and AI tools are being
recognition used in chatbots and other developed for use across society. Tools are
in place, for example, so consumers can
applications.
make an appointment or book a reservation
Deep learning, an evolution of machine by speaking to a machine. Natural language
learning, delves into vast stores of data, tools support conversations with customers
performing a task over and over again, seeking an appointment with their dentist
“thinking” its way through and constantly or a reservation at a local restaurant. The
improving the output measured against a COVID-19 global pandemic accelerated
dataset of human decisions. Deep learning is testing that was already underway,

Deep learning, an evolution of machine learning, delves into vast stores

of data, performing a task over and over again, “thinking” its way through
and constantly improving the output measured against a dataset of
human decisions.

essential in speech and language recognition, resulting in early implementation of

translation, facial recognition, automated systems by a broad range of users.
vehicle cameras, chatbots, and online AI pioneers, like Amazon, found that these
shopping. Deep learning is responsible for tools provided a competitive advantage
the product and video recommendations that through a period of great uncertainty
pop up on computers and mobile screens. The for consumers. Systems are available for
more deep learning algorithms are used, the businesses in a broad range of circumstances,
better they perform. with increasing simplicity to implement,
declining cost to acquire, and growing
Both machine learning and deep learning are acceptance by consumers. Breakthroughs
essential to the field of artificial intelligence.
in online payment capacity and security
Although neither type of learning is capable accompanied the increasing availability
of the human thought processes that AI aims of AI tools, further speeding up consumer
to duplicate, their ability to dig into the data acceptance of new systems.
has made possible the advances seen in
recent years. The military has developed sophisticated
training tools that simulate an enemy attack.
Concerns sometimes arise about AI systems Meteorologists have significantly improved
that inadvertently learn to make biased the accuracy, narrowed the geographic
decisions that are unfair to certain customers. focus, and extended the duration of weather
These algorithms learn to make decisions or forecasts using these tools. Amazon and
provide advice from the examples of previous Google train their systems to learn about
human decisions provided by the data. individual customers and anticipate their
future needs.

15
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

Artificial intelligence can deliver industry expectations through machine learning and deep learning.

Artificial intelligence
The science and engineering of Machine learning
making intelligent machines A major approach to realize Deep learning
artificial intelligence A branch of machine learning

1950s 1960s 1970s 1980s 1990s 2000s 2010s

Artificial intelligence (AI) Machine learning (ML) Deep learning (DL)

Intelligence exhibited by machines, Major approach to realizing AI by Branch of ML where algorithms attempt
whereby machines mimic cognitive learning from, and making data-driven to model high-level abstractions in data.
functions associated with human minds; predictions based on, data and learned DL connects artificial, software-based
cognitive funcitons include all aspects of experiences. ML comprises several calculators that approximate the function
learning, perceiving, problem solving, categories including reinforcement of brain neurons. Neural networks,
and reasoning. learning, supervised learning, and formed by these calculators, receive,
unsupervised learning. analyze, and determine inputs and are
informed in determination is correct.

Source: McKinsey & Company, "Insurance 2030–The impact of AI on the future of insurance," based on "The history of artificial intelligence,"
blog entry by Rockwell Anyoha, sitn.hms.harvard.edu.

The OECD observed that “when AI or making process of the algorithm becomes
machine learning is being applied for important and the fact that it is a black box
business purposes, such as online retail and causes concerns.” 15
recommender systems, the outcome would
There should be transparency and
be to show products, social media posts and
responsible disclosure around AI systems
search results, which would require accuracy
to ensure that people understand AI-based
of data, but may not lead to considerations of
outcomes and can challenge them.
an ethical nature or on its criticality.
When an AI decision has a large stake,
such as a financial decision…the decision

15 OECD, “The Impact of Big Data and Artificial Intelligence (AI) in the Insurance Sector.”

16
 Six Critical Questions Emerging Issues Research Series

Leading voice assistants, including Amazon’s Alexa (named for

the ancient library in Alexandria), Apple’s Siri (meaning ‘beautiful
woman who leads you to victory’ in Norse) and Google Assistant,
are female by default.
A recent United Nations study has highlighted that female-voice AI
reinforces harmful gender biases and has made recommendations
to limit the practice of making digital assistants female by default
and most importantly, “to develop the advanced technical skills
of women and girls so they can steer the creation of frontier
technologies alongside men.”16

The OECD identified five principles for the responsible stewardship of

trustworthy AI:
1. AI should benefit people and the planet by driving inclusive growth,
sustainable development, and well-being.
2. AI systems should be designed in a way that respects the rule of
law, human rights, democratic values, and diversity, and they should
include appropriate safeguards—for example, enabling human
intervention where necessary—to ensure a fair and just society.
3. There should be transparency and responsible disclosure around AI
systems to ensure that people understand AI-based outcomes and
can challenge them.
4. AI systems must function in a robust, secure, and safe way
throughout their life cycles, and potential risks should be continually
assessed and managed.
5. Organizations and individuals developing, deploying, and operating
AI systems should be held accountable for their proper functioning
in line with the above principles.17

16 UNESCO, “I’d blush if I could: closing gender divides in digital skills through education.”
17 OECD, “The Impact of Big Data and Artificial Intelligence (AI) in the Insurance Sector.”

17
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

AI is not expected to replace A widely cited 2018 paper by Joy Buolamwini, a researcher at
human workers in the insurance the MIT Media Lab, and Timnit Gebru, a member at Microsoft
industry, as has been evident Research, found that the error rates for three commonly used
in industries that include commercial facial recognition systems (Microsoft, IBM, Face++)
agriculture and manufacturing. were much higher when identifying darker-skinned individuals
Tedious, unsatisfying tasks may than when identifying white-skinned individuals, and that
increasingly be completed with darker-skinned females are the most misclassified group. The
support from these tools, allowing disparity appears to be in part because the data sets used to train
existing workers to focus on better the systems are predominantly male and white. The research
serving the needs of customers. paper supports earlier findings that algorithms trained with
biased data result in algorithmic discrimination and pushes for
algorithmic transparency and accountability.18

Debate about the ethics of AI often focuses by concern about job security. The
on concerns about the threat of mass potential for significant positive change will
unemployment and the danger to humanity result in uncertainty for risk-averse employees
if machines that will never be able to achieve inclined to be skeptical about change. This
emotional intelligence and empathy should may be more evident in industries like
develop cognitive intelligence that grows insurance where most employees are not
to exceed humans. These issues have been familiar with the rapid change evident in
subject to extensive academic study and many other industries. A key will be the
become part of popular culture through potential for rapid change. AI may reshape
movies and writing. Recent concerns about the workplace by replacing some jobs with
the application of these algorithms focus on others that require different skillsets.
the fair treatment of consumers in terms of
Warren Buffet expressed concern at an
explainability, unintended bias, and impact
annual meeting of Berkshire Hathaway
on diversity.
about the importance of business leaders
Augmentation versus automation being prepared to address questions about
the “plight of displaced workers”:
Over the next 10 years, artificial intelligence
and machine learning are expected to support “If you cut everybody’s hours in half, it’s
existing workers to become more productive, one thing. If you fire half the people and the
effective, and satisfied with their work. AI is other half keep working [then] we need to
not expected to replace human workers in have policies that take care of the people that
the insurance industry, as has been evident become the roadkill in the process.”19 Business
in industries that include agriculture and leaders planning to use artificial intelligence
manufacturing. Tedious, unsatisfying tasks and other emerging tools must develop
may increasingly be completed with support expectations about the anticipated changes
from these tools, allowing existing workers across their business plans and communicate
to focus on better serving the needs of this information with employees, consumers,
customers. Nevertheless, the promise of the and other stakeholders.
new systems will likely be accompanied

18 Buolamwini and Gebru, “Gender Shades: Intersectional Accuracy Disparities in Commercial

Gender Classification.”
19 Carrier Management, “What Warren Buffett Said about AI.”

18
 Six Critical Questions Emerging Issues Research Series

Data scientists and others Appropriately trained employees will increasingly be supported by AI
developing the new artificial tools over the next 10 years.
intelligence tools need to be directed
Through the implementation stages established formal processes to identify and
to explain their new findings—a anticipated over the next 10 years it is directly support new approaches to finance,
task that may be difficult when expected that existing employees will need to data, and ideas.
assessing large datasets. be trained to implement the findings from the
Deployment also presents challenges for the
new learning systems across most industries.
insurance industry. Once a model is developed
This would include the insurance industry,
where initial tools are likely to focus primarily and tested there will be challenges to integrate
on supporting existing employees. it into existing systems and processes. This
will require close collaboration between
The major impact of robotics and other analytics and information technology teams.
technologies that brought job losses in
manufacturing and agriculture due to A challenge in the application of machine
automation were the result of decades of learning and artificial intelligence in insurance
change. Appropriately trained employees will involves explainability. Systems that learn
increasingly be supported by AI tools over the to better anticipate outcomes of interest to
next 10 years. the insurance industry must also be capable
of explaining how they arrived at those
conclusions. Explainability is important if
Lessons for the insurance industry the insurance industry is to confidently apply
Big data analytics is a natural next step for the new approaches. Consumers, regulators,
the insurance industry, but the application and other stakeholders will demand an
of machine learning and artificial intelligence explanation why the new tools generate
represents a greater change to conventional different outcomes. Data scientists and others
practices. Some insurers in Canada are leading developing the new artificial intelligence
the way through the establishment of world- tools need to be directed to explain their new
class research labs and by fostering a culture findings—a task that may be difficult when
of innovation. Others will follow as the new assessing large datasets.
systems are proven to improve efficiency and Moreover, training will be important for
make it possible to better serve customers the industry. AI is unlikely to change the
and reduce costs. Some change is driven by number of people working in the Canadian
innovative companies that sometimes operate insurance industry over the next 10 years, but
at arm’s length from the industry. Some it is expected to change the skills required
strive to disrupt current practices through in many circumstances. The success of
the application of new approaches and organizations, like the Insurance Institute of
techniques. Successful innovators will likely Canada, to provide training will be important
be drawn into mainstream industry practices in supporting the success of the industry in
over time. Many insurance companies have the transition to a digital future.

19
Six critical questions There may be no limit to the extent artificial
intelligence and big data analytics can
doing its part to flatten the curve and was
able to showcase its adaptability in a changing
Why are AI and big data analytics transform the insurance industry. The environment. This included remote claims
expected to transform the potential to deliver custom coverage tailored resolution and innovative approaches to
to individual customer needs matched welcome new customers, demonstrating that
insurance industry? with accurate pricing is fast approaching. the path forward to a digital future for the
Responsible application of emerging insurance industry may come sooner than
data science tools holds great promise to widely anticipated.
“Clinging to the past is the
better serve the risk management needs
problem. Embracing change is the of customers. Insurers will be able to
answer.” confidently offer coverage for risks that Insurance analytics
were not previously insurable, sometimes Analytics has been applied to insurance
in collaboration with government agencies for many years. An initial focus involved
— gloria steinem and other partners. Change and improvement rigorous actuarial assessment of recent
will be evident throughout the insurance industry loss data. Pooling of industry data

It is an exciting time for those who embrace a culture of change

and innovation—terms not typically associated with the insurance
industry in Canada.

industry, including insurance and reinsurance was essential for effective risk assessment
carriers, brokers and agents, adjusters and and development of actuarially sound pricing
appraisers, and others across the industry. for consumers. Development of industry
It is an exciting time for those who embrace datasets was promoted by the industry and
a culture of change and innovation—terms insurance regulators. Some jurisdictions
not typically associated with the insurance required insurance companies to regularly
industry in Canada. file detailed analyses of industry loss data to
justify decisions about auto insurance rates.
The digital transition began many years ago, Actuarial analysis and data analytics were
accelerated over the last five years, and is central to insurance practices and embraced
expected to speed up over the next 10 years. by regulators as being in the public interest.
Initially, change was led and driven by a small
number of industry innovators in Canada and Over time, companies developed tools
elsewhere around the world. Increasingly, to assess pooled industry data blended
insurance consumers have begun to press with additional information provided
the industry to adopt approaches proven by their consumers. Change was most
evident with larger companies that had
to be successful in other industries. For
access to extensive information about
example, many in the insurance industry
their consumers and applied it to the
identify digital-first retailers like Amazon as
development of a more granular assessment,
pioneers in the development of customer-
which is important for risk selection, rating,
friendly experiences they would like to bring
and strategy. Privacy regulations largely
to Canadian insurance consumers.20
accepted and endorsed evolving industry
The COVID-19 pandemic forced the practices, focusing on the assurance that
industry to accelerate the introduction of customer information would be protected
some changes. The insurance industry was and only used for insurance purposes.

