For the last couple of decades, Google has been in an odd situation:
- They make a ridiculous amount of money.
- They already have a ridiculous amount of money, and also engineering talent and compute capacity.
- Almost all that money comes from one single revenue stream.
- That one revenue stream is very vulnerable. (If everyone in the world started using ad blockers today, Google would start going out of business tomorrow.)
So they were (and still are) extremely interested in trying to diversify that revenue. And, reasonably, tried to use those strengths to shore up those weaknesses.
That meant taking a lot of gambles on fairly wacky ideas in the hopes that at least one of them would pan out in a big way. It's honestly not very different from how a VC works, except that instead of investing naked dollars they were investing engineering time and infrastructure.
So far, none of these projects has turned into anything big enough to meaningfully diversify their business model; they are still almost exclusively an ad company. But it's probably still wise for them to keep trying.
(Source: I worked for Google for a long time.)
IMHO the two biggest factors in Google's reputation for cancelling projects are:
1.) A bottom-up culture that grants engineers and low-level managers a lot of autonomy and encourages them to generate their own ideas, with the expectation that not all ideas are going to survive.
2.) A performance review system shaped like a pyramid scheme.
20% time is another manifestation of #1, but it's #2 that's particularly insidious. To get promoted, you have to demonstrate impact (usually in the form of launches), challenge (in the form of tackling hard problems), and leadership (in the form of influencing others). The easiest way to get all 3 of them is to launch (impact) a hard problem (challenge) with other people (leadership). Except that once you've done that, the maintenance work required to actually polish it up is not flashy enough to generate impact, not hard enough to be challenging, and usually done by solo developers without leadership. Nowhere in the performance review system are things like "bugs fixed" or "minor user annoyances averted" ever rewarded.
So if you are an IC, your incentives are to get yourself appointed as TL of a major initiative, launch it, get promoted, and then repeat with an even harder project. If you are a manager, there are not enough genuinely hard & impactful projects to get all of your people promoted, so your incentives are to make your next project hard & impactful too by redoing a large chunk of what you just launched with subtly different requirements. If you are a director, you need to keep all your people happy by having equal chances at promotions, so you rotate around your top performers and give them successive hard & impactful launches that usually involve wild swings in direction. If you are a VP, you kill all competing projects so that projects under your current superstars are the ones that actually succeed.
Managers who do not do this find that they lose all their reports, because their reports do not get promoted, nobody wants to stick around on a team with no room for advancement, so they all leave and work on the projects which do get them promoted, which happen to be the ones that cannibalize Google's existing product portfolio. (You generally cannot cannibalize external competitors' product portfolios, because the timespan that promotions & re-orgs play out on is shorter than the time needed to win a billion-dollar market, and Google isn't interested in anything that can't become a billion-dollar business.) Thus, the entire company becomes dedicated to destroying the company's existing product portfolio and ruining its brand name. Anyone who isn't eventually gets sidelined and leaves for lack of advancement opportunities.