Abstract
The increasing capabilities of artificial intelligence (AI) are changing the way organizations operate and interact with users both internally and externally. The insurance sector is currently using AI in several ways but its potential to disrupt insurance is not clear. This research evaluated the implementation of AI-led automation in 20 insurance companies. The findings indicate four business models (BM) emerging: In the first model the insurer takes a smaller part of the value chain allowing others with superior AI and data to take a larger part. In the second model the insurer keeps the same model and value chain but uses AI to improve effectiveness. In the third model the insurer adapts their model to fully utilize AI and seek new sources of data and customers. Lastly in the fourth model a technology focused company uses their existing AI prowess, superior data and extensive customer base, and adds insurance provision.
Keywords
Citation
Zarifis, A., Holland, C.P. and Milne, A. (2023), "Evaluating the impact of AI on insurance: the four emerging AI- and data-driven business models", Emerald Open Research, Vol. 1 No. 1. https://doi.org/10.1108/EOR-01-2023-0001
Publisher
:Emerald Publishing Limited
Copyright © 2019 Zarifis, A. et al.
License
This is an open access article distributed under the terms of the Creative Commons Attribution Licence, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Introduction
Insurance is changing and sources from inside the sector1 and outside2 expect around a third of insurers and their models to disappear within a decade. The sector faces several challenges such as more unpredictable weather and fierce competition, and opportunities offered by technology such as artificial intelligence (AI), the internet of things (IoT), big data and blockchain. AI-led automation has the potential to mitigate many of the challenges and be a catalyst to harnessing the opportunities. The journey and destination of AI led automation however is unclear. This research focuses primarily on business-to-consumer (B2C) insurance. The move of B2C insurance online separated parts of the value chain differently. AI is now separating and re-arranging the value chain again.
AI is already used in several ways in the insurance sector3,4. Examples are fraud detection4, virtual assistants (chatbots)5 and calculating insurance risk6. These applications of AI are having some influence on the business processes. As these islands of AI innovation start merging the influence will be greater, possibly transformative. AI is a broad term that refers to a technology that has the ability to learn. Machine learning and deep learning, where learning can be supervised or unsupervised7,8, are making AI more capable and more accessible to organizations. There is a spectrum of capabilities of AI and there is a spectrum of degrees of implementation of AI. As more capable AI is implemented across the value chain more data is required, and more processes are changed. Some organizations cannot make this step because they do not have the necessary data or willingness to change their processes. This suggests that some organizations can be propelled to new BM by AI while others need to implement new BM to utilize AI.
Along with the insurers and AI a third factor in the diffusion of AI in the insurance sector is the consumer. Purchasing insurance and making claims are part of a relationship between the consumer and the insurer. Most current applications of AI are used to strengthen the capabilities and knowledge of the insurer and not the consumer. As the information asymmetry increases, the insurer can provide lower quality services at lower prices. The higher quality more expensive services may stop being offered as they can no longer compete on price. More information relating to how the decisions are made can be concealed by the insurer increasing the moral hazard.
This raises the question how the user will react to their weaker position and how they should be “compensated”. One approach is to avoid increasing the information asymmetry. Other approaches are to offer lower prices or a better service. Lastly, the fourth option is to also use AI to strengthen the consumers knowledge on the insurance they will be using.
The insurers certainly have a role in shaping AI in this sector. The consumers have their role as does the government that needs to develop and adapt laws and regulation. It is unclear if there should be a fourth stakeholder along the three obvious ones. In many areas, including e-commerce, there are additional institutions like third party certifications that fill in the gap between the retailer and the government in building trust. They are more unbiased than the retailer and more specialized than the government. Who will these third parties be and what will their role be in AI in insurance? Will technology companies be the fourth stakeholder and provide the re-assurance that their AI is trustworthy? Alternatively, will the additional trust building come from within the insurance industry with self-regulation and standards? These issues that emerge from the use of AI in insurance are, potentially, sufficiently transformational to change B2C insurance BMs. Therefore, the research questions of this exploratory research are:
What are the applications of AI in the insurance sector?
What is the effect of AI on insurance value chains and business models? Is it improving existing models or disrupting them?
The next section reviews the literature. This is followed by the methodology section that explain how the cases studies and focus group were implemented. The analysis identifies the new AI and data driven value chain and four AI driven BMs for B2C insurance. The last two sections discuss the implications of these models and sets an agenda for future research.
Literature review
The review is divided into two sections, AI as a technology and the challenges it faces to become a catalyst for new BMs and the existing BMs and value chain in insurance.
AI and AI driven automation
The challenges to AI in insurance come from several sources including the insurance organizations themselves, regulation and the consumer. The unpredictability and difficulty to explain decisions is at the center of many challenges9. The general increase of data is also a challenge exacerbated by AI.
The sociotechnical challenges to AI from within the organization include the data, the people and the processes. The data is not enough in volume or quality in some cases to train AI sufficiently or for AI to use to make decisions. This is one reason why insurers are seeking out collaborations with organizations that can bring more data or use the existing data more intelligently.
In relation to the people, there are not always enough skills and expertise in AI. Furthermore, the insurance sector is historically risk averse and traditional. It is unclear how the user would react to extensive use of AI but the current backdrop of privacy concerns and challenges to trust are an indication that there may be some resistance10. Efforts to regulate AI intend to avoid misuse but also aid its adoption and center on respect of human autonomy, prevention of harm, fairness and explicability11.
In addition to these broader challenges, from the user’s perspective, there are also challenges more specific to insurance. As insurers link their services to offer a “one stop shop” they often penalize loyal users. The information asymmetry is tipped towards the insurer and this can lead to worse contracts for the user. While this is not the way all insurers use technology, this is a user concern that should be considered. It is an example of a recurring theme of how new technology, like AI, changes the relationship between the user and the insurer.
In addition to the increasing role of technology, there is also the increasing role of the technology providers. The knowledge shifts from the insurance professionals to the systems. From the insurers perspective new skills and training are needed to implement AI.