20 Malik, “COVID-19 Has Accelerated Changes to the Claims Industry.”

20
 Six Critical Questions Emerging Issues Research Series

Eventually, the industry sought information to devices in the home or place of business.
that was not provided directly by consumers. Some companies have begun to access
Beyond auto coverage, this involved social media information. The objective is
developing models to assess low-probability, to learn more about consumers. When this
high-consequence tail events, like potential information is used properly it can provide

Telematics and connected devices are a visible demonstration to

consumers of the emerging capacity in a digital age to directly link
behaviour and use with insurance practices.

claims from a catastrophic earthquake.21 a stronger foundation to resolve claims and

This included acquiring information develop products that better anticipate
purchased from third-party vendors and using consumer needs and expectations.
independent models. Insurers can also learn
This evolution of analytics was accompanied
from broader data that may focus, for example,
by a remarkable increase in the capacity to
on the neighbourhood or community rather
store, assess, and quickly access enormous
than the individual.
datasets. These emerging tools are well
Some insurers assess data from telematic suited to improve a broad range of insurance
devices installed in vehicles and connected practices. Much more data is readily available
devices in homes or on industrial equipment. and accessible to support interactions
Real time information is provided directly throughout the insurance process, develop
to insurance companies. The information new coverage, and improve assessment of
is used to link insurance coverage and consumer risk. Increased capacity has been
pricing with consumer behaviour. Drivers, accompanied by a significant reduction in the
homeowners, and businesses can also use cost of the new systems and a growing pool
the information to monitor and improve of talented analysts.
their behaviour and sometimes prevent
losses. “Today, they can price on the basis of Insurance companies seek more information
real usage and driving behavior, such as how about current and potential consumers.
fast a vehicle is being driven and whether it Tools to assess and understand the data are
is being driven at night.”22 Telematics and evolving rapidly, including digital ecosystems
connected devices are a visible demonstration that provide easy access through the cloud to
to consumers of the emerging capacity in a third-party providers. Capacity to manage
digital age to directly link behaviour and and process growing volumes of quality
use with insurance practices. This includes information have increased remarkably over
devices within vehicles, mobile apps on the past 25 years. Also, the analytical tools to
the telephone of consumers, and direct assess and understand the data continue to
connection through the Internet of Things grow in capability.

EY found that “While it is still early, insurers are exploring how to deploy
AI and ML technology…some are beginning to invest substantially,
while others are more cautious. What problems does the industry think
it needs AI to solve?”23

21 Kunreuther and Grossi, eds., Catastrophe Modeling.

22 Balasubramanian, Libarikian, and McElhaney, “Insurance 2030.”
23 Tapestry Networks, “Three Ways AI Will Transform the Insurance Industry.”

21
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

Digitalization and the insurance value chain

• Customer-specific “targeted” • Automated (including non- • Platform business models

marketing human) product service • 360 degree view of customer
• Robo-advice, AI and chat bots centres using robo-advice, for consultants
chat-bots and AI
• Internet sales and Price • Continuous real time data
Comparison websites • Big data enables ability to enabling focus on high value
predict what customers want customers
• Social media and SMART and need before they ask for it
phone/device channels for • Unstructured data (e.g. voice)
direct distribution • Continuous real-time analysis learning
customer communication
and U/W

Marketing, Sales Pricing and Product Claims Customer

and Distribution Underwriting Management Handling Interactions

• Telemetrics — customers and insurers understand • Fraud detection using Big Data and Blockchain
risk much better (wearables, IoT, SMART phones, • Blockchain facilitating trust-worthy and timely
apps) claims information
• Big data enabling more granular and accurate • AI and drones in assessing processes
pricing and faster U/W
• Claims cost efficiencies re online/SMART device
• Blockchain in technologies to seemlessly manage claims lodgement, AI/automated assessing,
and instantly verify data sources optimised payouts, reduced labour costs
• Peer-to-peer insurance models • Supply chain management efficiencies, vertically
• Granular, customer-specific product offerings integrated claims processes
including usual-based insurance

Source: IAIS, “Issues Paper on Increasing Digitalisation in Insurance and Its Potential Impact on Consumer Outcomes,” page 8.

Growth bar graph and composition of patents (donut chart, 2018) at insurers

800 6%
6%
693
700 Customer service/VAS

Claims
30%
600 11% 11%
Operations
500 Early warning system
51%
400 Underwriting
38%
300 225
Distribution

200 121
Motor 23%
82 95 92
100 25 38
Multiline
24%
12 Property
0
2010 2011 2012 2013 2014 2015 2016 2017 2018
Number of filings

Source: Swiss Re Institute, sigma No 5/2020, page 14.

22
 Six Critical Questions Emerging Issues Research Series

Underwriting applications Smaller organizations can learn from the

findings of others or purchase tools and
Big data analytics and, to a lesser extent at
algorithms to use at their level.
this time, artificial intelligence are essential
to support underwriting decisions. Analytical Aggressive, long-term analytics initiatives
capacity and data quality have emerged as involve large datasets. The presence of

Analytical capacity and data quality have emerged as critical factors to

determine the performance of insurance companies.

critical factors to determine the performance older, legacy information technology systems
of insurance companies. may serve as a constraint for companies
that aspire to bolder data management
Development of datasets and analytical
and analysis efforts: “Many insurers have
tools were initially put in place to support
enormous amounts of valuable data in
industry actuaries. Over time, the insurers
numerous systems, some of which are
began using statisticians and data scientists to
legacy systems. But they were often not
support underwriting decisions. Research by
the Insurance Institute of Canada and others recorded in an easily-interrogable form with
set out company and industry strategies consideration for future use. So many data are
to find, hire, train, and otherwise develop not readily accessible.”25 Companies involved
individuals trained to implement this work. in recent acquisitions may have multiple
The Institute’s “Demographics” report systems in place with distinct histories and
concluded that “technology is redefining processes. Telematics and connected devices
workforce needs,” noting that “analytics and generate large datasets. Companies focused
data analysis skills are growing in importance on managing climate risks have access to
and urgency as organizations look to increase enormous volumes of climate history data.
their ability to collect, interpret and monetize Some consumers use social media and
data and their competitive advantage.”24 produce text and video information useful
to insurers. Data management is a complex
Some companies seek to secure as much high- challenge for insurance companies, and it will
quality data as possible about consumers. The grow in complexity over time.
data are organized and assessed to support
The goal for insurers is to develop products
risk selection, rating, and product design. The
information can be analyzed to better assess aligned with consumers’ evolving risk
risk for new and existing consumers. The management needs with a price adequate
gathered data can more accurately determine to cover anticipated claims and expenses.
what each customer should pay to properly Investment in analytics is a priority for
cover their risks. Potential new customers innovative companies seeking to be early
will be welcomed with an accurate estimate to offer new products and services and to
as their risk is proactively assessed. support the modernization of legacy systems.
This is evident for insurance companies,
Other companies, in contrast, seek only brokers, adjusters, and others across the
minimal information to support decisions insurance industry. Confident and successful
and seek to avoid acquiring and managing leadership in the insurance industry requires
additional data. Simply put, companies can an understanding of the needs of consumers,
choose to take as much or as little data as and increasingly analytics is critical to identify
they are comfortable gathering and analyzing. the best path forward.

24 IIC, “Demographics of the P&C Insurance Industry in Canada 2017–2027.”

25 Raman and Lam, “Artificial Intelligence Applications in Financial Services.”

23
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

The COVID-19 pandemic Claims resolution • Build models to anticipate the frequency
and severity of claims through the
accelerated the introduction The greatest opportunity for insurers with
insurance cycle and link this to product
the new technology is to help consumers
of new tools, as conventional design, distribution, and marketing to
avoid claims. Warnings about an approaching
assessment of claims was difficult hailstorm could help insurers get their
better anticipate consumer needs.
or impossible. consumers to move their vehicles to a safer Beyond the three priorities identified by PWC,
location. Water sensors in a home may many also stress the importance of models to
help insurers inform homeowners before a reduce claims fraud.
water damage claim develops. Telematics
may be used to help drivers adopt safer The COVID-19 pandemic accelerated the
driving practices. introduction of new tools, as conventional
assessment of claims was difficult or
Adjusters and claims managers are also
impossible. Intact, for example, reports that
beginning to use artificial intelligence tools
10 to 20 percent of vehicle-collision damage
and analytics to improve customer support
estimates in 2019 were based on photographs
and resolve claims. New approaches are in
development to collect and process damage and videos submitted by customers, and this
claims. PwC, for example, identifies early, increased to more than 60 percent during
intermediate, and advanced stages in the the pandemic. “Intact said photos are sent
application of artificial intelligence to claims electronically for about 60% of auto claims,
management:26 giving them the opportunity to create
estimates and pre-order parts for repair.
• Build predictive models and analyze
Adjustment to these estimates can later be
claims process flows.
made when cars are being repaired.”27
• Build deep learning tools that process
New approaches to claims resolution
images and estimate repair costs and
build capacity to monitor input from were introduced sooner or their use was
remote sensors and prevent losses expanded. Importantly, consumers got to
for consumers. partake in the completion of their claim.

Claims innovation timeline

Claims journey Past Present Future

FNOL (First Notice of Loss) Complex forms which Customer uses the phone, post, Automatically: IoT satellite,
require explanation online / mobile apps weather stations, etc.

Claims admin Manually by insurer staff Data entry by insurer staff Digitally augmented support in
complex situations; AI advice

Data gathering / Rudimentary fraud analytics Combination of manual and Automated using analytics like
fraud detection models to identify fraud big data and AI

Claims estimate Manual claims adjudication Mix of manual inspection or Automated claim adjudication
advanced technologies

Settlement Reimbursement via bank Direct to vendor (garage, hospital) Instant payout, automated
cheques / drafts and reimbursement validation, options (eg,
replacement, second hand)

Source: Swiss Re Institute, sigma No 1/2020, page 38.

26 PwC, “AI in Insurance.”

27 Malik, “COVID-19 Has Accelerated Changes to the Claims Industry.”

24
 Six Critical Questions Emerging Issues Research Series

Additionally, the wait time for them to receive significantly reduce criminal activity. New
payment may be reduced because they do tools are also emerging for companies to
not have to wait for an adjuster or appraiser identify and confront individuals who make
to be available in person. Companies have fraudulent or exaggerated claims. Company
established a new relationship with the efforts can be strengthened through analysis

The application of analytics to detect fraud “is among the fastest areas
of tech adoption in the insurance industry.”
consumers they plan to build on, with an of social media postings and third-party data.
expectation that proven practices will not Rules, regulations, and case law setting out
return to previous approaches. acceptable investigative practices continue
to evolve as these approaches are increasingly
The application of analytics to detect fraud
adopted, which benefit greatly from the
“is among the fastest areas of tech adoption
advancements made in AI. A priority for
in the insurance industry.”28 EY observes that
the industry is to reduce false positives and
“AI technology is increasingly deployed to
the resulting terrible claims experience for
detect and prevent [insurance] fraud, which
consumers wrongfully accused of fraud. AI
costs the industry an estimated $40 billion
can address this through its ability to learn
a year.”29 Machine learning and AI tools
and shift to address emerging fraud patterns
can assess company data and information
rather than constantly finding signals of
pooled across companies to detect suspicious
where fraud used to be found.
criminal practices. Most fraudsters follow
patterns that AI systems can be trained A number of companies are providing AI tools
to identify. The new information can be to support the claims management process

A priority for the industry is to reduce false positives and the resulting
terrible claims experience for consumers wrongfully accused of fraud. AI
can address this through its ability to learn and shift to address emerging
fraud patterns rather than constantly finding signals of where fraud
used to be found.

designed to better inform and support early of insurers. Canadian insurance publications
and forceful response by the police. Company regularly report on companies operating
and industry investments in fraud-reduction in Canada. More than a dozen US-based
tools are expected to increase over the next companies were identified in a 2019 article
10 years because they hold great promise to by Mike Thomas, including the following:30

28 Faggella, “Artificial Intelligence in Insurance.”

29 Tapestry Networks, “Three Ways AI Will Transform the Insurance Industry.”
30 Thomas, “25 AI Insurance Companies You Should Know.” Slava Kurilyak provides another list
in “Artificial Intelligence (AI) in Insurance.”

25
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

Examples of companies supporting claims through AI

AI Insurance provides software tools for small insurance companies to automate aspects
of claims management;

Avaamo is an artificially intelligent chatbot that collects and compiles claims

information from consumers;

Betterview uses AI to estimate repair costs based on historic experience;

CCC Smart Estimate provides a prediction of vehicle repair costs;

Claim Genius offers an AI platform to assess collision photos and videos and
estimate damage;

Clearcover uses photos to estimate claims;

Galaxy AI’s Galacticar provides an app so customers can report claims;

H2O.AI provides insurance supports that include systems to detect suspicious claims;

Lemonade has developed chatbots and AI algorithms to rapidly process claims;

Nauto seeks to avoid collisions by providing loss avoidance advice to feed drivers
in real time;

The RiskGenius platform streamlines insurance claims using custom algorithms;

Solaria Labs’ Auto Damage Estimator uses AI to process vehicle damage photos; and

TROV has developed chatbots that support claims management.

26
 Six Critical Questions Emerging Issues Research Series

Examples of value-added and preventive services

Service type Auto Property
Preventive services • Safe driving alerts • Remote monitoring and alerts

• Maintenance rewards • Flood monitoring in

home basements
• Anti-theft and breakdown alerts
• Automatic shutdowns of
• Geo-fencing (eg, alert if driven outside
gadgets during fires
of safe neighbourhood)
Value added services • Assistance to buy car • Property security advice

• Concierge services • Facility maintenance

• Alert when vehicle is moved, towed or • Emergency repair services

hit when parked

Source: Swiss Re sigma No 1/2020, page 32.

