Is claims fraud subject to a sex discrimination exemption?

In identifying claims fraud, some insurers have traditionally used sex/gender as a risk factor.

Whether this is allowed depends, of course, on the jurisdiction and the laws that apply. I can’t cover all here, so will focus on a selection:

• EU: the Gender Directive forces “unisex” premiums and benefits.
• US: no single federal rule. A handful of states ban sex in insurance pricing; others allow it if backed by data.
• Canada: insurers may use sex in pricing if they publish credible data.
• Australia: the AHRC and the Actuaries Institute produced this guide. It says discrimination laws recognise that insurers may need to apply some discrimination (sex, age, disability, but not race).

Some don’t allow it at all. Others do, as an exemption to discrimination laws, if you use credible data to justify it.

Now, the guidance from the AHRC and Actuaries Institute focuses on pricing and underwriting. It is tempting to take the guidance and apply it to other aspects of insurance business, including claims fraud.

In this article, I’ll explain why I don’t think a sex exemption can be applied to insurance claims fraud.

How the exemptions work

Where separating between the sexes for pricing and underwriting is allowed, it needs to be based on data that proves the differences in risk. There must be data, as this is the only type of exemption allowed.

Whether that is ok, or morally justifiable, is beyond the scope of this article. But let’s assume that it is ok. The reason would be that the observed behaviour is not the same between sexes. The risk associated with one sex is different to the risk associated with others; therefore, someone of the riskier sex category could be charged more. Again, no judgement on the validity of this line of thought.

However, when determining the likelihood that a claim is fraudulent, this argument doesn't hold the same weight.

But men are more likely to commit fraud

The prevailing theory was that men were more likely to commit fraud. But there are at least five problems with this:

1. Norms: it is based purely on gender (sex at birth), and sex data is no longer binary. Nowadays, which of them would you use? Is sex a better indicator, or should you use gender, or both?
2. Opportunity: most of the research comes from a historic time when there were significant imbalances in opportunity; for example, with men over-represented in the workplace, and at higher levels of seniority (where the opportunities existed).
3. Not targeted: the sentiment is quite broad, largely driven by employee fraud, and not focused on insurance claims fraud.
4. Survey data: some surveys focus specifically on insurance fraud, like the UK Insurance Fraud Bureau’s “Fraud Statistics” (IFB YouGov survey May 2024). But these are statistics based on public perception, not actual fraud.
5. Data size: the total sample size in the IFB survey was 2,072 adults (against a population of over 40m). Similarly, an ACFE report (we explore that below) was based on 1,921 cases of fraud; not insignificant, given that the cases were considered in detail; but can that be extrapolated to all known and unknown fraud globally? They represent interesting information, but don’t warrant an exemption.

There is more recent research that appears to back up the original claim. A 2024 report by the Association of Certified Fraud Examiners (ACFE) stated that “Women committed fewer frauds and caused lower losses”, and by quite a significant margin (men: $158k median loss and 74% of cases vs. women: $100k median loss and 25% of cases). However, subsequent research suggests that the differences are not as stark as the report suggests. This 2024 research report that used the ACFE data found that the numbers converge once role and authority were controlled for. 

The bottom line

Focus on behaviour, not personal characteristics. Don't rely on an exemption that isn't relevant.

Beyond fraud claims, if you are using an exemption, consider whether it is based on reliable data.

Disclaimer: The information in this article does not constitute legal advice. It may not be relevant to your circumstances. It was written for specific algorithmic contexts within banks and insurance companies, may not apply to other contexts, and may not be relevant to other types of organisations.