Insurance business models
BMs cover an organization, what it does, how it captures and creates value. While the focus is the organization, a BM can include the partners of that organization12. One important distinction between insurers is those that attempt to cover all the insurance needs of the consumer and act as a “one stop shop” and those that focus primarily on one type of insurance. Another distinction is incumbent, traditional insurers, in contrast to new entrants. Incumbents insurers often cover a broad range of services while new entrants often focus on one service such as car insurance. New entrants often rely on technology to provide a fully digital stand-alone service which is integrated with other fully digital services like mobile payments. Insurers regularly rethink and change their BMs to overcome the typical challenges of too much complexity and weak efficiency, usually measured by cost to income ratio.
There are several ways of explaining insurance BMs including by mapping the value chains13. As we are in a data driven economy, data, its sources and how it is utilized is another way to explain a BM14. Another influence on BMs is the prevalence of powerful platforms such as Google and Alibaba. These platforms fulfil several functions traditional insurer’s BMs used to, like payment processing and data storage.
Before we can identify the new BMs we must step back and ask ourselves what the new value chain is. It can be argued that the impact of AI and other technologies like IoT and blockchain are of such importance that the sociotechnical capability to utilize them is significant enough that it should be included as part of the value chain. Technology has been included in insurance value chain models as a supporting function along with human resources and the legal department and not as a primary function13.
Methodology
An exploratory approach was chosen because the impact of AI is potentially far-reaching and can create a new environment in insurance, where some current theories are no longer valid. This exploratory research analyzed case studies of insurance companies using AI and carried out a focus group with experts. The data sources included academic literature, practitioner-focused sources, insurer websites, websites of organizations that support the insurer’s supply chain, parent companies of insurers and employees of organizations in the insurance value chain. Case study research was used to build theory15,16. In total, 20 cases were selected for this inductive approach. The selection was based on how theoretically useful they were, and it was not random. The cases were chosen because they covered different geographic regions, incumbents and disruptors and different levels of technology focus. More cases were not explored because there was a repetition and a saturation of the themes and it was unlikely that more cases would bring new insights. From the 20 cases, seven are UK insurers, three are from other European countries, six are from north-America and four are from Asia.
The cases were analyzed across six constructs identified in the literature. The constructs were refined and verified in an iterative process15. The constructs were used for within case analysis and across case analysis. To enhance the validity one of the researchers with a background in finance and insurance acted as a “resident devil’s advocate” on the analysis of the cases and the theory that was built17.
Lastly a focus group was held with 12 experts from six large organizations involved in InsurTech. All the experts had experience in implementing AI in the insurance sector. Their roles are senior manager in the IT department of insurers and insurance comparison websites, insurance technology providers and consultants on implementing technology in insurance. They all had over five years of experience. The selected experts were recruited with an email that explained the project and the purpose of the focus group. The discussion was moderated by one of the researchers. To ensure privacy and anonymity no voice recording was taken. One of the researchers took notes. The issues identified in the case studies were put forward for discussion, but the experts were also encouraged to introduce their own topics. The issues put forward for discussion covered the roles of AI, how their organizations currently implemented AI, future plans for implementing AI and the influence on their BM. The workshop lasted two and a half hours. The notes taken during this focus group are available as Underlying data18.
The research received ethical approval by Loughborough University as part of the TECHNGI project (Technology Driven Change and Next Generation Insurance Value Chains) that is funded by the Economic and Social Research Council (ESRC) in the UK. Informed consent was given through the email invitation and subsequent email exchange, reaffirmed by attending and further reaffirmed in an informed way when the purpose of the focus group and how the data would be used was explained before the focus group started.
Case study analysis
The first step was to get a rich understanding of each insurance company using AI and make a within case analysis. This was followed by the cross-case analysis.
The 20 cases selected are currently utilizing AI in insurance. Some, like Lloyd’s of London, are incumbents with a broad range of insurance activities and AI applications. Others, like FRI:DAY are new entrants with a narrower focus in their operations and systems. Beyond some basic descriptive analysis done in the following section the rest of the analysis avoids identifying specific organizations. The purpose of the research is to identify the overarching BMs and not to produce findings on a specific organization. The case studies are available as Underlying data19.
Within-case analysis
All the organization evaluated were currently using AI. Most current implementations of AI replaced specific processes that had specific challenges. One of the earliest implementations was using AI to evaluate damage to cars such as cracked windshields. Table 1 lists the 20 cases, the position of these organizations in the insurance sector and some examples of how they utilize AI. An analysis of the AI used, follows. It is important to understand each organization’s present state but also their journey as an insurer and their relationship to technology in general and AI specifically. The insight gained in the within-case analysis is developed further in the cross-case analysis. First there is an overview of the large incumbent, multichannel insurers, then the small new entrants that only operate online and, lastly, the technology companies that have a role in insurance.
Large incumbent, multichannel insurers
Lloyd’s of London is a large incumbent, B2B, multichannel insurer that focuses primarily on re-insurance. They have several approaches to adopting AI such as developing solutions in their labs, supporting startups and purchasing solutions from leading vendors like Expert System. They are currently turning their physical market place to a digital one.
Willis Towers Watson is primarily a B2B, insurer. They use AI and machine learning to enhance their analysis and predict future risks more effectively. AI is used to develop and optimize behavioral models.
Bupa is a health insurance and healthcare provider. They utilize technology extensively to aid cross selling and sharing of intelligence. They also use an app from Babylon Health that allows virtual consultations with doctors and the prescriptions are provided within the app. The app uses AI to replicate the process a doctor goes through, interpret symptoms and give prescriptions. The data also enables AI to predict disability claims more accurately. Like Bupa, Ping An Insurance offer innovative services that utilize AI to be proactive and shape behavior like improving health.
AVIVA, AXA and Zurich Insurance are primarily B2C insurers. AVIVA engages with AI in several ways including collaborating with TESLA to create new insurance services for AI supported self-driving cars.
AXA employees use an AI enabled virtual assistant by Veezoo for easier access to relevant information. Zurich Insurance uses AI to helps detect fraud cases that are less obvious and can be missed by a rules-based system.
Manulife uses AIDA, an AI underwriting tool that was trained with machine learning and is allowed to underwrite life insurance without a human making the final decision.