Loss prevention will grow in importance The new tools are unlikely to have a
in claims management. Connection with significant direct impact on companies that
the Internet of Things, sensors, and other adopt the most conservative investment
devices support development of value-added strategies. Nevertheless, insurers will need
and preventative services for vehicles and to understand how these tools are used by
homeowners. Swiss Re, for example, has third-party advisors.
identified the opportunities above.31 Some insurance companies are active in the
management of their investments. This may
Investment management include purchasing bonds with different risk
Insurance companies are major investors. and reward characteristics that may be traded
AI and analytics can be applied directly by before maturity, or investments in equities,
insurers to support investment decisions real estate, or other assets. AI and analytics
or can be used by third parties to advise can be applied directly by insurers to support
insurers. The skills and capacity of these management of investment risk and be used
tools are expected to improve considerably by third parties to advise insurers.
over the next 10 years. They can be applied to Over the next 10 years, all insurance
support actions to minimize investment risk,
companies are expected to focus
improve returns, and measure compliance
increasingly on environmental, social, and
with established strategy.
governance (ESG) issues in their investment
Some companies prioritize preservation of management, including climate risk.32 This
capital, choosing to minimize the risk of loss will include conservative and active investors.
in the portfolio. This may involve investing Several Canadian insurance companies are
exclusively or predominantly in high-quality international leaders in the development of
bonds that are frequently held to maturity. tools to measure and manage climate risks.

AI and analytics can be applied directly by insurers to support

management of investment risk and be used by third parties to
advise insurers.

31 Swiss Re, “Data-Driven Insurance?” page 32.

32 Kovacs, “Climate Risks.”

27
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

The Task Force on Climate-Related Financial eliminate the tedious task of eliminating
Disclosure established an insurance those who are not qualified. Aspects of
companies working group to develop background checks can also be automated
disclosure guidance focused on underwriting AI tools can be used to support human
and investment activities.33 resources professionals to further narrow the
pool of candidates using information found
Moreover, companies that support the in applications supplemented by additional
United Nations Principles for Sustainable information collected from references, online
Insurance commit to “integrate ESG issues searches, or analysis of social media.
into investment decision-making and
ownership practices.”34 Tools are emerging Expanded data about staff experience and
to assess and disclose ESG and climate risks interests can be used by human resources staff
in insurance investments. to develop more effective training programs,
establish career pathways, and strengthen
retention of critical employees. Analysis of
Hiring and staff support previous work experience can better match
existing staff with open roles. Technology
Machine learning and AI are becoming
can be used to better connect companies
mainstream in human resources practices
with current and prospective employees on
in all industries, including insurance. They
a personal level.
are increasingly embedded in recruitment,
onboarding, performance analysis, training,
and retention. Automating some processes Marketing
allows human resources professionals to
Marketing teams can leverage customer
focus their time on making better decisions
datasets using customer relations
to establish and maintain the best workforce.
management technology and AI functions.
It is now easier to inform existing employees Optimization models can support cross-
about vacancies and, where appropriate, marketing campaigns and the matching of
reach outside the company to other qualified niche insurance products to better serve
people. Employee referrals can be tracked to the needs of consumers. Chatbots can
recognize staff that are successful in helping provide an AI-empowered conversation
to attract others and understand the kinds interface. These approaches can be applied by
of employees identified. Candidate résumés insurers, brokers, agents, managing general
collected online can be subjected to an agents (MGAs), adjusters, and others in the
initial screening by automated systems that insurance industry.

AI tools can be used to support human resources professionals to further

narrow the pool of candidates using information found in applications
supplemented by additional information collected from references,
online searches, or analysis of social media.

33 UNEP FI, “UNEP FI Working with 16 Global Insurers to Better Understand Risk & Implement
TCFD Recommendations.”
34 UNEP FI, “PSI Principles for Sustainable Insurance.”

28
 Six Critical Questions Emerging Issues Research Series

Lessons for the insurance industry efficiency. Also, it is important to adapt to

evolving circumstances. For example, the
Artificial intelligence and big data analytics
COVID-19 pandemic served to accelerate
can be applied across the insurance industry
the digital journey for several companies.
to improve performance. This journey
of transition began several years ago and Using the right people is essential, but the
recently accelerated. The importance of best are in high demand. Companies compete
AI and analytics will grow over the next 10 with other insurers, banks, pensions, asset
years. Current industry leaders plan to press managers, and regulators to find skilled

A small number of companies account for most of the early

insurance industry focus on AI. Swiss Re reports that 10 insurers
account for more than 80 percent of the AI patents in the United
States over the past 10 years, and five companies account for
more than 85 percent of the patents in China.35 These early
adopters are striving for enterprise-scale transformative change.

forward and bring further change, but every analysts. It is critical that analysts have or
company does not need to aspire to lead. It is can acquire a solid understanding of the
essential that brokers, agents, adjusters, and factors that contribute to insurance risks
companies establish a strategy committing to and opportunities.
a path forward.
The remarkable experience of Amazon and
Insurers committing to innovation several other global leaders demonstrates
included strong leadership from the highest the enormous potential for change powered
levels within the company that helped by a commitment to artificial intelligence
spread a culture of innovation and digital and big data analytics. Learning algorithms
transformation. Innovation was established designed by creative data scientists can
as a priority, and experimentation was improve the performance of the Canadian
encouraged. Discussion focused on insurance industry. However, it is critical to
overcoming barriers and lessons learned. allow sufficient time for new approaches to
Importantly, these efforts were consistently mature, and the journey must allow for some
tied to establishing a better experience for failures along the way.
consumers and increasing operational

35 Anchen, Bohn and Sharan, “Machine Intelligence in Insurance.”

29
Six critical questions Should AI and big data analytics be introduced
in the insurance industry? It is critical for the
propose solutions that were shared with
interested stakeholders.36 Regulatory bodies
How can the insurance industry industry to focus on insurance consumers and like the Office of the Superintendent of
best anticipate and manage their perspective. Innovation leaders involved Financial Institutions or the Canadian
in the application of artificial intelligence Council of Insurance Regulators could
emerging risks? and big data analytics consistently identify impose requirements. The Property and
a number of challenges that extend across Casualty Insurance Corporation, which was
“Your scientists were so an array of industries. Four risks that are established more than 30 years ago to protect
addressed here include explainability and insurance consumers in the unlikely event
preoccupied with whether they transparency, fairness and bias, availability, that a company fails, could evolve to also
could, they didn’t stop to think if and privacy and security. Solutions to assume responsibility to determine consumer
they should.” facilitate the adoption and consumer acceptability of evolving insurance practices.
acceptance of AI and big data analytics Presently, it is unclear how the industry would
include lessons that can be learned from other seek to determine the social acceptance of
— jurassic park industries and early adopters, but these must new approaches.
be tailored to the specific needs of Canadian
insurance consumers.

The UK Centre for Data Ethics and Innovation warns about three potential
detrimental outcomes that could arise from the implementation of AI
for personal insurance:
• The collection and sharing of large data troves, which could impinge
on privacy, if done without the express consent of customers.
• Hyper-personalized risk assessments, which could leave some
individuals “uninsurable” by revealing previously unseen
indicators of risk.
• New forms of nudging, where insurers use AI to alter the behaviour
of consumers in a way that could be viewed as intrusive.37
While consumers, regulators, and a number
of other stakeholders will ultimately provide Explainability and transparency
judgment about the social acceptability of Explainability is the ability to explain how
decisions made by the insurance industry, it is assessment of large datasets with complex
unclear what forum would be best to explore models and algorithms arrived at a conclusion.
and test new ideas. The OECD has endorsed Analysts may struggle to explain why a factor,
the approach of regulatory sandboxes, or combination of factors, appears to have a
where new ideas are tested in a controlled significant impact on the assessment of risk.
setting with regulatory supervision. TD Bank Consumers deserve to be informed about
established an expert panel to participate in changes in their expected risk of loss, as
a roundtable to identify critical issues and reflected in the price they must pay for

36 Contant, “Focus on These Three Areas When Developing AI Best Practices.”

37 CDEI, “AI and Personal Insurance,” page 4.

30
 Six Critical Questions Emerging Issues Research Series

Black box models that predict coverage, and it is important that this can to make a flood insurance claim than those
be resolved if insurers are to rely on the without a boat.
outcomes without an explanation emerging tools. Explainability emerged as a
An assessment of causality may find that
are incomplete and not ready for significant challenge in other industries with
homes located in the flood plain are more
use.
An expert panel advising TD Bank found three critical elements involved
in addressing explainability:38
• Identifying what needs to be explained
• Expecting the unexpected
• Educating and reaching consensus
likely to flood due to the location of the
the introduction of AI and big data analytics, building and, coincidently, homeowners
and insurers must learn from this experience. located near water are more likely to own a

The OECD stresses that analysis should be boat. The risk of flooding does not increase
with the purchase of a boat—it is the result
informed by insurance knowledge and be
of location in a flood plain. While humans
directed to assess the root causes of loss.39
are able to distinguish this relationship,
Predictive models reporting correlation of AI may not be able to. It is the industry’s
findings are not as effective in explaining prerogative and ethical responsibility to
the expected risk as are models focused on properly manage how it uses the data that
causality. For illustration, analysis may find are collected and analyzed. Black box models
that homeowners with a boat are more likely that predict outcomes without an explanation
are incomplete and not ready for use.

In 1992, the Supreme Court of Canada ruled that the use of age
and gender as a rating criterion was discriminatory but was
permitted because reliable data that would allow the insurance
industry to set premiums based on non-discriminatory criteria
were not available. Two dissenting judges argued that an
insurer must not only prove a statistical correlation but a causal
connection.41 Young drivers, particularly young male drivers,
experience more collisions, but is it socially acceptable to use age
and gender as rating criteria? Age and gender contribute to the
understanding of the risk of loss, but can the industry dig deeper
to identify and assess the root causes of collision risk? Younger
drivers in Canada are involved in more collisions because, in
part, they do not yet have sufficient driving experience. Novice
drivers of any age are perhaps at greater risk. The industry
championed the development of graduated licensing as a public
policy response to confront the experience of more collisions
for newer drivers.41

38 TD, “Responsible AI in Financial Services,” page 8.

39 OECD, “The Impact of Big Data and Artificial Intelligence (AI) in the Insurance Sector.”
40 Gomery and John, “Gender Neutral.”
41 Mayhew, Simpson, and Singhal, “Best Practices for Graduated Driver Licensing in Canada.”

31
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

AI and big data analytics can support delays in the time required to secure
real-time monitoring of connected risks approval and uncertainty in the rates that
in vehicles and homes. This can be used will ultimately be approved; these are risks
to shift the focus of insurance from an that would be expected to increase the values
actuarial analysis of historic correlations of requested relative to a system that relies
loss with socioeconomic factors to a deeper on competition to best serve consumers.
assessment, monitoring, and measurement of Nevertheless, transparency is one element
underlying factors contributing to the risk of where insurance companies in Canada can
loss. Big data analytics conducted by skilled learn from rate regulation.
analysts with expert support will allow for
Rate regulators are trying to help insurance
deeper study of the root causes of risk. This
consumers, and this often includes better
will improve the capacity of the insurance
communication about insurance practices.
industry to anticipate the risk of future losses,
This is increasingly evident for climate-related

The debate about the use of credit scores by insurance companies

provides a warning that effective communication is challenging.
risks. Alabama, Connecticut, Florida, Georgia,
which is good for consumers, particularly
Louisiana, Maryland, Mississippi, New Jersey,
those with low risk. But acceptance of the
New York, North Carolina, Oklahoma, Rhode
new tools requires that the findings are
Island, South Carolina, and Texas are some
understood by insurers and can be explained
states where insurance companies provide
to consumers. Analysts must identify the
premium discounts to homeowners that
path between the identified change and
install certified loss reduction measures.42
the resulting adjustment in the risk of loss,
Residential consumers pay a higher price for
and this information must be shared with
insurance when these features are not present
consumers. The debate about the use of credit
and are offered a discount when they are.
scores by insurance companies provides a
Some regulators require that insurers offer
warning that effective communication
a discount, and the value is determined by
is challenging.
the company. Others have established a
Consumers typically do not understand minimum discount.
how the insurance industry functions in
Insurance companies in Canada typically
Canada. Many speculate about behaviours
charge a lower premium for consumers
that increase or reduce the cost of insurance,
with lower risk, but the justification is often
but actual industry practices are not evident
unclear. The promise of insurance discounts
nor made clear to consumers. This disconnect
is one of the most important factors causing
contributes to the view that insurance is a
more homeowners in Texas to install
necessary expense rather than a critical
hail-resilient roofs and thus reduce their
form of protection for drivers, homeowners,
risk of hail damage. Actions taken by rate
and businesses.
regulators to improve the transparency of
Rate regulation harms insurance consumers. the insurance process provide insights into
It is expensive for insurance companies to practices that can better inform insurance
comply with rate regulatory processes, and customers—practices that can be introduced
this cost must be recovered from consumers. in jurisdictions fortunate not to have rate
Rate regulation stifles innovation and delays regulation. This improves the relationship
introduction of new practices that would between consumers and brokers; building
benefit consumers. Moreover, insurance trust and enforcing safer practices have a
company rating proposals must anticipate financial benefit for both parties.

