Fairness – I know it when I see it. Do you?

The idea of fairness is integral to almost every aspect of our lives. Our latest research shows consumers having strong inherent understanding of what fairness means, however dictionary definitions are somewhat one-dimensional and generally ‘under-baked’.

The Oxford dictionary gives us “the quality of treating people equally or in a way that is reasonable”. Webster is not much better, “the quality or state of being fair. Especially: fair or impartial treatment: lack of favouritism toward one side or another”. And Collins, “Fairness is the quality of being reasonable, right, and just”. None are particularly helpful or comprehensive in the round.

Consumer Duty outlines a version of fairness in requiring “good outcomes for customers” across four different duties, communications, price, and value, meeting needs and offering support. Importantly, firms need to provide evidence that these outcomes are being achieved.

Fairness also raises its head high in the ethics of AI. There are well-publicised examples of AI achieving poor outcomes in terms of fairness and a great deal of work has been done since to define principles for ethical AI.

So, what about insurance? Duncan Minty has written an insightful paper (https://actuaries.org.uk/media/z5rh5jhl/revolutionising-fairness-to-enable-digital-insurance.pdf) outlining key elements that should be considered for insurance offers to be fair. These include the fairness of access, need, merit, time, and crowds. Importantly, he also considers how these relate to common themes in insurance, namely the concepts of asymmetry of information, adverse selection, and the principle of pooling.

Given how fundamental the issue of fairness will be for insurers over the coming decade, we have broken the concept down into a granular set of dimensions. Dimensions built not only on centuries of philosophical thinking, but also on more recent work on ethical AI. They can be used to build a set of tests across the actuarial journey from product design and pricing through to underwriting and claims.

Reciprocity - fair exchange of value - “do as you would be done by”

In insurance this means paying valid claims in full. Customers have paid their premiums each month, now it’s the insurers turn. The use of loss adjusting to actively reduce valid claims may seem like standard practice – especially if everyone else is doing it – but it is a corrosive example of how the industry fails on a basic idea of fairness.

Reciprocity means delivering the peace of mind that insurance is based on. The often-quoted safety net insurance provides is only true if consumers trust and believe it will be there when they need it.? Consumers need to feel, believe, and trust the value in the safety net, regardless of their need to claim.

Process – transparent, just, distributive

This is not simply about easy and efficient systems. It is about clarity in why questions are asked, what the consequences are for incorrect information, and how decisions are taken throughout. It is particularly important in insurance since the value purchased is usually not immediately obvious (unlike most consumer goods, biscuits for instance). Fair process is an initial proxy for fair outcomes.

Reducing friction at the point of sale may reduce barriers to sale but if friction is simply being shunted along to claims or renewal then there is an obvious risk of unfairness by stealth. Yes shorter, lighter, quicker underwriting can give benefit to customers, but if it’s quicker to buy, but harder to (successfully) claim the process is clearly unfair. The fairness of process means understanding and managing the full process.

Process also means that cheaters/fraudsters should be prevented, and, if identified, suffer consequences. It cannot be fair for the honest majority to simply pick up the detriment of fraudsters not obeying the ‘rules of the pool’. Insurers not fully cracking down on fraudulent whiplash claims is as much an afront to customer fairness as it is insurer ratios.

Respect – non-discriminatory, unbiased outcomes

Human beings have equal worth. An obvious statement, but one which needs actively protecting and promoting when insurance decisions are being made. Processes or decisions that (even passively) elevate some above others have profound impact, materially and otherwise.

Examples abound. Prioritisation of organ donation using an algorithm based on “maximising increased likelihood to survive 5 more years” was found to systematically prioritise older people simply because of base life expectancy. There is no malice here and no conscious bias, but it is still not fair and still does not absolve the owners of the model from outcome accountability.

In practice, this means putting governance and design guidance in place to ensure that models do not systematically result in some groups having poorer outcomes or detriment based solely on their membership of a group (e.g. race or age), or models that build in a bias or systematically disadvantage people based on design decisions.

Merit – riskier people should pay more

Most agree that you should not be punished for things over which you have no control, but the vast majority of people also agree that riskier people should pay more for insurance. Merit is a complex and nuanced dimension of fairness. ?

This higher risk, higher premium mentality is more subtle depending on the context. If in charge of selecting a football team for a friendly match, people would not just select the best players – they are much more likely to select those who came most to practice beforehand. This means that the desired outcome or purpose significantly impacts how we view merit. In insurance this has big implications for pooling vs personalisation, moral hazard, and adverse selection.

Examples of merit in practice are also wildly context reliant and nuanced. A simple example of premium increases based on a percentage increase over time (5% say), rather than an absolute increase (￡10 say) means the person identified early on as riskier pays an exponentially increasing premium over time compared with others. Should commercial models be required to prioritise absolute margin over percentage margin? Regulators in the US appear to think so.

And what about living in a flood area? If you chose to buy a house somewhere knowing it was riskier but cheaper, should you be required to agree to high flood premiums at the point of sale? What responsibility does the government have to maintain flood defences vs. communities? Should insurers take back Flood Re customers? Or should they exit sectors where these choices are complex and political? Exit too many areas because they are difficult, and the question raises itself – what is insurance for?

The very concept of fairness goes back to early philosophers and has been enriched and developed over time. What societies view as “fair” changes, but the building blocks they use to make those choices does not. Having a defined set of criteria for fairness offers a practical and more complete approach to designing, implementing, and measuring compliance with the principles of Consumer Duty and ethics. For insurance and elsewhere.

?

?

?

?