Allstate, GEICO and Progressive use virtual assistant that utilize AI. They have different names “ABIE”, “Kate” and “Flo” respectively, but they have similar capabilities. They can answer general questions around insurance and specific questions on the user’s claims and deductibles. They can also set up payments. The virtual assistant can interact with consumers by voice or text.
In addition to the more common uses of AI seen above Tokio Marine applies AI to identify complex handwriting in claim documentation.
Small new, online only insurers
Lemonade Insurance is online and mobile app. based without physical stores. It is a relatively new disruptor that is technology focused and offers innovative B2C services. The whole interaction with the consumer can be carried out with an AI enabled virtual assistant.
FRI:DAY is a small start-up that is mobile app. based without physical stores. It offers innovative services like paying for insurance by the kilometer travelled. Despite the innovative services provided they use a popular insurance enterprise solution. This enterprise solution currently uses some AI for behavioral analysis. CUVVA is similar to FRI:DAY. It also offers innovative B2C services like paying for insurance per hour, in an automated interaction online supported by AI.
E-commerce and tech companies in insurance
Comparethemarket.com, of the BGL Group, is an incumbent, tech-focused B2C, company offering a platform to the user with several insurance providers. Unlike most implementations of AI in the insurance sector, that enhance the knowledge of the insurer, AI is implemented here to enhance the knowledge of the user so they can make the best choice for themselves. This reduces the information asymmetry as opposed to increasing it.
Confused.com offers similar services to Comparethemarket.com. They use AI solutions provided by Microsoft, including virtual assistants for customer support that can analyze sentiment.
Alibaba is a tech and e-commerce giant that offers several insurance services directly and through several partnerships. It has a large user base that purchase from its marketplace and it has entered the insurance sector with innovative services. Alibaba uses AI heavily in healthcare and it is being integrated with new insurance products. As this technology giant uses AI extensively and the several partners it has in insurance also use AI extensively it can be challenging to have a transparent image of how AI and data is used across the many actors involved in providing the insurance. Achieving auditable “explainability” is harder with AI and complex partnerships.
Tencent, like Alibaba, is also a tech and e-commerce giant that offers several insurance services. It has a large user base from its social media platforms like WeChat. It is entering the insurance sector with innovative services. Tencent has extensive AI capabilities including a leading AI lab that integrates the use of AI, blockchain and other technologies across their services.
TESLA is a technology company and car producer that is influencing the insurance sector primarily through its Insure MyTesla program. Tesla use AI for their current cars and are developing self-driving cars that use AI more extensively. There are new opportunities and challenges for insurance by AI in both current cars that can drive independently, but even more so for self-driving cars. Therefore, TESLA takes a proactive approach in shaping the insurance their cars need and partnering with traditional insurers to deliver it in different countries.
Cross-case analysis
The cross-case analysis offers further insight on each insurance company’s use of AI but also on the potential of AI as a whole to disrupt this sector. The analysis of the current implementations of AI and the plans for further implementing AI, indicate that AI is driving automation and is enabling the insurer to be more proactive rather than reactive. The analysis is separated into six areas: User-facing and back office applications of AI in insurance, sources of data and users, new insurance and technology ecosystems and lastly the capability to implement AI driven automation.
User-facing applications of AI in insurance
The current processes that are enhanced by AI are presented in Table 2 for the user-facing applications and Table 3 for the back-office operations. While the distinction between the two is not always clear it is a useful way to group the many current AI applications. The hundreds of applications of AI supporting the interaction with the user, across the 20 cases, fall into 11 applications. While this number could be further reduced with larger groups, these 11 applications represent sufficiently distinct and useful applications to be separate.
Proactive processes and active loss prevention: Several capabilities of AI, such as the speed of analysis and the volume of data that can be analyzed, support tailored proactive processes. Utilizing the data from sensors and IoT further enhances this. AI can identify emerging threats, warn customers and support active loss prevention. It can also encourage and monitor better behavior.
Virtual assistants, chatbots and robo-advisors: They use natural language comprehension and recognize facial expressions and feelings. They are intuitive and keep a consistent state throughout a conversation. In other words, they remember the previous questions made by a user and answer subsequent questions accordingly. Chatbots and robo-advisors on websites or apps can interact with the user with spoken language or text.
Fast initial offers and underwriting: The AI led automation can speed up the evaluation of the request, calculate a price in 60–90 seconds and underwrite it.
Faster and more accurate claims processing: IBM Watson and similar AI are used for calculating the claim payout amount and making the payment automatically. In some implementations a reimbursement is triggered by a sensor on an IoT before a claim is even made. Furthermore, AI can process unstructured data, including pictures. For example, a picture of cracked windshield can be evaluated for a claim.
Improving process flow with clients: Beyond the insight, the order and timing of the engagements with the customer across multiple channels is coordinated and optimized.
Enabling and supporting new more flexible services. New services like paying car insurance per KM or per hour require more data to be collected and processed. The AI led automation also leads to paperless processes.
Direct marketing and customer retention: These existing processes, can be optimized and tailored to each user and their current circumstances.
Tailored insurance advice: AI can harness information from disparate sources like social media, IoT sensors and big data for more accurately tailored advice.
Understanding the user’s emotions: Natural language processing and facial expression processing can identify user’s emotions, feelings perspectives and attitudes towards risk.
Identify legal parameters across countries: While some insurers want to standardize their operations, different countries and regions have different laws, regulations and norms. This is a problem that AI can help with by analyzing local data and localizing the services offered.
AI insuring AI such as driverless cars: AI has a bigger influence in cases where humans are still in the loop and makes more decisions without humans in the loop. In many cases the decision making of AI is not predictable, explainable or transparent. The unpredictable nature of some AI and the large volumes of data it can use, make evaluating its decision challenging in some cases. The scale and complexity can be challenging for humans, so AI can support the process of insuring AI.
Back-office applications of AI in insurance
The 13 primary uses of AI that were identified are listed in Table 3.
Automating simple, low-value tasks: Automating simple, low-value tasks such as simple analysis or rearranging data. AI can replace the simplest administrative tasks.
Processing large volumes of data: The ability of AI to process different types of data enable it to process large volumes of data such as multiple policies and information from IoT devices and social media.
