From the moment you start typing your details into an online quote, to the loss details you give to a claims handler: across the lifecycle of a policy, insurers are looking for signs of suspicious things. Such signs then trigger a change in premium, a different type of service or even a declined claim.
Remember though that this is not about hard facts. Suspicions are beliefs or impressions that someone or something is questionable or dishonest. So those beliefs or impressions could range from a ‘little inkling’ to a ‘virtual certainty’. The latter usually kick starts a human assessment of the claim and an ultimate decline.
Before a human becomes involved, there are lots and lots of digital assessments being made, across all sorts of parameters. These digital assessments are invariably brought together in some form of ‘fraud score’. That score then turns the dial on things like inception or renewal premiums, and the nature of the service response.
Getting it Wrong
Sounds all very straightforward. What could go wrong? Well, Liberty Mutual found out to its cost that things can go very wrong. It was fined £5.28 million by the Financial Conduct Authority in 2018 after it plugged in some counter fraud software into its digital decision systems for a mobile phone scheme. What they forgot to do was check on how the dials and levers within that software had been set. It turned out that they’d been set very wrong, resulting in lots of complaints about claims being unfairly turned down.
My point here is that suspicions vary in strength and depth. How you set the dial for how your digital decision systems handle them will directly influence the outputs that are generated. So what then is a fair level of suspicion?
There isn’t one. It depends on the circumstances. Different scenarios will need their fairness to be individually judged. This is a human judgement, not a statistical calculation. Digital systems can’t set fairness parameters (more here).
Is this a new process to add into a probably already complicated system design? Not really. Part of any process design is to set the significance levels for a variety of parameters, either individually or in conjunction. Those significance levels are equivalent to fair levels of suspicion. What is needed though now is for the setting of those significance levels to be understood in fairness terms, not just statistical terms.
Surges in Suspicion
Let’s bring in some data to shine a light on this. Here are some numbers from 2022 about application fraud in UK personal motor insurance, compiled by the Insurance Fraud Bureau.
What stood out for me in these figures was the increase in the number of suspected cases from 2017 to 2019. It rose from 226,522 cases to 631,631. That’s a near threefold increase; in three years.
It turned out to be impossible to get any detail about what was behind this huge increase in suspicious cases. Remember this is application fraud, not claims fraud. This means that it’s not down to external factors such as an increase in accident rates or the like.
I was left to conclude that the near threefold increase was down to the sector in some way resetting the dial on what it judged to be suspicious. It has reset the significance levels within its digital decision systems. Yet to what extent had fairness played a part in this?
There’s two ways of looking at this. On the one hand, the dial might previously have been set too low. Or new forms of fraud necessitated a finer tuned decision system. Or 2019 might have been an aberrant year.
On the other hand, a move of the dial significantly upwards would support the sector’s public relations campaign to strengthen support from the Government on counter fraud measures. I appreciate that that’s a rather ‘pot half empty’ view, but I’ve come across that view on a number of occasions.
Closed Doors
So which way of looking at this is correct, or at least the more likely? I don’t know, and that’s my key point here. The sector and its suppliers of counter fraud decision systems make these decisions behind closed doors.
Two reasons are given for that closed door. One is that these are ‘commercially secret’ decisions, so there’s no need for them to be disclosed. The other reason is that fraudsters would exploit such disclosures. There’s something in both arguments, but not enough. I’ll explain why.
Fraud is a crime. Parliament decides what it means, the courts apply those meanings and the police collects the evidence to bring before those courts. This is the criminal side; the legal system replaces the police on the civil side. Yet only a tiny tiny proportion of suspicious insurance applications or claims come before a court. The vast vast majority of them are handled by insurers.
Concentration of Power
So what, you might ask. Surely it makes sense not to flood the courts with hundreds of thousands of little cases. And there is sense in that, subject to an important caveat.
The suspiciousness of an insurance application or claim is being defined, applied and judged upon by insurers. Or to put it another way, insurers are acting as policy makers, courts and police combined, all in one. That’s a significant concentration of power.
So what, you may ask again. Well, it matters because how those dials are set for application fraud in digital decision systems affects lives. Everyone one of the more than 400,000 additional people brought within the scope of motor application fraud between 2017 and 2019 will have paid more, endured less empathic service or have received less or nothing in settlement. Because of how the suspicious dial was been moved. Remember that this is mandated cover, so if you can no longer afford it, the outcome could be a lost job, college place or the like.
Two Key Words
There are two key words here: fairness and accountability. How insurers operationalise suspicions has to be done fairly. And insurers have to be prepared to be held to account in how they go about this.
At the moment, there is very little pressure on the sector to demonstrate that fairness and accountability. For sure, both are key words in regulations but they have not so far being brought together and applied to application fraud in insurance. I find this surprising, given how every motor and household inception, MTA and renewal is subject to that suspicious once-over.
Will that change though? I think there is a reasonable prospect of this happening at some time in the next five years. We’re not going to see descriptions and numbers put into the public domain. Instead, independent authorities will be placed within the governance structures for institutions overseeing counter fraud. For example, in the Insurance Fraud Bureau and the General Insurance Fraud Committee.
Interim Steps
In the interim, it would make sense for insurers to set down the rationale for how the dials on their digital ‘suspicion machines’ have been set. This should include how this rational has been reviewed and exposed to challenge, for example by independent board members in that firm. And this should also include something about output testing to check that the suspicion machine is actually doing what it was intended to do.
The danger here is that if this is not done, then when asked how they decide what is suspicious or not, insurers will have little more to say than ‘what we think is suspicious or not’. And while the sector may find this a reasonable response, policy makers will not.
The irony here of course is that policy makers will look suspiciously at how insurers handle suspicion. Both will be subjective, but only one will have in their hands the levels of real power.
