Are some insurers going too far with conviction data?

Counter fraud has been blossoming in recent years. Once focussed on catching fraudulent claimants, the scope is now much wider, encompassing ways to both avoid selling them a policy and avoid even marketing to them. All this relies a lot on data analytics and its integration across insurer/consumer interactions. Insurers are looking for signals of fraud potential, and one of those I understand some insurers have turned to is data relating to convictions.

Insurers have always been able to ask a proposer about any convictions they may have. Since 1974 however, that right has been tempered by the Rehabilitation of Offenders Act. That allowed anyone with a conviction that was spent (defined by the act) to not have to declare that conviction, including to an insurer.

The purpose of the act is to facilitate the rehabilitation of offenders back into society. For not too serious crimes, and after a number of years had passed, the person would be able to act as if they had never been convicted in the first place. Without the act, people with convictions would struggle to gain a job and a home. This then increased the risk of them turning to crime again. The act helped reduce reoffending and that was good for everyone, including insurers. After all, they’re the ones who invariably have to pay for stolen cars or stolen contents.

Trawling for Signals

So what’s happening now?

I understand that some insurers are using predictive and inferential analytics to trawl for signals in their vast lakes of data that someone might have at some time been on the wrong side of the law. And they’re then incorporating those signals into decisions about that person’s potential fraud risk. That then influences whether they’ll be offered a ‘no quote’ or ‘go away quote’, or even appear on quote screens.

The logic behind this, so I understand, is built upon the principle of moral hazard and its notions about character. From this has emerged the idea that a history of crime, even if now ‘spent’, still indicates a propensity to act outside of the law. And in the hunt for insurance fraudsters, some view that as worth taking into account.

Remember that this doesn’t necessarily involve actual data about convictions. Inferential analytics draws out what data might infer about you. The data exhaust from the daily life of an ex-offender might just send out the odd signal or two that points to their past lives.

So are some insurers acting in contravention of the Rehabilitation of Offenders Act? Well, they’re not asking proposers to declare spent convictions. Instead, they’re simply trying to work it out for themselves, at a level of accuracy tailored to the business decision to ‘quote’ or ‘no quote’.

Three Levels to Look At

So does that mean those insurers are still acting within the law? That can be looked at on three levels. The first level approaches this from the angle of what the law says about information relating to convictions. So if the Act doesn’t say you can’t use inferential analytics on everyday data, then surely it’s within the law to collect it? That’s a neat question, but it’s also the wrong one to ask.

The second level looks at the intent of the act. The act says that someone whose convictions are spent should be treated as a rehabilitated person. They keyword here is ‘treated’, which is measure of outcome. If an insurer treats someone whose convictions are spent differently from someone who has no such convictions, then that would be in contravention of the act. A ‘fraud potential’ score influenced by inferential data on someone’s past conviction record would seem, to a non-legal mind like mine, to be treating them differently.

The third level is a logical one. Insurers benefit from reduced crime. Initiatives that help reduce crime, like the rehabilitation of offenders, will benefit insurers. So while some insurers think that identifying people with spent convictions will reduce insurance fraud, the market overall needs to remember that in the long run, increases in crime will result in them paying out for more theft claims.

Moral Hazard

So where does this leave the moral hazard justification sometimes associated with the use of data relating to convictions? The reality is that moral hazard is not something only of interest to insurers. They may have long been at the fore of narratives about it, but they’ve never had free rein to determine how it should be applied. The Rehabilitation of Offenders Act is an example of that. And while the act may have been written in analogue times, its intent and the rationale behind it are still very relevant in today’s digital times.

So what should insurers do? I would suggest these three steps to start with.

Three Areas for Challenge

Firstly, what controls are in place to ensure that people with spent convictions are not being treated differently? Note that I am not talking here about the direct or indirect use of data that sends some sort of signal about convictions. I’m talking about how consumers are being treated.

Secondly, look at My Licence. This is a facility provided to motor insurers by the UK Government, giving them instant access to driving licence information at the individual level. It so happens that certain motor endorsements are left on a licence record for longer than the period for them to become spent. This means that some endorsements will be on the record, but also spent. What controls do insurers use to ensure that spent endorsements on My Licence are not used in underwriting and claims decisions?

Is that a rather too detailed question? Not when you consider that in some circumstances, someone who acts upon information about spent convictions would themselves be guilty of an offence.

The third step takes that second step and extrapolates it into the data analytics world that counter fraud now sits in. How did the due diligence carried out by the Insurance Fraud Bureau for its recently publicised partnership to develop predictive counter fraud AI handle the question of treating people with spent convictions? And what controls has the insurance distributor who last year launched a self-learning AI fraud detection process put into place to address that same question?

Micro Outcome Data

Until a few years ago, I might have asked those last two questions with the nagging feeling that we might never know the answer. Outcomes data has for a long time been hard to come by. However, we’ve since had the pricing super-complaint, evidenced by Citizens Advice’s use of mass micro outcome data and confirmed with the FCA’s own data analytics.

Outcome questions like ‘are people with spent convictions being treated differently’ are now no longer too difficult to answer. Supervisory technologies have changed that. Organisations who can harness micro outcome data have changed that. This is just one part of the shift in power relations that is emerging around data and analytics. And insurers need to recognise this, and make sure they’re asking themselves some challenging questions about outcomes.