Populating and improving data sets: AI can populate and enhance the quality of a data sets that will be used later to make decisions.
Processing structured, semi-structured or unstructured datasets: Several sources of information are used, such as engineers reports and geological information, which is not always in the ideal form for what it will be used for.
Claims processing and adapting to new risks: New risks, such as the recent increase of thefts of cars with keyless entry due to a vulnerability in that technology, influence the overall risk to an insurer. As the overall risk is influenced the way specific claims are processed is also influenced. AI can support evaluating and adapting to new risks.
New insights on current clients: Analyzing current users to avoid premium leakage can be enhanced. The risks current users face change over time. The quicker their cover adapts the better it will cover new and evolving risks.
Utilizing the data from the IoT: The number of IoT devises will continue to expand both for individuals, for example with smart homes, and organizations, for example with connected sensors on freight containers. Furthermore, some insurers sell devices like health trackers and use their own devises like telematic “black boxes” in cars. A large proportion of the data the insurer will use in the future will come from IoT devices and AI can support analyzing it. This enables behavioral insurance, where the terms of the insurance change dynamically with behavior.
Virtual assistants for analysts: These assistants for analysts use AI to keep a consistent state throughout a conversation. In other words, they remember the previous analysis requests made by an analyst and answer subsequent questions accordingly.
Fraud detection: Expert systems that use a rules-based approach do not capture all frauds, particularly new types of fraud. The capabilities of AI can reduce subrogation, multiple unlawful claims and scams, such as written off cars that are later sold second hand.
Faster risk detection for automated services: As AI can analyze risks quickly and 24 hours a day, seven days a week, risks can be identified instantly. Therefore, automated services and humans can respond quicker. This fast reaction can also support a more proactive approach and less reactive approach.
Improved risk analysis: AI can process large amounts of disperse data which is beneficial in risk analysis. This improves risk analysis by analyzing the customer and other risks, such as natural disasters more effectively.
Identifying and analyzing new data: Unlike most rules-based systems, machine learning and deep learning can identify new data that is useful. As more data is generated in many ways like social media and IoT devices, identifying which of this data is relevant is beneficial.
Audit: AI can support an insurer to audit its processes in several ways. AI can prepare deeper insight for the human analyst by drawing on more data. Cases of special interest can be identified for a human to review. Potential fraud within the organization can also be identified that is too innovative or subtle, for rule-based systems to detect.
Sources of data
Based on the case studies of the insurers, the volume of data is important but so is the quality. If AI uses inaccurate insurance data, it will not perform effectively. The data needs to be checked and cleansed of inaccuracies. While AI can process unstructured data, it is more effective with data sets that offer the relevant information.
Sources of customers
In B2C insurance, AI driven automation using only online channels creates a large capacity. The only limitation is the amount of risk the insurer wants to take on. Furthermore, these users are new sources of data. Therefore, gaining access to new uses is important. This can be achieved through collaborations with partners from other sectors and cross selling insurance products to them or by the insurer moving into other areas.
While new sources of users should be pursued, it is equally important to avoid and mitigate the risks to losing users. From the user’s perspective some risks increase with AI driven automation: The capability of AI increases the information asymmetry between the user and the insurer. A larger volume of data that is collected from an expanding number of sources is processed. This amplifies privacy concerns. Lack of transparency may reduce trust and raise ethical concerns.
Insurance and technology ecosystem
Fully utilizing AI requires several sociotechnical capabilities across the value chain that one organization may not possess. Furthermore, there are new opportunities that AI offers beyond the traditional insurance value chain. Therefore, an ecosystem with other partners insurance and the technology sector may be necessary. One example is that insurers can combine their data so that AI can be trained better and have more data to make decisions. A second example is that it may be easier for one member of an ecosystem that already has data and a relationship with the user, to ask for more data and access, than another member of the ecosystem. Therefore, all members of the ecosystem can benefit from the data and access that they would not otherwise have.
Capability to implement AI driven automation
The changes necessary to fully utilize the capabilities of AI require new BMs, knowledge and skills. This requires a willingness to change from the employees who are technology users within the organization and the customer who are a technology user outside the organization. Based on the analysis of the case studies, the transition to new models may take several years. The transition includes sociotechnical challenges that go beyond the organization including the users. The users concerns over the increasing information asymmetry, and privacy concerns will require trust to be built. Some smaller organizations may make this transition more quickly while some large organizations may separate specific services, such as car insurance, and prioritize moving them towards the new AI and data driven model.
Focus group
The participants to the focus group raised five main interconnected themes. Firstly, this was the journey of AI in insurance, where we are now and where the final destination is. Secondly the challenges in implementing AI within the organization were emphasizes. These included the lack of skills related to AI and the importance of effective change management. Furthermore, the immaturity of both the ethics and regulation on AI where discussed.
The fourth topic was the current implementations of AI. There was some agreement that most current implementations of AI solved specific problems for specific processes. These specific applications stopped short of being enterprise systems or linking beyond one organization.
The fifth topic was the threat to the incumbent insurers from AI as it can be an enabler to new entrants and disruptors. The sixth topic was the importance of data and the increase in data from IoT. Overall the focus group supported the findings of the case study analysis although the importance of having enough, high quality data to support the training of AI but also support its decision making, came out more strongly. Most participants agreed that AI and an increase in the quantity and quality of data, could revolutionize BMs in the future but that this may be a gradual process over several years.
Discussion
Insurance value chain with AI
Traditional B2C insurance value chains need to be adapted to accurately capture the impact of AI and AI driven automation. The six areas that are important to AI driven automation identified in the 20 case studies form the AI and data driven insurance value chain are illustrated in figure 1. Figure 1 illustrates the first order themes of the value chain. Second order themes can also be developed to offer greater granularity.
Firstly, as AI, and the data it needs, cover more of the value chain a more data-centric value chain would represent the new B2C insurer better. A data driven model can have data central to the value chain, not in a supporting role, and can include internal and external data sources14.
Sources of consumers should be included in the new value chain as it should not be assumed that the consumer will come to the insurer directly. For example, tech-companies like UBER, TESLA, Alibaba and Tencent bring their consumers to be insured. Many value chains include marketing and distribution but there are also some that include the consumer20.
