It's not just about reducing the absolute number, it's about reducing the number of possible interactions.
Let's say there are 20 employees in the store fulfilling orders, and they work in two shifts during the day. So that's 40 people in the store that have the potential to interact with each other (though it's really more like two separate groups of 20). If they do the curbside dropoff in a safe manner (that is, without physically interacting with the customer), then it's mostly those 40 people risking transmitting the disease to each other, and that's it.
If you instead allow 20 random people into the supermarket at a time (in reality it's going to be more, which is even worse), and they spend an average of 45 minutes in the store, and their entry and exist times are randomized throughout the day, you have hundreds of people who could potentially interact with and infect each other.
Employees will just naturally be much faster and efficient at gathering groceries than customers will be. They know the layout of the store and where everything is better, and they may also know that certain things are not available, so they won't waste time looking for them. They can even multitask and take care of more than one person's order at a time. They won't be wandering around making impulse purchases, but will be sticking to a fixed list of items the customer ordered. Checkout and bagging will go much faster. This bit is optional, but they could even optimize things further by grouping the most common items in one place, and not stocking shelves like they usually do[0]. I imagine you can serve the same number of customers with much less than half as many people in the store.
There's also the matter of discipline when it comes to distancing. I imagine the supermarket employees are pushed hard to maintain social distancing with each other while working. When I went to the supermarket a week ago, probably a good 20% of the customers I encountered were failing at distancing, further increasing risk.
[0] Consider that placement of items in a supermarket designed for customers wandering around is intentionally not set up for efficiency! They want you to walk through the entire store when making typical shopping run, in the hope that you'll make a bunch of other impulse purchases.
To add to what kelnos said, curbside pickup is better for public health because it limits the number of customers in the store, reducing vectors. It's also better for the workers themselves. Wegmans and a bunch of other grocery stores are using Instacart shoppers for curbside order fulfillment. Comparing that to delivery, an Instacart shopper may go into 10+ different stores on any given day while they're working. Not only is it inefficient, it significantly increases vectors and endangers their health more.
As thecodemonkey has said, we made an intentional decision to only support curbside and not delivery. We've also encouraged people to add items to their own cart for elderly or carless neighbors who can't place their own pickup order. (We also organized a neighborhood effort, before this app, to help people get groceries and prescriptions.)
It would be ideal everyone were able to do curbside/pickup only. But that's not realistic given the short timeframe of this crisis.
In the UK, the government seems to have shared a list with Sainsbury's of high-risk people. We had a friend in the UK ask us to add it, so I tried to sign up, and they said I couldn't because my email wasn't on the list from the government. I don't see such a thing being possible here in the US.
This app isn't perfect, but we released it because we think it helps in a small way. And if we all help in a small way, in whatever our own way is, we'll get through this.
Does it involve something other than swapping out customers picking their groceries off the shelves for employees doing the same?