42 Smart Home America, “List of Mitigation Insurance Discounts and Tax Savings.”

32
 Six Critical Questions Emerging Issues Research Series

It will be crucial for the Fairness and bias odds with the approach adopted by the
insurance industry. The Merriam-Webster
industry not to act blindly on The insurance industry is committed to
online dictionary, for example, provides 28
the findings provided by these the use of decision support tools that are
definitions of “fair,” but does not mention
fair and unbiased in the assessment of risk,
new tools without a rigorous determination of pricing, resolution of claims,
actuarial valuation of risk in any of them.

understanding of how consumers and treatment of customers. Analysis of large If insurers triple the price of hurricane
and the public as a whole datasets and the introduction of machine insurance in Florida when new research finds

interpret these actions. Statistical methods for assessing fairness involve evaluating disparities
in error metrics for a variety of inputs. Machine learning fairness toolkits
include AI Fairness 360, Fairlearn, and Google’s ML-fairness-gym.

learning decision support tools hold great a tripling of expected hurricane damage, this
promise to reduce unconscious bias in is actuarially fair, but is it fair for a senior on
industry practices. Nevertheless, it is critical a fixed income who may no longer be able
that the industry be vigilant in the search for to afford the cost of staying in her home?
unanticipated bias in these complex new Actuarial valuations are a dimension of
analytical tools. determining fairness, but consumers seek
assurances that the industry understands
Statistical methods for assessing fairness
the broader context of modern life. Concepts
involve evaluating disparities in error metrics
like accident forgiveness are examples of the
for a variety of inputs. Machine learning
Canadian insurance industry adapting to
fairness toolkits include AI Fairness 360,
address consumer expectations.
Fairlearn, and Google’s ML-fairness gym.
These systems have been applied in other Systemic racism is a current issue in the
sectors to assess the fairness of systems in world. Implementation of AI and big data
simple environments, but their capacity is analytics in the insurance industry may
difficult to judge in a broader context. involve questions about the fair treatment of
visible minorities. The insurance industry has
A tool developed to support hiring at Amazon
had an ongoing struggle with its reputation in
was found to have learned to prefer male
general, including explainability and visibility.
candidates over female ones, penalizing
It will be crucial for the industry not to act
applications that included the word “woman”
blindly on the findings provided by these
or those attending all-female colleges. The
new tools without a rigorous understanding
software was trained on data submitted
of how consumers and the public as a whole
mostly by men over a 10-year period. Despite
interpret these actions. Analytic teams that
efforts to fix the bias when it was detected,
are diversified to reflect the makeup of
Amazon eventually abandoned the project
consumers are more likely to identify and
because it lost faith in the impartiality of the
correct biases.
system.43 The initiative was doomed from the
start because it was based on poor data about The application of emerging tools will test the
unbiased decision making. insurance industry’s capacity to respond to
the evolving needs of consumers. It will be
The insurance industry strives to align prices
critical to assess the expanding databases to
with the expected risk of loss. Insurers have a
ensure that the information is unbiased and
view of fairness based on actuarial valuation
will support equitable decisions.
of risk. Public views of fairness and acceptable
practices are complex and sometimes at

43 Hamilton, “Amazon Built an AI Tool to Hire People but Had to Shut It Down Because It Was
Discriminating Against Women.”

33
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

An expected outcome of Availability identified to the applicant. The industry

has prioritized many actions to safeguard
the application of big data Vigorous competition is found in most private
customer information from cyber criminals,
analytics is the prospect that insurance markets for vehicles, homes, and
hackers, and unauthorized staff. Best claims
business across Canada. Nevertheless,
insurance companies will begin company decisions to withdraw coverage
management practices are emerging with
respect to the ethical use of information from
introducing coverage of risks can have a devastating impact on consumers.
social media and other sources. Companies
that were previously seen to be Large shocks, like Hurricane Andrew in
have established processes to resolve
Florida, significantly disrupted the availability
uninsurable. of coverage as many companies sought to
consumer complaints.

exit the market. Withdrawal of coverage Federal and provincial privacy legislation
may threaten the livelihood of consumers if applies across all industries. The insurance
they are unable to secure insurance needed industry successfully argued for some
so they can drive to work or maintain a loan accommodations to best serve the specific
for their home or operate their business. It needs of insurance consumers, such as the
is important that companies considering the treatment of consent. The industry will need
prospect of withdrawing from a market or line to continue to adapt its practices to ensure
of business have more information available that the specific elements of big data analytics
than could be provided by the findings from and the development of machine learning
a new model. algorithms comply with (and can be shown
to comply with) evolving privacy regulations,
An expected outcome of the application
especially as additional data resources are
of big data analytics is the prospect that
used by AI, such as social media.
insurance companies will begin introducing
coverage of risks that were previously seen A primary focus of cyber criminals and hackers
to be uninsurable. The availability of more over the past decade was on banking, health
information and stronger analytical tools will care providers, and government agencies.
likely determine, over the next 10 years, that The insurance industry has adapted practices
new coverage is viable. Consumers will find tested in other industries to address the needs
that they are able to transfer additional risks of insurance companies.44 Ransomware
to the insurance industry for a reasonable and other cyber threats that target smaller
premium. An early example involves the organizations warn of the need to build
introduction of residential flood insurance. cyber awareness and protection across
the insurance industry, including brokers,
Privacy and security agents, and adjusters. Enhanced protection of
consumer information from cyber criminals
Federal and provincial agencies are actively
is needed by insurance companies and across
working with the insurance industry to
the industry and will grow in importance with
protect the privacy of consumer information.
the development of shared access to large
A focus has been on the protection of personal
industry databases.
information collected from customers and
potential customers. Information collected As previous studies by the Insurance Institute
must only be used for the specific purposes of Canada have shown, the insurance industry

The industry will need to continue to adapt its practices to ensure that
the specific elements of big data analytics and the development of
machine learning algorithms comply with (and can be shown to comply
with) evolving privacy regulations, especially as additional data resources
are used by AI, such as social media.

44 Kovacs, “Cyber Risks Report 2019.”

34
 Six Critical Questions Emerging Issues Research Series

must do more to prepare for cyberattacks.45 In better outcomes, some consumers will
the past, hackers and cyber criminals largely likely experience higher prices, reduced
ignored the insurance industry. However, availability, and other unwelcome changes.
as more data are gathered, stored, and Companies unable or unwilling to prepare
analyzed, there is greater motivation for cyber for these challenges will find that consumers
criminals to attack the insurance industry, are ready to move their business elsewhere,
and the consequences of those hacks will which might not be as simple as it was in the
be more dangerous for insurers, brokers, past, and will speak up about their treatment,
and consumers.46 which is much easier with social media.

Explainability and transparency could be

Lessons for the insurance industry challenged by regulators and the public.
Consumers deserve to know why their
Machine learning, artificial intelligence, and
expected risk of loss has changed. Leadership
big data analytics hold great promise and risk
on market conduct and fair treatment of
for insurance consumers and the insurance
consumer issues provided by the Canadian
industry. Early innovators in insurance and
Council of Insurance Regulators is typically
other industries report four critical challenges
triggered when a practice is adopted across
that the insurance industry will need to
most of the industry and then challenged as
address: explainability and transparency,
unfair by consumers. These conversations are
fairness and bias, availability, and privacy and
well managed but would be less disruptive
security. These issues are not new, but they
if they could start earlier. A concern is the
are amplified with AI and big data analytics.
absence of a forum where industry leaders
Introduction of the evolving technology
can work together with key stakeholders
is expected to increase the importance of
to acknowledge, anticipate, and resolve
the industry demonstrating to consumers,
issues. The Property and Casualty Insurance
regulators, and other stakeholders that
Corporation (PACICC) was established
insurers understand and are proactively
more than 30 years ago to protect insurance
working to address these important concerns.
consumers in the unlikely event that a
Emerging tools have the potential to company fails. PACICC could evolve to
improve relations with insurance consumers also assume responsibility of determining
if introduced responsibly. While most consumer acceptability of evolving insurance
consumers are expected to experience practices and as a centre to showcase best
industry practices.

45 Kovacs, “Cyber Risks Report 2019.”

46 Kovacs, “Cyber Risks Report 2015.”

35
 The application of machine learning, artificial Association will monitor the implications for
Six critical questions intelligence, and big data analytics is presently auto insurance rate filings. Other regulatory
under active study by insurance supervisors organizations will supervise cyber security,
Why should insurers in Canada and elsewhere. They seek to fraud, money laundering, and other risks.
expect increased attention understand how insurers and other financial
from regulators?
David Collingridge, in his 1980 book The Social Control of
“We’re charged by Congress Technology, sets out the “Collingridge dilemma”: The impact of
with regulating financial new technology cannot be predicted until it is in wide use, but
institutions. We take that mission regulatory oversight of the technology is no longer possible
very seriously. We are tough when it has become entrenched. Larry Downes’ 2009 book,
supervisors and regulators.” The Laws of Disruption, sets out the “pacing problem,” which
argues that technology changes exponentially, but social,
— janet yellen economic, and legal systems change incrementally. Technological
innovation often outpaces the capacity of regulators to keep up.
One solution to these issues involves prohibiting or curtailing
innovation until proponents can prove that the changes will not
cause harm. This interpretation of the precautionary principle
has been widely criticized, but it nevertheless provides insight
into the challenges facing regulators responsible for protecting
consumers from harm.

institutions plan to use these emerging While most regulators have expressed their
tools to provide appropriate oversight and commitment to address AI and big data, the
consumer protection. AI and data science near absence of early action is surprising.
will also be important to support continuous International experience suggests that
improvement in supervisory practices. monitoring and preparation in Canada
will soon result in regulatory action. The
Regulatory interest is widespread, although
various Canadian regulators are informed
there has been little early regulatory action
by a continuing stream of advice provided
to address the application of AI or big data
by the OECD, the International Association
analytics in Canada. Federal and provincial
of Insurance Supervisors (IAIS), and others,
privacy regulators are working to modify
including regulatory changes applied to
their current activities. The Office of the
the banking sector. The OECD has set out
Superintendent of Financial Institutions
principles for the responsible implementation
(OSFI) has adopted a prudential perspective
of AI. IAIS reports on the experience of
to the supervision of AI and big data
insurance regulators around the world as
issues, including a focus on operational
they address AI and the digital economy. AI
risk. Members of the Canadian Council of
and big data analytics have a high profile
Insurance Regulators are expected to address
within the Canadian and international
planned changes from a market conduct and
regulatory community.
consumer fairness/explainability perspective.
The Canadian Auto Insurance Rate Regulators

36
 Six Critical Questions Emerging Issues Research Series

A consistent message involves regulators looking to understand and

judge the social acceptability of new practices introduced by financial
institutions. Regulators seek evidence that the application of AI and
analytics will benefit consumers.

A consistent message involves regulators safety and soundness of financial institutions.

looking to understand and judge the social The benefits of advanced analytics, digital
acceptability of new practices introduced technology, and innovation are mentioned in
by financial institutions. Regulators seek the paper, while the focus of the consultation
evidence that the application of AI and is on managing risk. Nine risks set out in the
analytics will benefit consumers. Some, like paper include the following:
the OECD, have endorsed the concept of
• Confidentiality
regulatory sandboxes and innovation hubs,
where new approaches can be tested before • Availability
their widespread application. Indications
• Integrity
of flexibility and awareness of the benefits
are consistently followed by statements • Soundness
of concern about the risk of harm to some
• Explainability
insurance consumers and warnings about
the potential for intervention. Swiss Re • Accountability
observes, “the regulatory framework will play
• Transparency
an important role in shaping the integration
of new technology into the insurance space.”47 • Reliability

• Substitutability
Prudential regulation The paper indicates that “OSFI believes that
In 2016, Autorité des Marchés Financiers its existing model risk guidance remains
(AMF) established the AMF Fintech Group, relevant, but could be better aligned and
which now has more than 60 employees. The strengthened to address increased use of
lab helps the regulator better understand the advanced analytics.” This suggests oversight
new technology used by financial institutions of property and casualty insurance practices
in Quebec and identify circumstances where that may be distinct from that applied to
these tools should be applied by the regulator banks and life insurance companies. Also,
to improve its own practices. AMF is working prudential oversight of companies actively
to support industry innovation, provide committing to emerging technologies may
effective guidance to financial institutions, differ from the supervision of companies less
and modernize its supervisory operations. involved in these issues.

In late 2020, OSFI initiated consultations It has been common for OSFI to develop
that included AI and analytics and released initial guidance for the major banks that is
a discussion paper: “Developing Financial later modified for life insurance companies
Sector Resilience in a Digital World: and, eventually, general insurance. Discussion
Selected Themes in Technology and Related is traditionally followed with guidance
Risks.”48 The paper focused on cyber security, about expected company practices and
advanced analytics (including AI and machine increasing requests for reporting. Assistant
learning), third-party ecosystems, and data. Superintendent Jamey Hubbs noted that
These issues were framed as operational “AI presents challenges of transparency
risks that must be addressed to ensure the and explainability, auditability, bias, data
quality, representativeness and ongoing data
47 Swiss Re, “Data-driven insurance,” page 23.
48 OSFI, “Developing financial sector resilience in a digital world: Selected themes in technology
and related risks.”