The four insurance business models for AI driven automation
From the 20 case studies four AI and data driven BMs emerged as illustrated in Figure 2. AI is the catalyst in creating this new taxonomy for insurance companies. These four models cover all of the case studies. Some insurers had a hybrid model using one model for some services and another model for others. These hybrid models could remain hybrids in the future, or when more AI is absorbed, they may transition into one uniform AI and data driven model.
Focus strategy and disaggregation
In this model the insurer implements a smaller part of the value chain, has a larger ecosystem and may join a platform. In this strategy “less is more”. The insurer has a reduced part of the value chain either by choice or because other organizations, such as tech companies, adopt processes they used to do. The insurer focuses on their existing core competencies and disaggregates others. A tech company or e-commerce platform can implement other parts of the value chain. Examples of tech companies taking a larger part of the value chain are TESLA and UBER. UBER agreed with insurers such as AVIVA and AXA to insure the passenger for one trip.
The “streaming generation”, may make more subscriptions for cars than purchases, so these collaborations will be helpful to insurers to access these temporary users.
Another example is where the insurer focuses on the marketing and distribution. In this model the insurer accepts that other organizations are better positioned to benefit from many of the changes AI brings. Other organizations may be better positioned because of better in-house AI capabilities, applying external AI better, having more, and better-quality data for AI to use, or a larger customer base to apply AI to. Becoming smaller across the value chain can be seen as an effective way to become part of a larger ecosystem.
Same model and value chain but improved by AI
In the second BM the insurer utilizes AI across their value chain but does not change their BM to fully utilize the opportunities AI led automation can bring. AI-driven automation may be implemented in some processes and services but not at enterprise level, across the BM, so that risk is reduced. AI is mostly absorbed into current processes. An advantage of the insurer applying the AI themselves, “in-house”, with their data, is that the industry specific knowledge is not lost or given to partners in the AI implementation. Therefore, the insurer has control over their data, their industry specific knowledge and how they use them to shape their competitive advantage. A traditional insurance company that is risk averse with technology and processes would be in this category.
Expanding beyond the insurance value chain, seeking more data, becoming the platform
An example of an insurer in the “go big or go home” model is selling IoT hardware or using the data from it. There are examples of this in health, such as Bupa, and transport. In health people are tracked by IoT sensors and in transport containers and goods. New forms of big data create new opportunities. Some traditional risk averse insurers may use startups they own to test these new opportunities.
Tech company expanding into the insurance value chain
The case studies illustrated the increasing role AI and data will have in the future. This is also reflected in the value chain proposed. This prominent role of technology across user-facing and back office processes, accessing data and implementing AI means technology companies and e-commerce platforms are in a strong position. Furthermore, companies like Tencent and Alibaba have access to customers. As these tech companies do not have a long legacy in insurance, they are arranging their processes to deliver insurance with a focus on AI and data from the start. While tech companies and e-commerce platforms are in a strong position to implement the new value chain effectively, there may also be other innovative companies or startups coming from outside insurance that can harness AI, data and users sufficiently. New incumbents, with no reputation to lose, can accept a higher risk and use technology in new untested ways.
Conclusions and further research
This research gained insight at industry and company level. The 20 case studies and the focus group indicate the disruptive potential of AI and reinforce other research that suggested this21. Some processes change in a specific and fundamental way, while in other cases AI drives a broader automation in a similar way to the Internet. In a similar way to the disruptive impact of the Internet the transition is a journey for the insurance sector.
This research makes three primary contributions: Firstly, the current applications of AI in the insurance sector were identified. Secondly, a new AI and data driven value chain for B2C insurance was developed. Thirdly, four AI and data-driven BMs for insurance emerged. The choice of model and value chain influences the approach to implementing AI. In the focus and disaggregation strategy more off the shelf solutions are used with less customization. In the second model where AI is absorbed into the existing model some customization is implemented. In the third and fourth model extensive new capabilities of AI are developed and implemented.
Further research can explore how the transition to new AI and data focused models can be implemented and how the users inside and outside the organization can be encouraged to support this change.
Data availability
Underlying data
Figshare: 20 case studies on AI evaluating the impact of AI on insurance. https://doi.org/10.6084/m9.figshare.9845015.v219.
Figshare: Notes from workshop on AI evaluating the impact of AI on insurance. https://doi.org/10.6084/m9.figshare.9845048.v118.
Data are available under the terms of the Creative Commons Attribution 4.0 International license (CC-BY 4.0).
Publisher’s note
This article was originally published on the Emerald Open Research platform hosted by F1000, under the Fairer Society gateway and Information systems collection.
The original DOI of the article was 10.35241/emeraldopenres.13249.1
Author roles
Zarifis A: Conceptualization, Data Curation, Formal Analysis, Investigation, Methodology, Resources, Validation, Writing – Original Draft Preparation, Writing – Review & Editing; Holland CP: Conceptualization, Funding Acquisition, Investigation, Writing – Review & Editing; Milne A: Conceptualization, Funding Acquisition, Investigation
Funding statement
This research is part of the TECHNGI project (Technology Driven Change and Next Generation Insurance Value Chains) that is funded by the Economic and Social Research Council (ESRC) in the UK, grant number ES/S010416/1.
Competing interests
No competing interests were disclosed.
Reviewer response for version 1
Joao Alvaro Carvalho, Department of Information Systems, Centro ALGORITMI, University of Minho, Guimaraes, Portugal
Competing interests: No competing interests were disclosed.
This review was published on 12 December 2019.
This is an open access peer review report distributed under the terms of the Creative Commons Attribution Licence, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Recommendation: approve
The article addresses the consequences of the use of artificial intelligence (AI) and other technologies in the insurance sector. It involved a set of case studies and a focus group aiming at answering 3 research questions: What are the applications of AI in the insurance sector? What is the effect of AI on insurance value chains and business models? Is it improving existing models or disrupting them?
By the end the authors claim that they contributed with: an identification of “current applications of AI in the insurance sector” (first question); a “new AI and data driven value chain for B2C insurance” (related to the second question); identified “four AI and data-driven BMs for insurance” (related to the second question).
