But online sales is made much more for books than pet supplies. The library of pet supplies is nothing compared to the vast variety of books. So selling books online with a good search and novel distribution model is what set Amazon apart from pets.com
Good question, and I think the answer has to be subjective.
My attempt at an answer: companies that either do things that couldn't be done before they were online because of the nature of the product/service, or where it couldn't be done before because the code has is complex.
So pets.com, really they were doing what shops had been doing for years, they just happened to do it online. Reddit on the other hand, there's not really an offline equivilent. Hipmunk, there's offline equivilents (travel agents) but the tech behind Hipmunk is what makes is a good company, not just "we've moved our travel agency online".
But yeah, no doubt there are YC companies that aren't necceasrily "tech companies", and on the other side there are companies not thought of as tech companies where technology plays a huge, huge role.
I had a door to door salesmen sell me a papa johns coupon book about 5 years ago. It cost 20$ and it probably saved me 10 times that over the next year. Honestly, I have trouble seeing Groupon as anything other that 'that' + email.
PS: I think people have been doing the same thing from at least the 1950's if not earlier.
Quick question: has YC funded any fabless semiconductor shops? (this isn't snark, I legitimately don't know). That has always been my representative sample of 'a real tech company.' I may be biased though, since I was brought up as an EE :).
I'd be very surprised, as the capital outlay to start even a small ASIC team is not small. A single simulator license costs considerably more than what the fine folks at YC invest in the companies they have under their wing.
The hardest part about fabless semiconductor startups is the lack of flexibility to pivot.
The time to design chips correctly is 10x-20x higher than doing the same in c++ (let alone a higher level language).
Example: in my previous company we needed to do some Jpeg decode in hardware. You have to buy a library for that. The library aint cheap so you have to evaluate vendors first, then work with lawyers to negotiate a licensing agreement. Then you design, implement, and test integration into your chip. All of this takes 2-3 man-months.
By contrast, we prototyped it in sw in an afternoon, and could be out the door with tests and production polish in less than a week.
You can use fpgas or low-NRE ASICs like eASIC to reduce the build costs, but engineering hardware is really slow.