37
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

“If you’re creating models that can’t be explained to regulators, you

shouldn’t be using those models.”50
governance.”49 The CEO of Concentra, Don risks, data management practices, and
Coulter, said, “if you’re creating models that analytics by insurance companies in Canada
can’t be explained to regulators, you shouldn’t is certain to follow in the near future.
be using those models.”50
Assistant Superintendent Neville Henderson
OSFI Superintendent Jeremy Rudin said, “We get a lot of questions around AI
has spoken about how data science can and what we are to do about it.” He added,
and should be applied to improve the “We really are a trust and verify regulator. We
supervision of financial institutions. He believe you. It’s a great system, I am sure, but
believes that solvency supervision that relies you have got to show us…we have to have
predominantly on judgment and experience an independent review to satisfy us that
can be improved by adding learning from the the system does what it says it does and it’s
application of data science. He noted that accurate and companies don’t do something
“my colleagues and I already do considerable that could be systemically discriminatory.”54
data analysis. By combining increased
computing power with more granular data,
we can increase the speed, accuracy, and Market conduct regulation
level of detail of our existing analysis.”51 The Canadian Council of Insurance
Rudin views data science as the opportunity Regulators (CCIR) has set out formal
to further develop current tools. He found expectations about the fair treatment of
that existing tools, however, are not set up to insurance customers.55 Insurance companies
explain why they predict what they predict: are ultimately responsible to ensure the fair
“I am a prudential regulator, I am not in the treatment of consumers. The guidance also
prediction business, I am in the prevention sets out expectations for brokers, MGAs, and
business…What I need to know is the most other industry intermediaries. Insurers and
likely causes of future failure…The algorithm intermediaries are obliged to ensure there
does not know that there is anything wrong is no conflict of interest by always putting
with these cases; it just knows that they consumer interests ahead of their own
stand out in some way. That said, these interests. Insurers are expected to disclose
cases might be good candidates for further relevant conditions, exclusions, restrictions,
investigation.”52 and fees of particular significance.
Assistant Superintendent Ben Gully spoke Work by the CCIR on fair treatment was
in early 2020 about OSFI’s supervision of independent of industry development toward
operational resilience and management of the application of AI and analytics. There was
non-financial risk by Canada’s banks. The convergence in the approaches developed by
discussion paper addresses supervision of the CCIR and those developed in Ontario,
artificial intelligence and machine learning but remaining differences warn that it can
models, cyber security, use of cloud services, be difficult to secure consensus on regulation
and the role of culture and conduct.53 of critical issues that will affect the adoption
Guidance about supervision of technology and acceptance of new technologies in the
insurance industry across Canada.

49 Zochodne, “What’s in the Black Box?”

50 Ibid.
51 Rudin, “Data Science and the Future of Financial Supervision.”
52 Ibid.
53 OSFI, “Developing financial sector resilience in a digital world: Selected themes in technology
and related risks.”
54 Meckbach, “Why solvency regulator is concerned about ‘black box’ underwriting.”
55 Canadian Council of Insurance Regulators, “Fair Treatment of Customers Working Group.”

38
 Six Critical Questions Emerging Issues Research Series

The CCIR’s 2020–2023 strategic plan is built around three strategic themes
in the insurance sector:56

• Technological innovation
• Climate change and natural catastrophes
• Conduct and culture

The CCIR observes with its first strategic Insurers should be able to explain to some
theme that “the need persists for appropriate consumers why price increases are required,
supervision of processes, products and for example, as the result of an at-fault
institutions in an evolving insurance collision or in response to the rising cost of
marketplace characterized by accelerating repairs for vehicles with sensors.
innovation, shifting consumer demands, new
In several provinces, auto insurers must
products and related vulnerabilities including
regularly file extensive analyses to secure
cybersecurity. Advancements in other
approval to adjust rates. This information
markets, such as development of autonomous
is processed by rate regulators, resulting in
vehicles and advances in bioscience, may alter
feedback to companies. It is unclear how
that traditional insurance product.”57
improving the capacity to assess risk will be
applied in jurisdictions with rate regulation.
Auto rate regulation Regulations about acceptable rating criteria
Analytics will improve the understanding limit the information that can be included
of the frequency and severity of collisions in filings. This interference may increase the
for Canadian drivers. Larger volumes of number of drivers forced into the Facility
data and increased capacity to process this Association, which limits the capacity of
information will improve the predictive insurers to reduce rates paid by low-risk
power of models. In jurisdictions without drivers. Ongoing discussion will hopefully
rate regulation this information will be used open pathways so drivers in jurisdictions
by insurers to improve the coverage provided with rate regulation can benefit from the new
to drivers. On renewal, low-risk drivers would analytical tools.
be offered similar coverage at a lower rate.

Recent experience in Alberta warns of industry challenges that

may occur with the increased use of analytics. The superintendent
of insurance in that province issued fines on 16 companies for
overcharging consumers. Media coverage sought information
about the size of the refunds provided to drivers and explanations
about why companies failed to comply with established rate
approvals. One company spoke about system errors, and another
identified a technical error.58

56 Canadian Council of Insurance Regulators, “Strategic Plan 2020–23,” page 6.

57 Ibid.
58 Labby, “16 Insurance Companies Fined More than $1.5 Million for Overcharging
Alberta Motorists.”

39
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

International guidance for big price would become evident for segments of
the population as a result of AI and analytics,
data analytics resulting in decreased consumer confidence
The IAIS set out advice for the regulation of in the insurance sector.
the use of big data analytics in the insurance
They also warn of the risk that consumers
industry.59 The direction set out by the IAIS
may be harmed when insurers make decisions
can influence practices in Canada, and the
based on data not provided directly by
IAIS reports at length about the risks for
consumers that may be incomplete or
insurance consumers of AI and analytics.
inaccurate. If customers are not aware these
They urge supervisors to learn about the
data are being used for insurance purposes,
benefits and risks presented by rapid
they would not have an opportunity to correct
advancements in analytics so they develop
inaccuracies. Supervisors are encouraged by
appropriate responses to promote and
the IAIS to consider “how best to mitigate any
encourage the industry to consistently deliver
potential prejudice to consumers as a result
fair outcomes for consumers. They advise
of the use of this type of alternate data for
regulators to consider independent audit
insurance purposes.”60
and integrity checks of machine learning and
algorithm processes to ensure fair outcomes
for insurance consumers.
International guidance for artificial
The IAIS warns that granular customization intelligence
may provide more product availability, but
The OECD has published principles for the
some customers may find themselves facing
use of AI.61 The IAIS also provided advice
higher premiums or not being offered
about the increasing use of digital technology
coverage at all. In Canada, for example,
in the insurance industry independent from
some condominium owners are presently
advice about analytics.62 Digital technology
adversely affected by the reduced availability
includes artificial intelligence, robo-advice
or affordability of insurance. The IAIS is
mechanisms, and other emerging tools.
concerned that disruption in availability or

Here are the OECD’s recommendations, which identify five

complementary values-based principles for the responsible
stewardship of trustworthy AI:

• AI should benefit people and the planet by driving inclusive

growth, sustainable development and well-being.
• AI systems should be designed in a way that respects the rule
of law, human rights, democratic values and diversity, and they
should include appropriate safeguards—for example, enabling
human intervention where necessary—to ensure a fair and
just society.

59 IAIS, “Issues Paper on the Use of Big Data Analytics in Insurance.”

60 Ibid., page 36.
61 OECD, “The Impact of Big Data and Artificial Intelligence (AI) in the Insurance Sector.”
62 IAIS, “Issues Paper on Increasing Digitalisation in Insurance and Its Potential Impact on
Consumer Outcomes.”

40
 Six Critical Questions Emerging Issues Research Series

• There should be transparency and responsible disclosure

around AI systems to ensure that people understand AI-based
outcomes and can challenge them.
• AI systems must function in a robust, secure and safe way
throughout their life cycles and potential risks should be
continually assessed and managed.
• Organisations and individuals developing, deploying or
operating AI systems should be held accountable for their
proper functioning in line with the above principles.63

In addition, the OECD put forth seven key requirements that AI systems
should meet in order to be deemed trustworthy:

• Human agency and oversight: AI systems should empower

human beings, allowing them to make informed decisions and
fostering their fundamental rights. At the same time, proper
oversight mechanisms need to be ensured, which can be
achieved through human-in-the-loop, human-on-the-loop, and
human-in-command approaches.
• Technical robustness and safety: AI systems need to be resilient
and secure. They need to be safe, ensuring a fall back plan in
case something goes wrong, as well as being accurate, reliable
and reproducible. That is the only way to ensure that also
unintentional harm can be minimized and prevented.
• Privacy and data governance: besides ensuring full respect
for privacy and data protection, adequate data governance
mechanisms must also be ensured, taking into account the
quality and integrity of the data, and ensuring legitimised
access to data.
• Transparency: the data, system and AI business models should
be transparent. Traceability mechanisms can help [in] achieving
this. Moreover, AI systems and their decisions should be
explained in a manner adapted to the stakeholder concerned.
Humans need to be aware that they are interacting with an AI
system, and must be informed of the system’s capabilities and
limitations.

63 OECD, “The Impact of Big Data and Artificial Intelligence (AI) in the Insurance Sector,” page 20.

41
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

• Diversity, non-discrimination and fairness: Unfair bias must be

avoided, as it could have multiple negative implications, from
the marginalization of vulnerable groups, to the exacerbation
of prejudice and discrimination. Fostering diversity, AI systems
should be accessible to all, regardless of any disability, and
involve relevant stakeholders throughout their entire life circle.
• Societal and environmental well-being: AI systems should
benefit all human beings, including future generations.
It must hence be ensured that they are sustainable and
environmentally friendly. Moreover, they should take into
account the environment, including other living beings, and
their social and societal impact should be carefully considered.
• Accountability: Mechanisms should be put in place to ensure
responsibility and accountability for AI systems and their
outcomes. Auditability, which enables the assessment of
algorithms, data and design processes, plays a key role therein,
especially in critical applications. Moreover, adequate and
accessible redress should be ensured.64

The IAIS recognizes the potential for and how this is catered for in the insurers’
digitization to improve the outcomes for risk management framework.”66
insurance consumers and reduce insurance
They advise regulators to “develop new tools
costs. Nevertheless, the focus of the advice
and skills for supervision of digital insurers,
is on supervision to prevent the unfair
enhanced cooperation with financial and other
treatment of consumers in the design,
authorities, safeguarding the supervisory
underwriting, pricing, marketing, claims
perimeter to prevent regulatory arbitrage and
resolution, and ongoing management of
enhancing information security.”67
consumers. “Supervisors will need to balance
the risks of new innovations against the Finally, they encourage supervisors
benefits for policyholders and the insurance to “consider establishing guidelines for
sector as a whole.”65 appropriate and responsible use of new
technologies to safeguard the fair treatment
Supervisors are encouraged to develop a
of customers and—for example in the use of
thorough understanding of how innovation
AI and robo advice mechanisms—promote
works and is applied “to ensure proper
advice and services that are suitable and
assessment of new product business
affordable for the consumer.”68
models, and the design and functioning of
IT architecture, infrastructures and processes,

64 OECD, “The Impact of Big Data and Artificial Intelligence (AI) in the Insurance Sector,” page 21.
65 IAIS, “Issues Paper on Increasing Digitalisation in Insurance and Its Potential Impact on
Consumer Outcomes.”
66 Ibid.
67 Ibid.
68 Ibid.

42
 Six Critical Questions Emerging Issues Research Series

Transition to a digital world Regulatory arbitrage these requirements. Transition to a digital

world increases the risk that insurance
increases the risk that insurance Some emerging products have effects
consumers misunderstand the regulatory
consumers misunderstand the similar to insurance but fall outside of the
protection in place.
jurisdictional powers of insurance regulators.
regulatory protection in place. In these cases, consumers would not be
provided with market conduct and prudential
Options for self-regulation
protection. If, for example, there is no In 2019, the Chartered Insurance Institute
assessment seeking to identify or measure based in the United Kingdom published
damage nor a promise of indemnity, this “Digital Ethics: A Companion to the Code of
product is not insurance. Nevertheless, Ethics.”69 The code sets out obligations and
it may be presented to homeowners as responsibilities for insurance professionals
insurance, and consumers may have an as they consider issues that include artificial
expectation that they will be protected like intelligence and big data analytics.
someone purchasing coverage from a licenced
There is scope for the insurance industry in
insurance provider.

