FWIW, it appears that at least some of the ancillary functions get done by other people. According to [1], social media marketing and accounting are two that he doesn't handle personally.
> I am in utter disbelief one person can do everything, even if I account for being an automation freak.
As someone who works in data acquisition, and someone who has evaluated BuiltWith along with other providers of tech install data, I can get a rough idea of how it stays as a one-man shop.
Whether deliberate or not, the more hairy type of signal analysis tends to be missing from BuiltWith's dataset. Unless it's exposed via DNS or direct webpage code, BuiltWith doesn't detect it. This isn't necessarily a bad thing, as it keeps it maintainable. Especially because you can use your historical data to identify changes in signals and be able to reactively update them as they change instead of use resources to proactively monitor all of them manually. But it limits it to a "public web facing" tech install base. You can greatly expand this by including secondary signals like parsing SEC files and job postings and whatnot, which will capture a lot of non web facing technologies. But those are a lot more complicated signals that require a lot of filtering/processing to reliably leverage, and would likely grow beyond a one-man shop like BuiltWith.
Salesforce is a classic example of this. It's generally an internally focused software, but if you use Partner Portals or Web2Lead forms, it'll be exposed publicly. Builtwith[2] identifies 12.1k live sites using it and 30k historical sites using it. HGData[3], which provides similar data but uses a wider range of signals, can identify 92k sites/companies currently using it.
> crawler needs to handle tons of edge cases
Not particularly. If rendering the page, sure. But when doing pattern recognition within the source code itself, edge cases that'd bork rendering are mostly moot.
I'm not at all disparaging BuiltWith - it's a solid product for what it is and a bargain for the price charged. But it's also a very limited product which, when destructured, hints at a lot of decisions made with the intention of being maintainable as a one-man shop. In the space it occupies, it also has a really solid pricing model if you can get by with using the limited dataset it provides. But if you need visibility into any internal software or tools or hardware, it has much more limited use.
[1] http://www.hostingadvice.com/blog/builtwith-more-than-how-we...