The authors also claim that the study reveals a “disruptive potential of AI” (related to the third question). Being a potential effect, the reader is led to the conclusion that there isn’t yet a disruptive effect of AI.
The problem addressed by the authors is very interesting, current and relevant. IA applications are still far from reaching stable configurations and it is of major importance to understand the roles they are playing in organizations, how they are altering the value creation in the insurance sector and they are reshaping the insurance business.
As an initial study (as I will argue I don’t think it corresponds to an exploratory study), it opens perspectives on the changes the insurance sector has been enduring and the role AI is playing in those changes.
The study has, however, some drawbacks, mostly related to the research design and to a weak association with theory.
1) Research design – nature of the study
The study described in the article is presented as exploratory. It is not clear, however, what the authors mean by exploratory.
An exploratory study paves the way for a confirmatory study. However, the future research presented by the authors aims at exploring other research avenues. It doesn’t focus on confirming the outcome of the presented study.
On the other hand, the authors claim that “Case study research was used to build theory”. If it is the case, the study led to the production of theory and, therefore, is not just exploratory work.
Furthermore, the authors refer that the “cases were analyzed across six constructs identified in the literature”. If it is so, the literature provided elements enough for the analysis of the cases. Thus, the study loses its exploratory nature.
2) Research design – information provided
In what concerns the information provided about the research design, there are some omissions:
• Case studies
There is no information about the “boundaries” of each case; the authors describe the data sources (“academic literature, practitioner focused sources, insurer websites, websites of organizations that support the insurer’s supply chain, parent companies of insurers and employees of organizations in the insurance value chain”). It is not clear whether the study involved interviews of employees of insurance companies (they are referred to as data sources). There is no information about the elements that were sought for each case study. Such omission compromises the comparability of the cases. It seems that most of the data sources are external to the companies. If so, how was it possible to get accurate, reliable and comprehensive information about the AI applications used by the insurance companies?
•Characterization of AI applications
The description of the AI applications is carried out in an ad-hoc manner. Sometimes the characterization focuses on its functionality, other times on the advantages they provide (faster, enabling. Improving, …). Thus, the characterization lacks systematicity.
•Focus group
The article mentions the five (actually six) main themes addressed. But no information is provided about the dynamics of the focus group. How is that the themes emerged? The participants were shown the results of the case studies? They were asked some questions?
The participants in the focus group were mainly IT specialists. It seems that no insurance experts were involved. So, a technological bias might be present in the study.
3) Results
The authors say that the “cases were analyzed across six constructs identified in the literature”. What are these constructs? When presenting the insurance companies, they are characterized according to their size, age, channels. Are these the constructs mentioned by the authors? Where do they come from (what theories)?
The authors frequently refer to value chains and business models. However, they don’t clearly present what they mean by these concepts. The references in these areas seem to be used to justify the use of the terms (“value chain”, “business model”) but their structuring concepts are not explained.
As a consequence, the contents of figures 1 and 2 reflect a naïve perspective of value chains and business models. Such perspective prevents reaching more interesting results.
- Is the work clearly and accurately presented and does it cite the current literature?
Partly
- If applicable, is the statistical analysis and its interpretation appropriate?
Not applicable
- Are all the source data underlying the results available to ensure full reproducibility?
Partly
- Is the study design appropriate and is the work technically sound?
Partly
- Are the conclusions drawn adequately supported by the results?
Partly
- Are sufficient details of methods and analysis provided to allow replication by others?
Partly
Reviewer expertise: information systems, information technology
I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.
Reviewer response for version 1
Alex Rast, University of Southampton, Southampton, United Kingdom
Competing interests: No competing interests were disclosed.
This review was published on 21 October 2019.
This is an open access peer review report distributed under the terms of the Creative Commons Attribution Licence, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Recommendation: approve.
This work analyses the different business models being used in the insurance sector to leverage or deploy AI. A broad range of insurers are analysed coming from different initial business models and sizes, and the resulting plans categorised into 4 distinct business strategies.
The strategies listed are intuitive and in addition to giving a good sense of where the sector is at present, may provide valuable decision-making guidance to other insurance firms in setting strategic policy for AI deployment. Particularly interesting is the authors' contention that the introduction of AI will fundamentally change the insurance business model, a supposition well supported by the entry of some smaller firms into the field as analysed here – Lemonade and FRI:DAY.
The indication also that players outside the traditional insurance sector can use their AI expertise to enter or influence the market is also interesting, although this reviewer would like to see a more detailed analysis of specific cases, e.g. Tesla Motors, whose influence over the market, while plausible, is not clear, and where furthermore it is left somewhat vague what their strategy is with respect to the insurance sector (other than a generalised understanding that policies may need to be altered in light of AI for self-driving vehicles).
Occasionally, it is unclear where the conclusions arise from; the data from the focus group study is (understandably) somewhat non-specific and the analysis, particularly with regards to use of AI in the companies involved, appears to be somewhat of an unstructured list of applications. The section 'case study analysis' could do with a little more classfication and ordering into coherent areas.
Sometimes the wording can be confusing. In “Cross-case analysis”, the authors indicate 6 analysis areas, but then the list immediately following appears to indicate 7 areas until a further reading clarifies the cases. This might be slightly reworded to make the distinctions clear.
The authors state that participants in the focus group raised 5 themes, but then go on to list 6 “topics”. Are topic and theme the same thing? If so then the numbers should be reconciled. If not, then the authors should clarify the distinction.
After the “take-home-message” figure 2 on the business models on Page 10, the article ends somewhat abruptly. A bit more reflection on the impact of these business models looking forward together with some projection of the state of the insurance sector in, say, 5 years and 10 years from now, would be welcome.
Apart from the clarifications and perhaps more systematic classification of their data sources, however (also indicated by the first reviewer), this work is a well-written article and should stimulate examination and refinement of AI models in the insurance industry
- Is the work clearly and accurately presented and does it cite the current literature?
Yes
- If applicable, is the statistical analysis and its interpretation appropriate?
Not applicable
- Are all the source data underlying the results available to ensure full reproducibility?
Partly
- Is the study design appropriate and is the work technically sound?
Yes
- Are the conclusions drawn adequately supported by the results?
