“Investing in innovative technologies like generative AI is one of many avenues Definity is pursuing to achieve its goal of becoming a top-five P&C insurer in Canada”

Jeffrey Baer, Definity

“The number-one critique of any sort of analytical model is its ability to amplify systemic biases that are already prevalent in our society”

Elizabeth Bellefleur-MacCaul, Definity

“Ultimately, this comes down to people and partnership and making sure that everybody understands the weight of the responsibility of using these tools appropriately”

Neil Bunn, Google

In Partnership with

Executive insights – winning with AI

Traditional and generative AI presents new insurance opportunities. Definity and Google representatives share how to make AI a success, how to navigate ethical challenges, and how to get started.

Read on

Jeffrey Baer

Definity

Elizabeth Bellefleur-MacCaul

Definity

Neil Bunn

Google

Industry experts

Definity Financial Corporation is the parent company of some of Canada’s longest-standing and most innovative property and casualty insurance companies, including Economical Insurance, Sonnet Insurance, Family Insurance Solutions, and Petline Insurance. It is an insurance holding company that invests in multi-channel operations to provide Canadians with optimal coverage options for their homes, vehicles, businesses, farms, and pets.

Find out more

Neil Bunn is a professional engineer and the director of client engineering for Google’s strategic verticals in Canada. He leads multiple teams of client-facing engineers covering Google’s most important focus areas in Canada, including financial & insurance services, telecommunications, media, retail, top manufacturing, and select emerging technologies clients. He has been with Google for five years, focusing on building a team of the world’s best cloud and AI engineers.

Director of client engineering for Google’s strategic verticals in Canada

Neil Bunn

Elizabeth (Liz) Bellefleur-MacCaul is a senior actuarial analyst at Definity. Since joining the company in 2018, Liz has been responsible for the development of predictive models for the personal insurance book of business and has more recently joined Definity’s advanced analytics team. As a passionate advocate for ethical AI and data usage, Liz is an original and active member of Definity’s dedicated task force for addressing bias and fairness in predictive modelling and analytics. Additionally, she is the vice-chair of the company-wide women’s empowerment employee group.

Senior actuarial analyst at Definity

Elizabeth Bellefleur-MacCaul

With an over 10-year tenure at Definity and Economical Insurance, Jeffrey Baer, VP, enterprise analytics and data office, leads Definity’s 200-person analytics community of practice and is accountable for Definity’s enterprise analytics strategy and high-performing advanced analytics, business intelligence, and enterprise data office functions. Jeffrey is also the co-founder of the Waterloo Region chapter of Data for Good, an organization that brings together local data scientists to support non-profit organizations in making positive change within the community.

VP, enterprise analytics and data office at Definity

Jeffrey Baer

AI in insurance

Published 15 Nov 2023

Specialty

Best in Insurance

Resources

Risk Management

News

RSS

Sitemap

About us

Conditions of Use

Terms & conditions

People

“Take a problem, try to develop a proof-of-concept to test out the hypothesis that AI would be able to solve that problem, and assess the risks of using AI,” Baer said. “If the proof-of-concept is successful, pilot the process and see whether AI is able to help out a small team of people within your organization. Also see whether it creates additional challenges that you might not have otherwise anticipated – and if that’s successful, look at scaling things out from there.”

There may be risks to the technology, but it also promises rewards. For Definity, generative AI is the next logical step following almost a decade of investment in ML, and Definity has already deployed generative AI-enabled use cases in its fraud detection, contact centre, and back-office functions. Getting this right will take focus, and in June Definity communicated with employees regarding appropriate usage of generative AI to ensure that the insurer can promote innovation in a safe and ethical way. Definity is one of the largest insurance carriers in Canada, giving it the capacity to responsibly invest in and implement AI technology. For insurance industry SMBs, the recipe for AI success could be in starting small and focusing on specific existing business pain points, rather than treating AI as a “hammer in search of a nail,” according to Baer.

Humans remain a critical factor in mitigating hallucination risk. “We can’t just let the technology run without people intervening and being actively involved in terms of the modelling process and reviewing the modelling outcomes,” said Baer. There are also data-privacy issues, and the Office of the Privacy Commissioner of Canada continues to investigate the collection and use of personal information by large language-model platforms. “Sometimes people have the tendency to interact with these new chatbots out there and think of them as a trusted confidant,” Baer cautioned. “But when you’re interacting with publicly accessible artificial intelligence, there’s no guarantee of data privacy.” Definity addresses leakage risk through its strategic partnership with Google Cloud, which gives it access to enterprise-grade generative AI technology within a secure cloud environment.

The rise of generative AI has ushered in risks not historically factored in when thinking about more traditional models. “If you’re looking to get into the space of using generative AI to advance your business, be mindful of the risks involved,” Baer cautioned. One top-of-mind risk is that AI is capable of inventing facts, known as the risk of “hallucination.” “[Hallucination] doesn’t happen very often, but when it does happen, because the generative AI seems so credible, people can take that information at face value, and they might use that inaccurate information in a business process or business decision,” Baer said.

