What you're working up to is a reverse auction for delivery service. Who is willing to do a batch of deliveries for the least price? Constrained time TSP+Knapsack, a very challenging problem. (Because you can choose many different subsets of delivery jobs, run TSP for each, and calculate a bid price for each job that will win, or if you don't, minimises regret.)
The real problem with instacart is that some orders are better than others. If the delivery fee reflected the real costs then this wouldn't be such a problem.
EDIT s/your/you're. I swear my phone's predictive text is getting antagonistic.
> The real problem with instacart is that some orders are better than others.
You don't need to eliminate differences to fix the problem - simply minimize them enough that it isn't worth building or buying a bot to give yourself an edge.
A simple solution would be to look at how quickly each order is snatched up, and reduce the price paid for orders that some AI decides will probably be snatched up quickly. I bet some simple linear model consisting of total travel distance, weight, number of unique items, and dollar value would suffice.
All of this is pretty simple. Instacart simply needs to block ability of those that make an order from decreasing the tip after the order is live and automatically charge the shopper a penalty in the amount of the tip and service fee for accepting an order and not completing it.
The first will remove incentives of customers to create orders that are tasty for shoppers under the false pretenses.
The second one will make it costly for shoppers to accept good orders that they cannot complete.
The real problem with instacart is that some orders are better than others. If the delivery fee reflected the real costs then this wouldn't be such a problem.
EDIT s/your/you're. I swear my phone's predictive text is getting antagonistic.