I'm basing it on a rough mental model. Let's assume there is zero difference between the quality of full time people and part time people, but, being people th

I'm basing it on a rough mental model. Let's assume there is zero difference between the quality of full time people and part time people, but, being people there is distribution of people that a) need to be trained over time to handle certain on-the-job situations (e.g. dog in yard, etc...), and b) the system needs time (or some number of deliveries) to discover that some people are basically not fit for the position - including people signed up with the intent to scam the system. Let's compare two groups of employees, a full-time group and a part-time group. The first thing that might surface in this model is that the part-time group requires more people to fill the same number of deliveries that the full time group can fulfill. So, in terms of numbers of bad deliveries, you'll have more failures just because you had to draw a larger number of individuals for the part-time pool vs the full time - even if quality of individuals for the full time vs part-time are exactly the same. Adding further assumptions doesn't improve things. Maybe one might assume that part-time employees are more likely to move on to other jobs and leave the Flex workforce - well that's more churn and you get more people in category b, and more people temporarily needing training/experience. Maybe the distributions aren't exactly the same between full time and part time - one might theorize that scammers are more likely to take jobs under flex (maybe even under multiple aliases to keep from getting caught)... again this makes the performance under Flex worse... Maybe even given the quality differences, it only ends up making sense to offer Flex delivery on the periphery of where it doesn't make cost sense to install or expand full-time infrastructure.