The debacle around how recent exam grades were decided in the UK provides insurance people with some lessons on what might befall them if current optimisation trends in claims continue. Both involve the use of algorithms in a way that fundamentally changes the assessment process. And like the exam scandal, claims people could find themselves accused of systemic unfairness. If claims people have to roll-back decisions as the qualifications bodies had to, the consequences for the sector will be enormous.
Some background first. The pandemic resulted in all UK school exams being cancelled. Teachers were asked to assess student performance and submit predicted grades for 16 and 18 year olds. Qualification bodies then used an algorithm to adjust those predicted grades, ostensibly to produce fairer results.
The outcomes that vast numbers of students experienced were quite different however. Many students were down-graded, some by several grade levels. Commentators were quick to point out patterns of distinct unfairness in these downgrades. As the scandal erupted, leaders in the four nations were forced to remove the algorithm’s influence from their assessment process.
How this relates to Claims
So how does all this relate to insurance claims? This is best explained by examining the influence of the exam grade algorithm. In previous years, the student took the exam and received a grade, in what was essentially a one-to-one situation. In essence, that student took that exam and received that grade. There was then some secondary adjustment of grade boundaries to contain what is called grade inflation. This was applied across the board, not by school or student.
This year, the algorithm took a different approach. It took predicted exam grades and adjusted them at the school and then pupil levels. This turned the previous one-to-one arrangement into a one-to-many arrangement. Instead of the student performance being put first, the historic school performance was used to set localised grade boundaries. The impact on social mobility was obvious. The media became saturated with stories of individuals whose life chances had been ruined by the algorithm.
Let’s now compare this to trends in how claims in many lines of volume insurance are being assessed. Historically, an insured loss resulted in that policyholder submitting their claim and receiving a settlement based upon that loss. This was essentially a one-to-one situation: that policyholder, their loss, their settlement.
Optimising Settlements
Move forward to the present and the increasing use of algorithms for optimisation of claims settlements. Some insurers are now looking at your claim in relation to similar ones that it has previously settled for other claimants. If that analysis shows that claims like yours, worth say £1,000, can be settled for say £950, then the claims system would put forward that reduced offer.
The logic is that if someone is willing to settle for £950, then what’s wrong with offering it? So while underwriting people talk about ‘willingness to pay’ for people who will pay a bit more each year, claims people talk about ‘willingness to accept’, for people who settle for a bit less.
So just with exam grades, insurers are moving their settlement strategies from a one-to-one situation, to a one-to-many situation. And in doing so, the insurance sector is creating for itself a reputational and legal cauldron which I believe, just as for the qualifications bodies, puts them into a lose lose situation.
Why Claims differs from Underwriting
Some of you will point to the one-to-many approach commonly used in retail GI underwriting. Why can’t it be reproduced in how subsequent claims upon those policies are settled? It can’t be because the two situations are fundamentally different.
A policy is a contact under which the basis of settlement is not premised on other peoples’ losses, but upon the policyholder’s own loss. Just because a person, due to their personal circumstances, is willing to accept an immediate but reduced settlement, doesn’t mean that insurers should then offer it. The notion that I hear talked about, that optimising claim settlements helps keep premium levels down for everyone, is flawed, and regulations confirm it to be so.
Insurance people need to think outside of their sectoral bubble. They need to take some time to consider, carefully and critically, the consequences of trends like claims optimisation. One technique they could use is the ‘front page’ test – would I be happy to see what I am doing explained on the front page of a leading business newspaper? Would I be prepared to defend the practice on leading news outlets like BBC Radio 4’s Today programme? If not, then why do it?
Some executives have defended such practices to me by arguing that a competitive market makes it impossible for them to do otherwise. This version of what is essentially the ‘everyone else is doing it’ rationalisation has been called ‘the five most dangerous words in business’ for good reason.
Some Obvious Next Steps
So what should insurers do now? Here are some obvious steps:
- Establish how optimisation in claims is positioned on your corporate risk schedule. Are you happy that its grading has been calibrated with enough of a critical eye?
- The ethical risk assessments currently being prepared for 2021 should not just log optimisation in claims as an ethical issue, but be unpacking it to critique the different techniques and data being used.
- The use of certain metrics for optimising claims should be seen as absolutely red ethical flags. Do your senior management functions understand, and are happy with, the level of personal accountability they’re carrying in such situations?
- If you’re a Chartered Insurer, have you prepared a case for claims optimisation, in the context of your primary obligation to always act in the best interests of the public? And have you tested it?
- If your firm makes use of external software houses and data brokers for claims optimisation, has your due diligence identified and reviewed the ethical issues that it raises?
There are many ways in which data and analytics can improve how claims functions work. What matters is that in weighing up how those improvements are designed and introduced, a critical eye, a customer perspective and an ethical point of view have to be considered. I predict that if such considerations are not now applied to optimisation in claims, then a huge mis-settlement scandal will erupt within five years. The exam grade debacle provides insurers with a sense of what it will feel like.
