Importantly, such discrimination may still be reasonable, even if LIPL could have made alternative decisions, or even decisions that are considered fairer to some customers.78 Provided it is reasonable, the data exemption should apply and such discrimination would not be unlawful. The inclusion of occupations and occupation categories changes the nature of any possible discrimination. If included, the relevant question to determine whether any direct discrimination occurred becomes whether men with particular occupations are charged higher premiums than women with those same particular occupations. Again, provided this discrimination was supported by appropriate data and was reasonable, it is likely that the data exemption would apply. Conclusion There are many variables that could be included into the analysis that might change the outcome. As set out above, inclusion of occupation categories changes the outcomes, and the inclusion of additional occupation subtypes within the categories further changes the outcomes. Inclusion of other datasets might change outcomes further. Insurers must make decisions regarding the collection and inclusion of relevant data, and the design of the model. Summary • An insurer does not necessarily need to include all data relating to risk in its model, where correlated with a protected attribute covered by an insurance exemption. The insurance exemption requires reasonableness, not perfection. • The existence of potentially better options (including the existence of potentially relevant data which remained unused) does not necessarily mean that any discrimination arising from chosen methodology is not reasonable. • The data exemption will apply where any discrimination arising from the methodology was based on data on which it is reasonable to rely and was reasonable having regard to the data and other relevant factors. The availability of the data exemption depends on discrimination being based on appropriate data and being reasonable having regard to the matter and any other relevant factors, which will turn on the relevant circumstances of each case. Guidance Resource: Artificial intelligence and discrimination in insurance pricing and underwriting • 2022 • 37