This is I think the core issue that almost all of the gig economy startups face, even the gigantic ones like Uber.
The trick with on-demand is that we're not the first into this space - low-tech versions have existed for years, but largely priced out of reach of the mainstream. This pricing was necessary because of the fundamental nature of on-demand.
After all, getting someone to drive to a restaurant, order for you, and bring you the food has some pretty immutable built-in costs.
Software can help reduce these costs, but IMO we've vastly overestimated the degree. It turns out in most cases the technology can't reduce costs enough to shift the economics from "upper class only" to "middle class".
There have been two responses to this reality - one is to force the issue by subsidizing the unit economics with VC money, and pray that you can actually figure out the economics of bringing on-demand to the middle class before the money train derails - this is the Uber strategy (specifically, UberX, which is heavily subsidized in many markets).
The other is to make the service suck - I'm not sure why any company thinks this is a viable strategy but we've seen many pursue it. Homejoy is the spectacular nuclear-explosion example but there are many others.
It turns out human labor is expensive, hard to scale, and has a lot of built-in costs that can't be eliminated by smart software.