Understanding how models are built is a “very intensive” process, and having access to the right resources and people is the largest concern for many businesses looking to tap into the technology, according to Bunn. “That takes a learning process involving two organizations [working together]; it takes deep trust between two organizations to talk openly about that being a challenge, especially when you’re dealing with a bunch of engineers who like to think technology can solve all things,” Bunn said. “Ultimately, this comes down to people and partnership and making sure that everybody understands the weight of the responsibility of using these tools appropriately.”

To support analytics practitioners in ethics-related decision-making, the insurer has introduced an escalation procedure that encourages collaboration with others in the organization to identify and address ethical concerns.

Bias risk is not just an insurance problem, though insurers need to be particularly conscious of the risks. “For an insurance company like us, our models help to inform business decisions that ultimately will affect our customers,” Bellefleur-MacCaul said. Ensuring compliance with regulations and laws is also essential to good model governance. To get ahead of potential risks, Definity launched its data usage ethics standard framework in 2021. The following year, it launched its internal task force on analytics bias and fairness, which led to the development of a suite of tools for analytics practitioners to detect and mitigate social bias in models. “As analytics practitioners who typically go through an educational process related to STEM, we’re not really taught about the impacts of social bias in our work, and so it’s one thing to understand bias from a statistical perspective – but not really understanding systemic bias, and how that can hamper our work, reinforces the need to invest more time to properly understand what we’re doing,” Bellefleur-MacCaul said.

Insurance companies may be reaping AI wins, but a host of potential concerns persists as the technology continues to mature. Ethics plays a pivotal role. The risk of bias amplification is top of mind when creating models. “There are many factors that need to be taken into consideration to ensure we are not doing undue harm with the AI that we’re implementing," said Elizabeth Bellefleur-MacCaul, Definity senior actuarial analyst. “To this day, the number-one critique of any sort of analytical model is its ability to amplify any sort of systemic bias that’s already prevalent in our society.” Research, including Cathy O’Neil’s book Weapons of Math Destruction, has demonstrated that even models intended to effectively “de-bias” a process can have the opposite effect. It’s important to understand the potential for bias in analytical models and address this in an effective way, said Bellefleur-MacCaul. “This would include the underlying assumptions in the data that we’re using, the tools that we are selecting, and the methodology related to predictive modelling, but also ensuring that we have a framework in place to ensure that once something has gone live, it’s not then doing the opposite of what we’re intending.”

Definity also uses analytics to identify when commercial property customers might benefit from consultative risk services or building inspections. First notification of loss (FNOL) benefits are also evident, with analytics being used to assign specialized adjusters to support a customer’s specific claim circumstances. Definity also uses machine learning (ML) to provide customers with personalized recommendations for repair shops and medical providers.

Definity, which established its strategic partnership with Google in 2022, has been reaping the benefits of integrating AI into its workflows for almost 10 years. The insurance company currently employs more than 200 data and analytics practitioners and possesses more than 25 years’ worth of high-quality digitized data. Investing in innovative technologies like Generative AI is one of many avenues Definity is pursuing to achieve its goal of becoming a top-five P&C insurer in Canada. Most insurance companies have already been using analytics to solve foundational underwriting and pricing use cases for some time, and Definity is no exception, according to Jeffrey Baer, VP, enterprise analytics and data office at Definity. However, progress spreads well beyond underwriting. Sonnet, Definity’s digital direct insurance arm, uses fraud detection analytics capabilities to detect potentially suspicious behaviour at point of sale.

ARTIFICIAL INTELLIGENCE (AI) has been used in the insurance industry for years, and has proven use cases, but recent conversations buoyed by generative AI developments, including the meteoric rise of platforms like ChatGPT and Google’s Bard, have propelled the technology to the forefront of executive and public consciousness. Senior representatives from Definity and Google joined Insurance Business for a roundtable, examining how AI is successfully being used in insurance; the risks and opportunities that generative AI presents; and how other industry players can make the most of the technology. Neil Bunn, director of client engineering for Google’s strategic verticals in Canada, notes AI has opened up new ways of communicating with clients and driving efficiencies across claims, compliance, fraud, underwriting, and customer experience. “AI is bringing all of these new technologies to business right now to enable better efficiencies and allow us to do a whole pile of things for clients that are much more natural to them than we’ve been able to use technology for before,” Bunn said.

Tackling ethical AI concerns

Generative AI – global CEO perspectives

70%

Percentage saying generative AI is their top investment priority

Percentage saying ethical challenges are the number-one concern when it comes to generative AI implementation

57%

Percentage who cited increased profitability as the number-one benefit of implementing generative AI

22%

Source: KPMG 2023 CEO Outlook

AI talent challenge

Generative AI brings new risks

Source: EY CEO Outlook Pulse – 2023

29%

Strongly agree

CEOs AGREE AI IS A FORCE FOR GOOD

CEOs were asked whether they agree or disagree that AI is a force for good, driving business efficiency and thereby creating positive outcomes for all, such as innovations in healthcare treatments

Somewhat agree

36%

Neutral

25%

Somewhat disagree

Strongly disagree