It seems that Artificial Intelligence (AI) is being used everywhere nowadays, from virtual health assistants in the medical fields to fraud detections in the finance industry and even crop forecasting in the farming sector. But – and perhaps surprisingly – the very first groundworks for AI have been established back in the 1950s/60s and although these were largely theoretical, AI started to enter businesses already 20 years later, mainly in the form of decision-making tools (e.g. credit risk assessments)(1).
So what makes AI such a prevalent topic in 2024? The answer is quite simple: its increased adoption across all industries, as more and more companies begin to realise the advantages. Even the insurance industry, which may still be perceived as more old-fashioned and traditional by some, has started to implement this technology. Its many uses include, for example, automated claims processing, natural disaster prediction, customer service chatbots and personalised pricing. As a result, many insurance experts are already moving into advisory roles as AI-powered automation frees up time that would be otherwise spent on manual tasks(2). This change also allows insurers to dedicate more time to value-add activities, provide better than average service or address any emergencies.
Nonetheless, implementing AI into your business is no easy task and you need to ensure that you are properly prepared and aware of all the obstacles you might encounter. We cover the main ones in our article: AI regulations, technology challenges and data quality.
AI Regulations in the EU and UK
As mentioned, AI has been around for some time now. However, the governments have started to regulate this technology only quite recently, and while the EU proposed its first regulatory framework in April 2021, the UK government published a regulation white paper only in March last year. The main underlying challenge for setting up the regulations is the high-speed progress of AI and its ever-changing landscape – meaning that any regulations and policies which are too specific or prescriptive can become quickly obsolete and stifle the development rather than encourage it(3). Another issue which tends to be overlooked is that some AI regulations cannot be applied across all sectors. This is particularly true for highly regulated sectors such as insurance, which already have many frameworks in place. The governments thus need to ensure that new regulations do not contradict the existing ones.
But overall, these factors are contributing to a wider problem within the insurance sector – according to Sapiens’ latest survey, 74% of senior insurance executives view regulations as a barrier to a successful implementation of AI(4). So, what exactly are the latest and upcoming regulations in the UK and EU that insurers need to be aware of? We have summarised them below.
European Union
Financial Data Access (FIDA) - Proposed in June 2023. It will support the Open Banking services and create a base for the implementation of Open Finance in the EU. Its aims to expand the scope of customer data held by financial institutions that can be shared with third parties(5)
EU Data Act – In force since January 11th, 2024. It clarifies and sets conditions for how data generated in the EU can be accessed and utilised. It aims to facilitate data sharing, while ensuring fairness and protection of personal data(6)
Digital Operational Resilience Act (DORA) - Coming into force in January 2025. It establishes standardised security requirements for network and information systems of financial sector entities as well as their critical ICT service providers. Additionally, it sets up a framework which ensures that organisations can maintain digital operational resilience during ICT-related disruptions and cyber threats(7)
AI Act – Coming into force in 2026. It is the very first European legal framework that comprehensively regulates AI, and divides AI applications into categories based on the level of risk involved for the users (unacceptable, high, limited and minimal risk) – each category will be subject to specific obligations and rules(8)
United Kingdom
As of today, there are laws that apply to the use of AI (e.g. data and consumer protection, product safety, equality law, and other sector-related regulations), however, a comprehensive legislation (such as the European AI Act) is not in place(9)
AI Regulation White Paper – Published in March 2023. Compared to the EU, the UK is adopting a more flexible, agile approach. The paper sets out a regulatory framework for AI that includes 5 key principles which are to be interpreted and implemented by the UK’s regulators, such as the ICO, FCA and CMA. Regulators have been asked to outline their approach toward AI by April 30th, 2024(10)
Some regulators are already aligning themselves with the approach of the UK government, like the UK Financial Authorities that published their response to the white paper in October 2023(11). It seems that regulators are trying to prepare as best as they can for an upcoming legal framework for AI. In particular, many are focusing on customer obligations and cloud-based regulations, which are already impacting regulated firms such as insurers.
Technology challenges
On top of navigating the legislative regulations, many insurers face another obstacle – legacy platforms, on-premise technology and outdated infrastructure, which tend to be incompatible with the AI technology. And although over 90% of banks and insurers in EMEA have started to adopt modern cloud solutions – which are usually capable of supporting AI – less than 50% have moved majority of their key applications to the cloud(12).
One of the reasons is that modernisation is still perceived as too expensive, despite the fact that incumbent insurers can spend up to 70-80% of their IT budgets on maintaining legacy systems(13). This raises a question of how cost-efficient these systems really are – especially since modernisation of core IT systems can lead to a reduction in IT costs per policy by 41%(14). It therefore seems that investing into innovations can bring many benefits, moreover, will soon become a requirement if insurers want to stay competitive, increase productivity and provide better customer service by using AI.
