For example, I'm currently looking to buy a car - my recent search history includes "buy z4" and "merc c class". Yet YouTube is showing me generic ads for the local supermarket and off-track-betting (and I've never gambled in my life). So it seems Google has a lot of room to improve the algorithm that matches viewer to advertiser
Yeaahhhhh... people need to work on their algorithms for sure.
Do they?
Most adspace buyers are utterly clueless about how Google algorithms work.
The flaws the OP describe are real, but my bet is 80 to 90% of adspace buyers have no idea of the shortcomings.
Nor do they care BTW: the ROI on their ad dollar when spent on GOOG is so much higher than traditional channels ever were that they just set the drip in place and never look back.
Source: used to work on prediction modelling in ads.
What?
Also, I'm fine with ads not being targeted at me. It's really annoying to look up information on a toaster and then get bombarded with ads for toasters.
If you randomly show ads, what are the odds someone just happens to be in the market for a new one toaster? 1 in 1000?
On the other hand, what are the odds that someone who just bought a new toaster ends up returning it and being in the market for a new one? Probably much higher.
I don't think it's a mistake if after you bought product P, you get ads for P, it probably does work in practice.
Either you have this information, then Rebelgecko's comment explains why it would make sense to still have such ads.
Or you don't have this information, then other information (such as "you searched for P") probably leads to this result as well.
In both cases this strategy makes sense intuitively, and I'm pretty sure that ad-tech knows what it's doing to maximize returns, if only by looking at the metrics.