Yes
- Are sufficient details of methods and analysis provided to allow replication by others?
Yes
Reviewer expertise: artificial Intelligence – hardware and systems, neural networks, parallel processing systems.
I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard.
Reviewer response for version 1
Aida Azadegan, Intelligent Systems Research Laboratory, University of Reading, Reading, United Kingdom
Competing interests: No competing interests were disclosed.
This review was published on 21 October 2019.
This is an open access peer review report distributed under the terms of the Creative Commons Attribution Licence, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Recommendation: approve.
The paper is aiming to review how AI is currently used in the insurance sector and potential opportunities to disrupt Insurance. The authors selected 20 case studies describing AI lead automation in 20 insurance companies. Their findings indicate four emerging business models that describes the role of AI and how it is positioned with respect to the role of the insurer.
Generally, the paper is well written. There are some minor typos within the narrative explained in the text e.g. This research evaluates rather than evaluated (referring to the abstract section of the paper).
The authors mention two Research Questions on page 3. It is suggested that authors explain how and to what extent their findings could address their generic Research Questions that clearly mentioned within the introduction section and make direct references to each question provided.
The authors decided to choose 20 case studies. It would be very helpful if the authors could explain their criteria of selection for these case studies and why they think the selected case studies were the suitable and most relevant ones for their purpose of research. Also why they think 20 cases would be adequate for them to get sufficient insights into their research research and address the Research Problems mentioned in the paper.
With regards to the focus group setting, it would be helpful if authors could explain further why they think a Focus Group would best serve the purpose of this research, why they decided to choose 12 experts and what were the criteria for selection of those experts? How they decided to reflect on focus group questions. On page 4 the authors mention that the issues identified in case studies were put forward for focus group discussions therefore it is needed to clearly explain what issues were identified and how they were classified.
Table 1 provided details that were gathered by the authors from the case studies in relation to each insurance company. Authors are advised to provide reference to all information provided in the table. The authors are suggested to explain how they gathered the information provided in the table and also how they could prepare and provide information to be able compare and contrast between different case studies. Also it is useful to explain why they believe such information is provided at the same level of abstraction to enable them to make such comparison and to classify them. Table 1 explains the position of companies in insurance, authors are advised to provide reference to the source of such classification model or if such positioning and classification is performed by authors themselves it would be helpful if they could explain what criteria they sued to address such positioning of companies and a reference that explains where they could get the related information from.
On page 6 again it would be helpful if authors could provide references to all information explained with respect to each named insurance company.
On page 7 authors explain how they could separate they analysis into 6 areas named in the paper.
- Is the work clearly and accurately presented and does it cite the current literature?
Yes
- If applicable, is the statistical analysis and its interpretation appropriate?
Not applicable
- Are all the source data underlying the results available to ensure full reproducibility?
Partly
- Is the study design appropriate and is the work technically sound?
Yes
- Are the conclusions drawn adequately supported by the results?
Yes
- Are sufficient details of methods and analysis provided to allow replication by others?
Yes
Reviewer expertise: e-learning, serious gaming, collaborative decision making, requirements engineering, HCI.
I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard.
Figures
The 20 cases of AI in insurance.
| Case | Position in insurance | Examples of AI applications | |
|---|---|---|---|
| 1 | Lloyd’s of London | Large incumbent, B2B, multichannel | Data entry quality control: AI analyses contracts to ensure regulatory compliance. |
| 2 | Willis Towers Watson | Large incumbent, B2B, multichannel | When processing a claim or evaluating a business customer AI can identify the right expert and bring in their expertise at the right point. |
| 3 | AVIVA | Large incumbent, B2C multichannel | A virtual assistant can be asked questions with natural language and reply. |
| 4 | Compare themarket.com | Incumbent, tech-focused B2C, company offering a platform | Digital Voice: Full or semi-automatic interaction with the consumer using natural language processing and big data are supported by AI. |
| 5 | Bupa | Incumbent, health insurance and healthcare provider | AI is used for Illness and disability claim prediction. |
| 6 | Confused.com | Online, tech-focused, B2C, low cost focused on car insurance | AI supporter virtual assistants (chatbots) support the customer with sentiment analysis and automation of time-consuming and repetitive tasks. |
| 7 | CUVVA | Small start-up, online, B2C, offers innovative services | AI supports underwriting, claims processing and fraud detection. |
| 8 | AXA | Large incumbent, B2C, multichannel | AI supports underwriting, customization, claims processing and fraud detection. |
| 9 | Zurich Insurance | Large incumbent, B2C, multichannel | Fraud detection and reduction. AI also supports underwriting and claims processing. |
| 10 | Tokio Marine | Large incumbent, B2C, multichannel | Claim documentation, AI “reads” hand-written documents. |
| 11 | Manulife | Large incumbent, B2C, multichannel | AI can underwrite insurance independently. |
| 12 | Allstate | Large incumbent, B2C, multichannel | AI supports underwriting, customization with a virtual assistant, claims processing and fraud detection. |
| 13 | GEICO | Large incumbent, technology focused, B2C, specialized in car insurance | Recruitment, AI carries out interview and matches the applicant’s skills to the right roles. |
| 14 | Progressive | Large incumbent, technology focused | AI is used to improve efficiency across a complete process so that humans are not needed. |
| 15 | Lemonade Insurance | Online, no physical stores, relatively new, technology focused, B2C, offers innovative services | Their virtual assistant called “AI Jim”. “AI Jim” can interact with the consumer during the buying a policy, switching from another insurer, claims process. AI supports behavioral analysis. |
| 16 | FRI:DAY | Small start-up, online, technology focused B2C, offers innovative services | The Guidewire system used are achieved by AI that is trained specifically to identify patterns that indicate fraud in insurance. |
| 17 | Ping An Insurance | Large incumbent, B2C, multichannel, offers innovative services | Innovative services utilize AI to be proactive and shape behavior like improving health. |
| 18 | Alibaba | Tech and e-commerce giant, with large user base entering the insurance sector with innovative services | Alibaba uses AI heavily in healthcare and this is being integrated with new insurance products. |
| 19 | Tencent | Tech and e-commerce giant, large user base entering insurance with innovative services | AI is used across this technology giant’s e-commerce and social media for analysis, facial recognition, natural language processing, fraud detection and security. |
| 20 | TESLA | Tech company influencing insurance sector | There are new opportunities and challenges for insurance created by AI in self-driving cars. |
Applications of AI in insurance (user-facing).