However, one still needs to keep in mind that there are risks involved in moving to the cloud, such as privacy breaches, data losses, potential operational disruptions and integration challenges associated with overlapping systems in place. These risks need to be managed carefully, for example by:
Selecting a reliable and competent system integrator
Using third party consultants which can help with migration planning, vendor selection, compliance, governance and cost optimisation
Building ‘digital twins’ - i.e. systems that operate alongside the legacy system and don’t cause interruptions to the business(13)
Data quality
Yet another issue to tackle is associated with data that is processed by AI to receive desired outputs. In summary, bad data quality equals bad results. Insurance executives are widely aware of this issue as almost 93% of them consider data quality to be a barrier to implementing AI, which makes it the most severe issue out of the 8 different factors(4). However, poor quality can come in many different shapes and forms, such as:
Incomplete data (e.g. missing information or values)
Inaccurate data (e.g. due to human errors)
Inconsistent data (e.g. using multiple systems to record the same data but in a different format)
Duplicate data (e.g. when using multiple different systems or due to human error)
Outdated data (e.g. data that is no longer relevant or accurate)
Inaccessible data (e.g. due to lack of documentation or technical limitations)
The impact of low-quality data can be significant and lead to missed revenue opportunities, slower reactions to market changes and lower customer satisfaction and engagement(15). If insurers want to achieve high-quality data and receive the most accurate, reliable, relevant and unbiased results, the following activities are recommended:
Set up data governance frameworks and policies that outline correct procedures for managing, collecting, storing, processing and accessing data
Carry out detailed assessments to determine the quality of data prior to its processing by AI
Implement cleaning and preprocessing techniques to address any data issues (e.g. missing values, outliers, duplicates) and to standardise and normalise the data. In many cases, you can use AI tools for these tasks to speed up and automate the process and reduce the manual effort required by your team
Provide training to equip relevant employees with knowledge, skills and techniques on how to ensure that data meets the required quality standards
Even if you manage to complete all of the above, the task does not end there. High data quality always needs to be upheld and improved to maximise the potential of your AI solutions. There are 3 key things that can help you achieve this:
Keeping up with the market regulations and technologies to ensure that your policies and frameworks meet the latest requirements, and that you are using an up to date and relevant tools
Documenting all aspects of the data pipeline, which is essential for securing transparency in your data processes and achieving reproducibility. The documentation should cover elements such as:
Purpose and scope of the data
Data sources
Data descriptions
Cleaning and preprocessing procedure (including detailed steps and techniques)
Tools used
Challenges/limitations
Future considerations
Creating a dedicated team which monitors and maintains the data quality. This team can ensure that all standards are met through the use of KPIs, metrics or dashboards which display real-time state of data quality (e.g. percentage of duplicate or missing values)(16).
A final note worth mentioning on data quality is bias that can be present in your data sets. This can pose a significant risk to you as an insurer since AI algorithms tend to amplify bias and lead to discrimination of certain groups(17). Even recently, it has been discovered that car insurers have been applying an ‘ethnicity penalty’ in areas of England with large ethnic minorities – compared to other less diverse areas, the quotes were higher by 33%(18). Naturally, this triggered a big debate in the industry and many are already trying to address the issue. For example, a group of insurance experts devised a voluntary code of conduct in January this year, which aims to establish a standard for ethical and responsible use of AI in the claims settlement process(19). Apart from this, there are also individual actions that you can take and implement, such as ensuring that your data is diverse and comprehensive, and that you are regularly utilising human review which can challenge the fairness of your selected AI model(17).
Conclusion
While the three obstacles we discussed (regulations, technology, data) are major factors that you need to be aware of, perhaps if you are at the very beginning of your AI journey and not sure where to start, consider asking yourself these questions:
What are my goals and objectives that AI can help me accomplish?
Can my suppliers support me on my AI journey and meet my regulatory, data and technology requirements?
Do I have the right people with the right skills that are able to maximise the benefits of using an AI solution? Do I even know what the necessary skills are?
What does the move to AI mean for my operating model?
Implementing AI into your business might be intimidating at first, but the key is to choose a tool that aligns with your business needs and data characteristics. Try starting with small, low-complexity projects that you can use as pilots to validate AI concepts and assess their feasibility and potential impact on your business. Don’t be afraid to experiment with different techniques and approaches, collect feedback and try again.
And finally, although it might seem as everyone is already utilising AI and you are the last one that doesn’t, you are not as behind as you think – even a giant tech company such as Google has started to use generative AI for Google Maps only earlier this year(20). Furthermore, a recent study revealed that only 43% of insurers are using AI nowadays(21) – so if you’re interested in adopting this technology, there’s no better time to start than now.