The issue, is to find a happy medium between pooling and more responsibility-inducing pricing.
Insurers have always used data and mathematical models to analyse and manage risk. As highlighted by Florence Picard, President of the École Polytechnique d’Assurances (EPA) and manager of the French Institute of Actuaries’ Big Data working group, “data constitute the raw material of actuaries”. Thanks to Big Data, AI and algorithms, insurance companies now have access to a much larger quantity of data, from various sources, and more importantly the means to process these data.
Finer risk assessment
The use of these digital technologies provides insurers with many benefits. Firstly, AI makes it possible to optimise internal processes and to improve customer experience, in particular in the event of a claim. Thanks to AI, certain procedures can be automated enabling a more rapid response. One can for example imagine implementation of a computer vision algorithm capable of analysing photos sent by the insured party, or a chatbot available 24/7. In this area, parametric insurance opens up promising opportunities. This solution, which is based on blockchain and smart contracts, enables automatic compensation that is triggered when an event occurs (generally, when a meteorological index, measured by a satellite or a weather station, exceeds a certain threshold). It does not require the intervention of an expert to estimate damages, which reduces cost and speeds up compensation. Originally deployed, notably by regional development banks, to protect African farmers against climatic hazards, parametric insurance is now starting to expand into other sectors that are sensitive to the weather (such as agribusiness, transport, energy, etc.) and is opening up to private individuals. For example, Fizzy insurance, offered by Axa from 2017 to 2019, enabled them to automatically obtain reimbursement of a delayed or cancelled flight. With Axa Climate, its entity specialised in climate risk management, the French insurance group is well-positioned in this promising market, the world over – starting with Africa – however, InsurTechs are also positioning themselves. In West Africa, we can mention Malian startup Oko, winner of the Orange Social Venture Prize in Africa and the Middle East in 2018, which provides small-scale farmers from emerging countries with an affordable and automated insurance solution.
But the great promise of insurance algorithms is better knowledge of the customer and finer risk assessment based on the combination of “traditional” data, from risk-declaration forms filled-out by customers for example, with new data on individuals’ behaviour or their environment.
The ethical issues
The use of these data nevertheless raises questions. In its “Report on the ethical implications of algorithms and of artificial intelligence” (Rapport sur les enjeux éthiques des algorithmes et de l’intelligence artificielle), the French data protection authority (CNIL) states that the personalisation of content and services resulting from the omnipresence of algorithms could potentially jeopardise collective logics. Several stakeholders fear in particular that this dynamic could call into question the principle of risk pooling, one of the pillars of insurance. In its report, the CNIL also underlines that “algorithms and artificial intelligence can create biases, discriminations, and even forms of exclusion”. Individuals that are judged to be “at risk” could see the cost of their premiums increase according to their state of health (or behaviour that is deemed risky, such as alcohol consumption, lack of physical activity, or poor diet), or even be rejected from insurance.
These questions arise in particular for the inclusion of health data in the scope of complementary health insurance. Quoted in the CCNE (French National Ethical Consultative Committee for Life Sciences and Health) and CERNA (French Committee for the Ethics of Research in Information Sciences and Technologies) “Digital & Healthcare” report, France Assos Santé notes for example that “If insurers know what health conditions their policyholders have, the categorial solidarity of supplementary insurance might be adjusted to take account of the level of risk of the condition in question. This is what is called customer stratification on the basis of risk.”
For Florence Picard, it is nevertheless in the insurer’s interest to maintain the pillar that is pooling. “The more this pooling is based on a large segmentation, the more the risk is manageable, as the difference between the tariff established or the provisions calculated is low. The more they refine the categories, the higher the risk of error.”
Yet, insurers are already offering certain contracts based on behavioural data to customers who so wish. For example, “Pay how you drive” consists in adapting the insurance premium to the driving style, a personalisation of car insurance that is made possible by the installation of a box in the vehicle or a mobile application on the driver’s smartphone. Could this kind of insurance be extended to other types of insurance? It’s unlikely, according to Florence Picard, to whom the idea of individual pricing doesn’t make sense. “The right price does not exist. Premiums are used to pay for accidents that happen to only a few. For all those who are not faced with an accident, the premiums will always be too high; for the others, they’re not high enough. It’s the actual principle of insurance. The issue, she continues, is to always find a happy medium between pooling and more responsibility-inducing pricing.”
The regulatory dimension
What about the regulatory framework? First of all, the law forbids the use of certain data, such as gender or “race”. Insurances and complementary healthcare insurances can thus not take these data into account in establishing their pricing. In addition, the General Data Protection Regulation (GDPR), which follows the main principles of the French data protection act (loi Informatique et Libertés de 1978), reinforces the protection of personal data and aims to give citizens more control over their data. It expands the notion of health data, the use of which is forbidden except in certain precise cases as defined by article 9. Article 22 also establishes the right to not be subject to entirely automated decisions, that could be the result of profiling, which significantly affect the person involved (such as the application of higher tariffs).
However, the law does not take into account all of the issues raised by algorithms according to the CNIL, which is in favour of the extension of the loyalty principle of algorithms to take into account their collective effects. This principle is based in particular on the idea that an algorithm’s criteria must not go into head-on opposition with certain large collective interests – such as pooling – nor have the effect of reproducing or reinforcing discriminations.
Insurance players themselves are making commitments. For example, the French Institute of Actuaries’ Big Data working group managed by Florence Picard has launched a reflection on the evolution of the actuaries’ code of ethics in the age of Big Data. They are creating a working group on ethics and a professional standard for the use of data.
The aim is to find the right balance between exploiting an increasing amount of data, which is necessary to develop the uses of AI, and certain high principles (privacy, non-discrimination, etc.) aimed at protecting citizens.