Afterpay uses a sophisticated behavioural quanititative credit...

Afterpay uses a sophisticated behavioural quanititative credit risk model. This model uses correlation analysis to figure out how risky a person/product/merchant is and these type of models are always learning as they receive more and more data. This is likely also the reason why default rates in newly entered markets are typically slightly higher because a consumer's characteristics in Australia don't match up perfectly with a consumer in the US. The model would be learning about each individual customer on an ongoing basis and use correlation to put blanket rules in place.

These models uses hundreds of variables to create a single score that ties to a limit. Each variable has a different weight depending on it's importance. For example, a late payment would be one of the highest weighted variable and instantly drop your score significantly lowering the limit, in some cases reducing it to zero. I'm not sure what other variables feed into the model but they would be quite sophisticated.

For example the model would probably note if you go from consistently making 3 purchases a month to making 1 or 0. This would flag that something is wrong. Of course the model doesn't know if this customer is being prudent or not, instead it would take the conservative view that the consumer has uncertainties on unemployment.

Anyway as mentioned before each "end score" would be tied to a limit. For example if I had a score of 88/100 then maybe my limit is 1.2k. If I had a score of 50/100 I might only have a limit of $750.

Management have two ways to use this for risk mitigation purposes.

1) One would be to drop all limits. Even if my score doesn't change from 88/100 my limit might drop to $1k with the purpose of lowering risk and exposure. It is quite likely that this might be in place for a limited period. For example if in a COVID-19 scenario I keep making purchases and keep paying on time, the model would recognise that, without exactly knowing my situation, that I still have a job or still have an income. It's quite likely my limit would increase (In fact my limit has indeed increased a couple of weeks ago after dropping about 5 or 6 weeks ago).

2) The second risk mitigation isn't really related to management decisions. It would simply be that the quantitative model picks up more inconsistencies or more people making late payments etc. So naturally you'll find people's "end score" drop in these kind of circumstances, again lowering limits and reducing Aftyerpay's exposure.

I hope this helps.