Some guidance set out by the institute includes:

• What would someone from outside of the profession think of

what I am doing?
• Reflect not only on what you “could do” but what
you “should do”
• The difference between the two can be where ethics lies and
trust is built
• Put the best interests of each client at the heart of the
decisions you take
• Reflect on the different perspectives of fairness that exist

Regulatory arbitrage may be definitional, as Canada to provide leadership to promote the

illustrated above, or jurisdictional. Some acceptance and adoption of new technologies
consumers in Canada purchase insurance to better serve consumers. A credible strategy
coverage that is available through the Internet must acknowledge and focus on resolving
but not authorized locally. Consumers may the risks facing consumers. For example, in
be unaware that products available for direct response to growing concerns by regulators
purchase are not sanctioned where they reside. and consumers about the use of credit scores
Providers, however, should understand by insurers, the members of the Insurance
the risks they accept and deny coverage in Bureau of Canada developed a code of
unapproved jurisdictions, but regulatory conduct to provide guidance to the industry.
oversight will be difficult. These challenges The guide discusses consumer consent, use of
increase for consumers who do not have current information, a commitment to protect
support from a broker or agent. Moreover, confidential information, and handling of
some fintech start-ups may be unaware of disputes. Moreover, the guide includes a
existing regulations or, as boldly modelled commitment that an insurer will not deny,
by Uber, may choose to purposely ignore cancel, or decline renewal of a policy solely
on the basis of credit information.

69 Chartered Insurance Institute, “Digital Ethics: A Companion to the Code of Ethics.”

43
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

Lessons for the insurance industry The regulatory environment for the insurance
industry to introduce AI and analytics is
Insurance regulators across Canada are
unclear, likely to change, and may include
expected to become increasingly involved in
conflicting expectations. The tone set by
the supervision of the application of artificial
the International Association of Insurance
intelligence and analytics in the insurance
Regulators and specific regulators, like the
industry. OSFI, AMF, and BCFSA are likely
Financial Services Authority in the UK,
to coordinate their efforts to enhance
implies considerable challenges for those in
prudential oversight and supervision of
the insurance industry looking to embrace AI
operational risk. Members of the Canadian
and big data analytics. Regulators would be

There will be a risk of conflicting expectations for market conduct and

privacy regulation, and possibly between prudential and cyber security
regulation. The industry will need to be proactive to champion the value
of consistent supervisory expectations.
Council of Insurance Regulators will work
expected to play a less-intrusive role if the
to coordinate change in market conduct
industry demonstrates that recognized risks
regulation, but differences in approach
are effectively managed and consumers are
may become evident across the country.
confident that they will be treated fairly. It
Oversight of privacy, cyber security, and
is important that the Canadian insurance
rate regulation are expected to remain largely
industry get ahead of potential regulatory and
independent from prudential and market
consumer issues, perhaps by establishing a
conduct regulation, unfortunately with few
forum where these concerns can be discussed
examples of cooperation and coordination.
and resolved. Industry leaders committed
There will be a risk of conflicting expectations
to technological innovation will want to
for market conduct and privacy regulation,
support awareness of local and international
and possibly between prudential and cyber
initiatives working to advance the benefits of
security regulation. The industry will need
responsible AI and analytics and organizations
to be proactive to champion the value of
like the Future of Life Institute that support
consistent supervisory expectations.
advances in AI ethics.
Growing regulator interest and action may
Application of AI and analytics will improve
slow the pace and scope of industry innovation.
the operations of insurance supervisors.
Regulators are technologically agnostic, but
Regulators will need to compete with
also frequently cautious about change. Their
insurance companies to secure skilled
first priority is to ensure solvency and sound
analysts. Analytics can be applied to better
market conduct. Insurance regulators seek to
identify risky behaviour and trigger early
protect consumers from risk and are often
regulator intervention. These tools can
skeptical of change, including new technology.
also be taught to provide assurance when
They seek proof that new approaches will not
company practices are sound and less
harm consumers—proof that may be difficult
intrusive oversight is warranted. Increased
to demonstrate. The industry will need to
use of the emerging tools by regulators will
continue to champion innovation and the
improve regulation, better serve consumers,
benefits to consumers. Regulatory sandboxes
and build confidence in the financial health
and innovation hubs are approaches that
of the insurance industry.
have been applied successfully in other
jurisdictions.

44
 Explainable AI and big data analytics, and acknowledge the legitimacy of other
Six critical questions implemented responsibly, hold great
potential to better serve insurance
perspectives of fairness.

Why must insurers focus consumers. But will the industry get this
The application of artificial intelligence
and big data analytics increases the ability
on better outcomes right? Industry experience dealing with
of the insurance industry to quantify risk
for consumers? millions of customers over many decades
of loss at a granular level. Pricing and
finds scope for improvement. An industry
coverage decisions will be informed by
disrupter speaks about re-establishing
better knowledge about risk, increasing the
“We had three big ideas at insurance as a social good instead of a
industry’s capacity to ensure actuarially or
Amazon that we’ve stuck with necessary evil.70 Insurance is viewed by
statistically fair treatment of consumers. Skill
some as unpopular and needlessly complex.
for 18 years, and they’re the Some consumers, regulators, and others do
to quantify the risk of loss, statistically fair
pricing, and availability of coverage across
reason we’re successful. Put the not understand decisions made by insurance
society is a powerful mechanism to help
customer first. Invent. And be companies, adjusters, agents, and brokers
society manage risk. Will implementation
when they offer coverage or resolve claims.
patient.” This may be evident from the customer’s
of artificial intelligence and analytics
result in better outcomes for insurance
first experience of learning that there may
consumers? Will improvement be expected
— jeff bezos be large, inexplicable differences in the price
for some consumers or for everyone? Is
various insurers charge to assume their risk.
there a company or industry plan to support
Unexpected changes in pricing or conflicts
customers that may be worse off? The risks
in resolving a claim can erode consumer
are found in the details. It will be important
trust, a serious concern for an industry
that a focus on consumers is applied to
founded on trust.
every aspect of the many specific elements
Concern about trust in institutions, including of the transition.
insurance companies, is expected to be
magnified in the digital age.71 Insurance A proud history
professionals are willing to commit to the fair
Insurance exists to serve the risk management
treatment of insurance customers, supported
needs of society. The insurance industry has
by the introduction of tools with greater
much to be proud of. Millions of Canadians
capacity to assess the individual risk of loss.
have been made whole following a loss. Risks
There are, however, different perspectives
for participants in a group are pooled. Most
about what is fair. The important debates
participants will not experience a loss, but
about the fair treatment of customers often
the revenue in the pool must be sufficient to
involve conflicting views of fairness. These
pay for the losses that some are expected to
conflicts are likely to grow in importance
incur, including a few losses that may be large.
as the financial sector increasingly presses
Assessing a fair premium is dependent on the
forward with the introduction of powerful
expected frequency and severity of losses.
tools based on actuarial and statistical
Insurance is the promise to participating
fairness. Insurance leaders must understand

The application of artificial intelligence and big data analytics increases

the ability of the insurance industry to quantify risk of loss at a
granular level.

70 Azhar, “Disrupting the Insurance Industry with AI.”

71 Kovacs, “Sharing Economy: Implications for the Insurance Industry in Canada.”

45
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

consumers that they will be made whole if in participation in a pool. Capacity in the
they experience a loss. industry to measure and assess individual
risks was very limited. Coverage included
Risk pooling is a powerful mechanism.
fire protection services through insurance
Widespread use of insurance has accelerated
company fire brigades that would respond
its development in advanced countries
and suppress fires.
around the world, including Canada. It is
hard to imagine a world without insurance Over time, insurance pooling concepts
to help society and decision makers manage supporting marine and fire insurance were
critical risks. How would society cope with applied to the development of additional
the risk of ruin from catastrophic loss? The coverage to serve consumers. This included
expectation of support from the insurance auto, business, and other insurance products.
industry is embedded in decision making Modest fees paid by many participating in
across society. a group were pooled to compensate the
losses experienced by a few. This was

The insurance industry has a proud and distinguished history of providing

consumers with protection across a growing number of perils through
the provision of pools and the active championing of risk reduction.
a sound business model for insurance
Informal risk pools were present in prehistory.
companies and welcome protection for
If a home were destroyed by fire, for example,
risk-averse consumers facing a range of
it was understood that neighbours would
risks. The insurance industry has a proud and
work together to rebuild the structure and
distinguished history of providing consumers
help replace the contents. When passing
with protection across a growing number of
through difficult waters, merchants would
perils through the provision of pools and the
redistribute their cargo across several ships
active championing of risk reduction.
to limit the risk of catastrophic loss if a
single vessel should capsize. These concepts
provide the foundation for property and Classification of risks
marine insurance.
The pooling of risk for a group of customers
More formal risk pools emerged over time. begins with the identification of individuals
Early merchants seeking to borrow funds with similar risk of loss to be included in
to finance trading voyages to distant lands the group—that is, risk classification. Past
secured loans that were bundled with experience for the group is coupled with an
a guarantee that a repayment would be analysis of trends to anticipate future losses
cancelled if a shipment were lost at sea. and establish actuarially fair pricing. The
Additional charges for loans to support expected average experience for the group
commerce by sea relative to goods carried is applied to the specific circumstance of
by land were condemned by Pope Gregory individual risks. Big data analytics increases
IX in 1236 as usury, so practices further the capacity for insurers to anticipate the
evolved toward those found in modern risk of loss.
marine insurance.
Each customer is unique. Concepts like the
Current property insurance practices law of large numbers finds overwhelming
emerged following the Great Fire of London uncertainty in small groups and robust
in 1666. Early fire insurance pools focused on outcomes for larger groups. Insurance
the identification of groups with similar risk actuaries find that “to the extent that the risk
of loss, typically property owners located in classification process attempts to identify
the same community, who were interested

46
 Six Critical Questions Emerging Issues Research Series

and measure every characteristic, it becomes American, severely limiting the availability
unworkable.”72 Nevertheless, it is important of loans and insurance coverage to people
that insurance practices take into account in these communities.75 The use of age and
some of the known differences in individual gender as rating criteria by auto insurers in
risk characteristics. The difficulty in risk Canada has been challenged many times over
classification involves determining a fair the past 40 years, including a Supreme Court
balance. The Casualty Society of Actuaries ruling urging the industry to develop data
observes that “most of the controversy that would support the application of non
surrounding risk classification involves discriminatory criteria. Use of credit scores
where the lines are drawn.”73 is a more recent controversy for the industry.

Actuaries have established five principles to achieve these objectives:

• The system should reflect expected cost differences.

• The system should distinguish among risks on the basis of
relevant cost-related factors.
• The system should be applied objectively.
• The system should be practical and cost effective.
• The system should be acceptable to the public.74

The objectives of risk classification are to Joi Ito argues that “the original idea of risk
ensure financial soundness, be fair, and spreading and the principle of solidarity
provide incentives for risk reduction was based on the notion that sharing risk
that result in the widespread availability bound people together, encouraging a
of coverage. spirit of mutual aid and interdependence.
By the final decades of the 20th century,
The insurance industry should anticipate
however, this vision had given way to the
and welcome public debate that serves to
so called actuarial fairness.”76 He goes on
determine acceptable industry practices
to state that “computer scientists of today
for risk classification. For example, banks
are more sophisticated in many ways
and insurers in the United States were
than the actuaries of yore, and they often
confronted in the 1960s about practices
sincerely are trying to build algorithms that
that resulted in the redlining of high-risk
are fair. The new literature on algorithmic
communities that were largely African
fairness usually doesn’t equate fairness with

“The original idea of risk spreading and the principle of solidarity

was based on the notion that sharing risk bound people together,
encouraging a spirit of mutual aid and interdependence. By the final
decades of the 20th century, however, this vision had given way to the
so called actuarial fairness.”76

72 American Academy of Actuaries Committee on Risk Classification, “Risk Classification

Statement of Principles,” page 1.
73 Ibid.
74 Ibid., page 5.
75 Ito, “Supposedly ‘Fair’ Algorithms Can Perpetuate Discrimination.”
76 Ibid.

47
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

accuracy, but instead defines various trade Insurance is a competitive business.

offs between fairness and accuracy.” Companies are driven to develop the capacity
to better identify and assess the risk of loss.
Access to enormous datasets and emerging
Leaders who demonstrate a superior capacity
data science tools is shifting the capacity
to understand, select, and properly price risk
of the industry to reclassify individual
can successfully sustain profitable growth.
risks. Consideration of change should be
accompanied by open discussion about the AI and big data analytics will further improve
acceptability of evolving insurance practices. industry capacity to assess the risk of loss.
Industry practices can be adapted to achieve This will benefit customers with low risk.
broader social goals when clear direction is A fair premium for low-risk consumers to
provided. Insurance practices must be fair participate in a pool is reduced when high-
and be seen to be fair. risk consumers contribute more or stop
participating. Insurance companies can
Consumers’ views of fairness offer low-risk consumers the same level of
protection at a lower price, and this change is
Risk classification is one of many factors sustainable if a new assessment tool is proven
that affect consumer perceptions about the to be superior in assessing risk.
insurance industry and fairness. Claims
resolution, distribution, marketing, and A new approach that benefits lower-risk
many other practices influence consumers. consumers will invariably harm those
Consumers will judge practices and changes members in a group with higher risk. These
across the entire system. higher-risk drivers, property owners, and
businesses must pay more to secure the same
For example, in 1992 Hurricane Andrew level of coverage. Indeed, some insurers may
resulted in damage claims in Florida that choose to withdraw the offer of coverage.
significantly exceeded industry expectations. Customers newly identified with higher risk
Some insurance companies became insolvent. will not be pleased with the change. They may
A new understanding of the risk of hurricane be more understanding if the change results
damage claims resulted in several insurers from a clearly identified action where they
withdrawing from the market while the rest were responsible, like an at-fault collision.
sought significant price increases. Consumer Also, they will appreciate it if there is a clear
frustration and anger about this disruption path to re-establish themselves as lower risk.
has been intense for more than 25 years. Consumer satisfaction increases when they
Debate about the fair treatment of insurance are able to control the relationship with their
customers often focused on ensuring that behaviour and practices.
affordable coverage would be available.
Regulations were introduced detailing steps Consumers deserve an explanation if the
to minimize the disruption for consumers if assessed risk of loss has increased. Vehicle
insurers withdraw from the market. owners are required by governments to

A new approach that benefits lower-risk consumers will invariably harm

those members in a group with higher risk. These higher-risk drivers,
property owners, and businesses must pay more to secure the same
level of coverage.