| AI application | |
|---|---|
| 1 | Proactive processes and active loss prevention |
| 2 | Virtual assistants, chatbots and robo-advisor |
| 3 | Fast initial offers and underwriting |
| 4 | Faster and more accurate claims processing |
| 5 | Improving process flow with clients |
| 6 | Enabling new more flexible services |
| 7 | Direct marketing and customer retention |
| 8 | Tailored insurance advice |
| 9 | Understanding the user’s emotions |
| 10 | Identify legal parameters across countries |
| 11 | AI insuring AI such as driverless cars |
Applications of AI in insurance (back-office).
| AI application | |
|---|---|
| 1 | Automating simple, low-value tasks |
| 2 | Processing large volumes of data |
| 3 | Populating and improving data sets |
| 4 | Processing structured, semi-structured or unstructured datasets |
| 5 | Claims processing and adapting to new risks |
| 6 | New insights on current clients |
| 7 | Utilizing the data from the IoT |
| 8 | Virtual assistants for analysts |
| 9 | Fraud detection |
| 10 | Faster risk detection for automated services |
| 11 | Improved risk analysis |
| 12 | Identifying and analyzing new data |
| 13 | Audit |
References
1. Bauer, A. and Gröninger, Y. (2017), “Dying, Surviving or Thriving – Strategic analysis of the future Swiss insurance market”, available at: Reference Source.
2. Business Weekly> (2018), “Microsoft chief warns companies to act now to maximise AI opportunity”, available at: Reference Source.
3. Hall, S. (2017), “How Artificial Intelligence is changing the insurance industry”, The Center for Insurance Policy & Research, Vol. 22, pp. 1-8, available at: Reference Source.
4. Wang, Y. and Xu, W. (2018), “Leveraging deep learning with LDA-based text analytics to detect automobile insurance fraud”, Decis Support Syst, Vol. 105, pp. 87-95, doi: 10.1016/j.dss.2017.11.001.
5. Riikkinen, M., Saarijärvi, H., Sarlin, P., et al. (2018), “Using artificial intelligence to create value in insurance”, International Journal of Bank Marketing, Vol. 36 No. 6, pp. 1145-1168, doi: 10.1108/IJBM-01-2017-0015.
6. Vassiljeva, K., Tepljakov, A., Petlenkov, E., et al. (2017), “Computational intelligence approach for estimation of vehicle insurance risk level”, Proceedings of the International Joint Conference on Neural Networks, Vol. 2017, pp. 4073-4078, doi: 10.1109/IJCNN.2017.7966370.
7. Ghahramani, Z. (2015), “Probabilistic machine learning and artificial intelligence”, Nature, Vol. 521 No. 7553, pp. 452-459, 26017444 doi: 10.1038/nature14541.
8. LeCun, Y., Bengio, Y., Hinton, G. (2015), “Deep learning”, Nature, Vol. 521 No. 7553, pp. 436-444, doi: 10.1038/nature14539.
9. Langley, P., Meadows, B., Sridharan, M., et al. (2017), “Explainable Agency for Intelligent Autonomous Systems”, 31st Conference on Artificial Intelligence (AAAI 2017), Proceedings, pp. 4762-4763, available at: Reference Source.
10. Saffarizadeh, K., Boodraj, M., Alashoor, T.M., et al. (2017), “Conversational Assistants: Investigating Privacy Concerns, Trust, and Self-Disclosure”, Proceedings of the International Conference on Information Systems (ICIS), pp. 0-12, available at: Reference Source.
11. European Commission (2019), “Ethics Guidelines for Trustworthy AI”, available at: Reference Source.
12. Zott, C., Amit, R. and Massa, L. (2011), “The Business Model: Recent Developments and Future Research”, J Manage, Vol. 37 No. 4, pp. 1019-1042, doi: 10.1177/0149206311406265.
13. Eling, M. and Lehmann, M. “The Impact of Digitalization on the Insurance Value Chain and the Insurability of Risks”, (2018), Geneva Papers on Risk and Insurance: Issues and Practice, Vol. 43 No. 3, pp. 359-396, doi: 10.1057/s41288-017-0073-0.
14. Hartmann, P.M., Zaki, M., Feldmann, N., et al. (2017), “Capturing Value from Big Data – A Taxonomy of Data-Driven Business Models Used by Start-Up Firms”, International Journal of Operations and Production Managment, Vol. 022, pp. 1-14.
15. Eisenhardt, K.M. (1989), “Building Theories from Case Study Research”, Acad Manage Rev, Vol. 14 No. 4, pp. 532-550, doi: 10.5465/amr.1989.4308385.
16. Miles, M.B. and Huberman, A.M. (2018), Qualitative data analysis, Sage Publications, Thousand Oaks, CA.
17. Sutton, R.I. and Callahan, A.L. (1987), “The Stigma of Bankruptcy: Spoiled Organizational Image and Its Management”, Academic of Management Journal, Vol. 30 No. 3, pp. 405-436, doi: 10.5465/256007.
18. Zarifis, A. “Notes from workshop on AI evaluating the impact of AI on insurance”, (2019), figshare, Dataset, available at: http://www.doi.org/10.6084/m9.figshare.9845048.v1.
19. Zarifis, A. (2019), “20 case studies on AI evaluating the impact of AI on insurance”, figshare, Dataset, available at: http://www.doi.org/10.6084/m9.figshare.9845015.v2.
20. Bonaccorsi, A., Giannangeli, S. and Rossi, C. (2006), “Entry Strategies Under Competing Standards: Hybrid Business Models in the Open Source Software Industry”, Manage Sci, Vol. 52 No. 7, pp. 1085-1098, doi: 10.1287/mnsc.1060.0547.
21. Maull, R., Collomosse, J., Brewer, S., et al. (2019), “Taking control: Artificial intelligence and insurance”, available at: Reference Source.