48
 Six Critical Questions Emerging Issues Research Series

purchase insurance, and property owners are with public authorities about how to best
required to purchase insurance by lenders. address the needs of those with high risk
The universality of insurance differs from where affordable private residential flood
most industries. Consumers must find insurance is difficult to introduce.
insurance coverage, and they will want to
Many years ago, improving capacity to
understand change in the assessment of their
identify drivers with a higher risk of loss
risk of loss.
resulted in concerns about the affordability
The insurance industry in Canada is subject and availability of coverage for high-
to extensive regulation. This includes risk drivers. This a predictable outcome
prudential supervision to assess capacity of company and public policy efforts to
to pay claims, governance practices, and identify and reward lower-risk drivers. It
solvency risk management. Importantly, was important that the industry was able to
the insurance industry is also subject to partner with policymakers and regulators
regulation of market conduct, including to develop risk pools to support vehicle
a focus on the acceptability of industry owners who temporarily did not qualify for
practices. Market conduct regulation seeks mainstream coverage.
to assure consumers and other stakeholders
Raj Shroff concludes that “AI has the potential
that industry practices are transparent and
to transform the insurance experience for
fair. Independent supervision of industry
consumers from frustrating and bureaucratic
practices by regulators is critical to building
to something fast, on-demand and more
consumer trust.
affordable. Tailor-made insurance products

Market conduct regulation seeks to assure consumers and other

stakeholders that industry practices are transparent and fair. Independent
supervision of industry practices by regulators is critical to building
consumer trust.

Public acceptability of tools that better will attract more customers at fairer
identify and benefit those with low risk prices…We will see insurance become more
must include consideration of the treatment personalized, because insurers using AI
of those with high risk. This mindset was tech will be able to understand better what
adopted by the insurance industry in their customers need…Most importantly,
Canada with the introduction of overland the AI solutions described above can make
residential flood insurance. Circumstances it easier for customers to interact with
for homeowners with low to moderate risk insurance companies.”77
of overland flooding were improved when
private flood insurance was introduced and
soon became widely available. This was
accompanied by the launch of discussions

77 Schroff, “How Are Insurance Companies Implementing Artificial Intelligence (AI)?”

49
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

Fraud Prevention
Early fraud detection

Operations Customer Analysis

Saving time
AI in Insurance Tailored & cheaper insurance

Claims Management
Faster claims processing

Source: Schroff, “How Are Insurance Companies Implementing Artificial Intelligence (AI)?”

Lessons for the insurance industry The near universal use of insurance by drivers,
homeowners, and businesses means that
The insurance industry exists to serve its inevitable changes from the application of AI
current and future customers. Changes that and analytics adopted by the industry will be
will become possible over the next 10 years judged by consumers, regulators, and others.
because of the development of artificial The industry should encourage and support
intelligence and big data analytics should informed debate about evolving insurance
be assessed by an industry committed practices. The industry has demonstrated
to introducing changes that will benefit great capacity to adapt its practices if and
consumers. Do not introduce changes simply when clear direction is provided.
because they are possible; make changes that
are expected to benefit consumers. Risk pools, statistically fair pricing of risk,
and other benefits of insurance have served
Increased capacity to assess risks at a more Canadians well. The insurance industry has
granular level will benefit some members of much to be proud of as it works to protect
a group and harm others in the pool. How drivers, homeowners, and businesses.
will the benefits be shared and explained to Demonstration that emerging AI and big
those found to be at lower risk? What is the data analytics tools are used to benefit
plan to support those found to be at higher consumers must acknowledge and work to
risk? How was public acceptability assessed address alternate views of fairness. Third-
for new company and industry practices? party verification may demonstrate increased
Has the industry explained the changes industry capacity to achieve statistical and
introduced by AI and big data analytics to actuarial fairness, but a broad perspective of
insurance regulators? fairness is critical to building consumer trust.

The near universal use of insurance by drivers, homeowners, and

businesses means that inevitable changes from the application of AI
and analytics adopted by the industry will be judged by consumers,
regulators, and others.

50
 Recommendations for the Insurance Industry Emerging Issues Research Series

Recommendations The insurance landscape is changing,

empowered by growing access to digital
thinking and innovative insurers will use
AI and big data to make insurance better
for the insurance information. Artificial intelligence and big for Canadians.

industry data analytics are among the most powerful

innovations that will affect the insurance
As the insurance industry embraces the
power of this new technology it is important
industry over the next 10 years. They can
to reflect on the risks of adoption and
“I’ve learned that people will organize billions of data points into clear
acceptance. Consumers lack clarity about
insights, create personalized experiences,
forget what you said, people will anticipate needs, and better manage risks
artificial intelligence and big data analytics,
what they can do, and where opportunities
forget what you did, but people facing drivers, homeowners, and businesses.
and risks exist. Demystifying the issues is
will never forget how you made When used responsibly these tools will
a step toward responsible implementation.
improve lives and solve problems. Forward
them feel.”
Here are eight recommendations for the insurance industry in Canada
— maya angelou related to AI and big data analytics:

1. Inform consumers: Insurance consumers need to understand

how the new tools will affect them. Big data analytics will
increase the capacity of the insurance industry to assess
individual risks at a granular level and this will benefit those with
low risk of loss. The industry must be proactive to demonstrate
that application of the new algorithms are statistically fair and
unbiased in the treatment of all consumers, including planning
for the needs of consumers that may face higher prices and
reduced availability of coverage options.
2. Embrace innovation: Seek out opportunities to change
traditional approaches to insurance. Celebrate the industry’s
proud history while encouraging new practices across the
organization. New tools will support actions that were not
possible in the past, and this is good for your company and for
consumers. Think of AI and big data analytics not as business
tools but as a commitment to innovation and modernization.
3. Be prepared for uncertain regulation: Deployment of artificial
intelligence and big data analytics introduces many questions
about the fair treatment of consumers, exclusion of coverage,
and difficulty in obtaining a quote. It is the responsibility of
insurance regulators to ask questions and provide guidance
through this transition. The insurance industry should be
prepared for detailed regulatory monitoring of change and for
challenges to practices that cannot be explained. Transparent,
open discussion is in everyone’s best interests.

51
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

4. Create new insurance programs: Big data analytics can support

the introduction of coverage for risks that were previously
not insurable, like residential flood insurance. Demand can be
created for new products in situations where consumers may
be unaware of the risks they face. There is scope, for example,
for the insurance industry to adapt coverages provided to large
businesses and homeowners to better serve small businesses
and tenants and to expand cyber insurance coverages.
5. Doing it right: Actions sincerely seeking to improve outcomes
for policyholders will be supported by consumers. Do not
be afraid to fail at first, as learning from the experience will
move you toward a better solution. It will likely be more
important to bring forward the right changes than being first
to market. Consumers will likely understand if you are willing to
acknowledge shortcomings and to move forward, learning to
provide better solutions.
6. Changing consumers: Insurers need to demonstrate their
understanding of the growing importance of environmental,
equity, and social issues for consumers. Consumers are also
increasingly comfortable with modern technology. Insurance
has successfully adapted to extraordinary societal changes
over more than 200 years in Canada. Insurers know how to
adjust; they just need to become comfortable with doing
so much faster.
7. Invest in new technology: Legacy systems can be a major
stumbling block for insurers trying to embrace artificial
intelligence and big data analytics. Leveraging digital process
automation platforms can insulate the front office from the
back office and provide the straight-through processing,
flexibility, and digital capabilities insurers need today to be
competitive and better serve consumers.

52
 Recommendations for the Insurance Industry Emerging Issues Research Series

8. Accept different views of fairness: The insurance industry

should be proud of the benefits to society of providing risk
pools supported by systems that result in actuarially fair pricing.
These approaches support risk management needs for drivers,
homeowners, and businesses. Nevertheless, the industry
must be prepared to participate in discussions about socially
acceptable classification of risk where some participants adopt
a perspective on the fair treatment of consumers that differs
from an unfettered commitment to actuarial fairness.
There is no silver bullet solution that is analytics. An insurer’s product or investment
certain to work across the insurance industry selections must be made with the main
in support of the introduction of machine goal of supporting and extending its core
learning, artificial intelligence, and big data business strategy.

Insurers should focus on the following:

• Data quality and management: Insurers must actively manage
data governance and quality for structured and unstructured
data used to support decisions and operations.
• Artificial intelligence: AI can develop a more personal
experience for policyholders. AI can give insurers greater ability
to control and learn from client interactions. AI can also manage
books of business by continuously monitoring performance.
• Digital automation platforms: Insurers need to select specific
platforms, tools, and applications that enable digital processing,
interface tools, and apps. Insurers are behind other industries in
this transition and can learn from sectors that include banking,
retail, and telecommunications.
• Omni-channel experience: Insurers need to interact seamlessly
with the insured, broker, or adjuster on whatever device or
app the user chooses. Whatever functions an insurer provides
need to work across channels and enable users to seamlessly
switch channels.
• AI for decision support: AI can do wonders to assist in
many judgment calls that happen every hour of every day in
insurance, like suggesting who is well-suited to respond to
a new claim, choosing the best vendor to repair a damaged
vehicle or home, or suggesting a repair flow or replacement.

53
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

Leading companies are pushing through these unlikely in the near term, providing some
challenges to establish a better understanding time to adjust. Nevertheless, profound
of the risks facing consumers and the coverage change is underway. Artificial intelligence
that can be provided. Informed insurers grow and big data analytics will bring benefits for
more confident in the products and services insurance consumers if companies deploy
they offer when supported by analytics. Data them with sufficient diligence, prudence,
science and analytics provide a foundation and care. The insurance industry should be
for sound decisions about product design, excited by the opportunity to responsibly
pricing, and willingness to offer coverage. bring new ideas and approaches forward to
better serve the risk management needs of
Significant change is inevitable for the
Canadian consumers.
insurance industry. Full-scale disruption is

54
 Appendix: AI Terminology Emerging Issues Research Series

Appendix: AI terminology
Term Description
Algorithms A list of computer-implementable instructions.
Machine A collection of programs and processes that enable a machine (eg. computers) to apply data and
intelligence information to solve problems.
Conventional and A basic form of MI (eg. generalised linear models). These rely on assumptions to understand how variables
curve fitting relate to each other, with the aim of creating a curve that best-fits the relationship between data points.
Conventional curve fitting can capture some types of non-linear relationships, also.
Machine learning (ML) Algorithms that learn from data and analyse more complex, inter-related and non-linear relationships
among variables. Commonly used in classification, regression and pattern recognition.
Artificial intelligence AI goes beyond ML by facilitating adaptive application of understanding. With these algorithms, machines can
store and apply learning flexibly, including to contexts not originally intended.
Supervised learning To train a machine using data which are labelled, ie. already tagged with the correct answer. These labelled
data act as supervisor. The machine infers relationships from this sample, which are used to map
new examples.
Unsupervised learning Unsupervised learning is used when labelled data are not available. With no teacher to train the machine, it
has to discover hidden structures in the unlabelled data on its own. Used for clustering and association.
Clustering A clustering algorithm seeks to discover inherent groupings in the data, eg, grouping policyholders by
and association purchasing behaviour. An association problem is when an insurer looks to find out rules that describe the
data, eg. policyholders that buy X policy also tend to buy Y policy.
Reinforcement Goal-oriented algorithms (agents), which answer the question, how can this be optimised? Eg. how can
learning marketing investment be optimised to extract maximym ROI? Learns by interacting with its environment.
Ensemble learning Uses multiple algorithms in combination to obtain better predictive performance than could be obtained
from any one of the algorithms alone.
Data engineering Data engineering is the process of collection, curating, storing, and transforming data for
analytical purposes.
Deep learning Imitates the human brain to learn without human supervision, with data that is unstructured
and unlabelled.
False positive A prediction which wrongly indicates that a particular condition or attribute is present.
False negative A prediction which wrongly indicates that a particular condition or attribute is absent.
Physics-based ML Machine learning that incorporates a model (eg, hydrodynamic) built using a valid scientific theory based on
physical systems understanding into an ML algorithm/process to provide more structure to the model than
would be the case for a less constrained ML model (eg, supervised or unsupervised learning.) This hybrid
approach is often easier to interpret and diagnose.
Generative Generative adversarial networks (GANs) involve learning patterns in data so that the model can generate
adversarial networks new examples that seem credible enough to belong to the original dataset. The original data and the
generated data can then be played off each other in the context of competing neural networks to develop
better models.
Causal inference Causal inference in ML refers to approaches that provide more structure for control and prediction by
building capabilities that identify actual drivers of outcomes to make an ML process more robust to changing
circumstances, eg, attempting to sort out causal drivers of obesity to distinguish what can be controlled
across different sub-populations or analysing what design choices lead to more clicks on a website.

Source: Swiss Re Institute, sigma No 5/2020, page 2.

55
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

Bibliography
ACM. “Fathers of the Deep Learning Revolution Receive ACM A.M. Turing Award.” (2018). https://awards.acm.org/about/2018-turing.

American Academy of Actuaries Committee on Risk Classification. “Risk Classification Statement of Principles.” (2014). http://www.
actuarialstandardsboard.org/wp-content/uploads/2014/07/riskclassificationSOP.pdf.

Anchen, Jonathan, Jeffrey Bohn, and Rajeev Sharan. “Machine Intelligence in Insurance: Insights for End-to-End Enterprise Transformation.”
Swiss Re, Sigma No. 5 (2020). https://www.swissre.com/dam/jcr:88e529ca-f0c9-46fc-95fd-0bc41c7655b8/sigma5_2020_en.pdf.

Anyoha, Rockwell. “The History of Artificial Intelligence.” Science in the News (August 28, 2017). http://sitn.hms.harvard.edu/flash/2017/
history-artificial-intelligence.

Azhar, Azeem. “Disrupting the Insurance Industry with AI.” Exponential View podcast (August 14, 2019). https://hbr.org/podcast/2019/08/
disrupting-the-insurance-industry-with-ai.

Balasubramanian, Ramnath, Ari Libarikian, and Doug McElhaney. “Insurance 2030—The Impact of AI on the Future of
Insurance.” McKinsey & Company (April 30, 2018). https://www.mckinsey.com/industries/financial-services/our-insights/
insurance-2030-the-impact-of-ai-on-the-future-of-insurance.

Buolamwini, Joy and Timnit Gebru, “Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification.” Proceedings
of Machine Learning Research (2018). http://proceedings.mlr.press/v81/buolamwini18a/buolamwini18a.pdf.

Canadian Council of Insurance Regulators. ”Fair Treatment of Customers Working Group.” (2018). https://www.ccir-ccrra.org/
FairTreatmentofCustomersWorkingGroup.

Canadian Council of Insurance Regulators. “Strategic Plan 2020–23.” (2020). https://www.ccir-ccrra.org/Documents/View/3579.

Carrier Management. “What Warren Buffett Said about AI.” (June 2, 2017). https://www.carriermanagement.com/news/2017/06/02/167765.htm.

CDEI. “AI and Personal Insurance.” Snapshot Series, Centre for Data Ethics and Innovation (2019). https://apo.org.au/sites/default/files/
resource-files/2019-09/apo-nid267086.pdf.

Chartered Insurance Institute. “Digital Ethics: A Companion to the Code of Ethics.” (2019). https://www.cii.co.uk/media/10122096/digital-
ethics-companion.pdf.

Contant, Jason. “Focus on These Three Areas When Developing AI Best Practices.” Canadian Underwriter (September 27, 2019). https://
www.canadianunderwriter.ca/insurance/focus-on-these-three-areas-when-developing-ai-best-practices-1004168445.

Faggella, Daniel. “Artificial Intelligence in Insurance—Three Trends That Matter.” Emerj (March 14, 2020). https://emerj.com/
ai-sector-overviews/artificial-intelligence-in-insurance-trends.

Gomery, Sally, and Noleen John. “Gender Neutral.” Canadian Underwriter (April 1, 2011). https://www.canadianunderwriter.ca/features/
gender-neutral.

Hamilton, Isobel Asher. “Amazon Built an AI Tool to Hire People but Had to Shut It Down Because It Was Discriminating Against Women.” Business
Insider (October 10, 2018). https://www.businessinsider.com/amazon-built-ai-to-hire-people-discriminated-against-women-2018-10.

Hilbert, Martin, and Priscila Lopez. “The World’s Technological Capacity to Store, Communicate, and Compute Information.” Science
332, no. 6025 (April 2011): 60–65.

IAIS. ”Issues Paper on Increasing Digitalisation in Insurance and Its Potential Impact on Consumer Outcomes.” (2018). https://www.
iaisweb.org/page/supervisory-material/issues-papers.

IAIS. “Issues Paper on the Use of Big Data Analytics in Insurance.” (February 2020). https://www.iaisweb.org/page/supervisory-material/
issues-papers.

56
 Bibliography Emerging Issues Research Series

IIC. “Demographics of the P&C Insurance Industry in Canada 2017–2027.” (2018). https://www.insuranceinstitute.ca/en/resources/
insights-research/changing-workforce.

IT Chronicles. “What Is Big Data?” (2020). https://itchronicles.com/what-is-big-data.

Ito, Joi. “Supposedly ‘Fair’ Algorithms Can Perpetuate Discrimination.” Wired (February 5, 2019). https://www.wired.com/story/
ideas-joi-ito-insurance-algorithms.

Kovacs, Paul. “Climate Risks: Implications for the Insurance Industry.” Insurance Institute of Canada (2020). https://www.insuranceinstitute.
ca/en/resources/insights-research/Climate-risks-report.

Kovacs, Paul. “Cyber Risks 2015: Implications for the Insurance Industry in Canada.” Insurance Institute of Canada (2015). https://www.
insuranceinstitute.ca/en/resources/insights-research/cyber-risks.

Kovacs, Paul. “Cyber Risks 2019: Implications for the Insurance Industry in Canada.” Emerging Issues Research Series Update, Insurance
Institute of Canada (2019). https://www.insuranceinstitute.ca/en/resources/insights-research/cyber-risks-update.

Kovacs, Paul. “Sharing Economy: Implications for the Insurance Industry in Canada.” Insurance Institute of Canada (2017). https://www.
insuranceinstitute.ca/en/resources/insights-research/sharing-economy.

Kunreuther, Howard, and Patricia Grossi, eds. Catastrophe Modeling: A New Approach to Managing Risk. Boston: Springer Science &
Business Media, 2006.

Kurilyak, Slava. “Artificial Intelligence (AI) in Insurance.” Produvia (November 2017). https://blog.produvia.com/
artificial-intelligence-ai-in-insurance-d9035ea2d0b9.

Labby, Bryan. “16 Insurance Companies Fined More than $1.5 million for Overcharging Alberta Motorists.” CBC News (September 29,
2020). https://www.cbc.ca/news/canada/calgary/alberta-superintendent-auto-insurance-fines-1.5742140.

Laney, Doug. “3D Data Management: Controlling Data Volume, Velocity, and Variety.” Gartner Blog (February 2001). http://blogs.gartner.
com/doug-laney/files/2012/01/ad949-3D-Data-Management-Controlling-Data-Volume-Velocity-and-Variety.pdf.

Malik, Adam. “COVID-19 Has Accelerated Changes to the Claims Industry.” Canadian Underwriter (June 15, 2020). https://www.
canadianunderwriter.ca/claims/how-changes-to-auto-claims-have-affected-the-customer-journey-1004179534.

Mayhew, Daniel R., Herb M. Simpson, and Deanna Singhal. “Best Practices for Graduated Driver Licensing in Canada.” Traffic Injury Research
Foundation (2005). https://tirf.ca/wp-content/uploads/2017/02/Best_Practices_for_Graduated_Driver_Licensing_in_Canada.pdf.

Meckbach, Greg. “Why Solvency Regulator Is Concerned about ‘Black Box’ Underwriting.” Canadian Underwriter (December 4, 2019).
https://www.canadianunderwriter.ca/technology/why-solvency-regulator-is-concerned-about-black-box-underwriting-1004171387.

Microsoft. “Empowering the Digital Insurer.” Microsoft Perspectives (2020). https://info.microsoft.com/rs/157-GQE-382/images/Digital%20

Insurer%20-%20Microsoft%20Perspectives.pdf?aliId=198418163.

OECD. “The Impact of Big Data and Artificial Intelligence (AI) in the Insurance Sector.” (2020). http://www.oecd.org/finance/impact-
big-data-ai-in-the-insurance-sector.htm.

OSFI. “Developing Financial Sector Resilience in a Digital World: Selected Themes in Technology and Related Risks.” Discussion Paper
(2020). https://www.osfi-bsif.gc.ca/Eng/Docs/tchrsk.pdf.

PwC. “AI in Insurance: Hype or reality?” Top Issues (2016). https://www.the-digital-insurer.com/wp-content/uploads/2016/06/716-pwc-

top-issues-artificial-intelligence.pdf.

Raman, Jayant P., and Rachel Lam. “Artificial Intelligence Applications in Financial Services.” Oliver Wyman (2019). https://www.oliverwyman.
com/our-expertise/insights/2019/dec/artificial-intelligence-applications-in-financial-services.html.

57
AI and Big Data: Implications for the Insurance Industry in Canada Emerging Issues Research Series

Rudin, Jeremy. “Data Science and the Future of Financial Supervision.” Remarks by Superintendent Jeremy Rudin to the 11th Symposium
on Asian Banking and Finance. San Francisco, California, June 25, 2018. Government of Canada (2018). https://www.osfi-bsif.gc.ca/
Eng/osfi-bsif/med/sp-ds/Pages/jr20180625.aspx.

Schroff, Raj. “How Are Insurance Companies Implementing Artificial Intelligence (AI)?” Medium (July 1, 2019). https://towardsdatascience.
com/how-are-insurance-companies-implementing-artificial-intelligence-ai-aaf845fce6a7.

Smart Home America. “List of Mitigation Insurance Discounts and Tax Savings.” (2020). https://www.smarthomeamerica.org/fortified/
discounts-and-incentives/list-of-fortified-discounts-and-incentives.

Smola, Catherine. “How Artificial Intelligence Is Transforming the P&C Insurance Industry.” CSIO (2020). https://www.csio.com/article/
how-artificial-intelligence-transforming-pc-insurance-industry.

Swiss Re. “Data-Driven Insurance: Ready for the Next Frontier?” Sigma No. 1 (2020). https://www.swissre.com/dam/
jcr:430c9b00-09ee-448f-8307-0ca484a2ad96/sigma1_2020_en.pdf.

Tapestry Networks. “Three Ways AI Will Transform the Insurance Industry.” EY (August 23, 2019). https://www.ey.com/en_gl/insurance/
three-ways-ai-will-transform-the-insurance-industry.

TD. “Responsible AI in Financial Services: Enhancing Trust While Adopting New Capabilities.” (2019). https://newsroom.td.com/uploads/
media/default/0001/04/ccc16614ca2791cc7a6856ba71ef1cfaa5dc5d92.pdf.

Thomas, Mike. “25 AI Insurance Companies You Should Know.” Built In (April 2020). https://builtin.com/artificial-intelligence/ai-insurance.

UNEP FI. “PSI Principles for Sustainable Insurance.” UNEP Finance Initiative (June 2012). https://www.unepfi.org/psi/wp-content/
uploads/2012/06/PSI-document1.pdf.

UNEP FI. “UNEP FI Working with 16 Global Insurers to Better Understand Risk & Implement TCFD Recommendations.”
UNEP Finance Initiative (November 13, 2013). https://www.unepfi.org/news/industries/insurance/
unep-fi-working-with-16-global-insurers-to-better-understand-risk-implement-tcfd-recommendations.

UNESCO. “I’d blush if I could: closing gender divides in digital skills through education.” (2019). https://unesdoc.unesco.org/ark:/48223/
pf0000367416.page=1.

Zochodne, Geoff. “What’s in the Black Box? Rise of Artificial Intelligence in Banking Presents Challenge for OSFI.” Financial Post (February 7, 2020).
https://financialpost.com/news/fp-street/whats-in-the-black-box-rise-of-artificial-intelligence-in-banking-presents-challenge-for-osfi.e

58
 Emerging Issues Research Series

professional designations

Chartered Insurance Professional (CIP)

Fellow Chartered Insurance Professional (FCIP)

certificate programs

Advanced CIP (ACIP)

Acord Certified Expert
Adjusters’ Training & Education Series
Associate Customer Service
Canadian Managing General Agent (CMGA)
Commerical Insurance The Insurance Institute of Canada
General Insurance Essentials (GIE) 18 King Street East, 6th Floor
Instructor Certification Toronto, ON M5C 1C4
Rehabilitation Benefits Administration toll free:1-866-362-8585
Risk Management phone:416-362-8586
fax:416-362-1126
insurance licensing & ce credits
[email protected]
Brokers & Agents
Independent Adjusters
CE OnDemand

professional development

Business/Management
Insurance/Technical
Interpersonal/Communications

networking & career development

Seminars, Symposium, Special Events Connect the Industry

Career Connections Provides Introduction to Careers in Insurance
mycareer Maps Career Pathways in the Industry

industry insight & research

Demographics of the P&C Insurance Industry in Canada, 2017-2027

Emerging Issues Research Series
Cyber Risks: Implications for the Insurance Industry in Canada
Automated Vehicles: Implications for the Insurance Industry in Canada
Sharing Economy: Implications for the Insurance Industry in Canada
A Changing Workforce: Implications of Technological Disruption for the Insurance
Industry in Canada
Climate Risks: Implications for the Insurance Industry in Canada

Empowering insurance careers

insuranceinstitute.ca

59