I can't believe that even MOTD has fallen into the trap of giving the total xG of each side as part of their stats. They're lending legitmacy to the completely false idea that xG "decides" who "deserved" to win a match (it doesn't, it can't, and it's not presently intended to).
It's important to understand how limited current xG models actually are. Two shots from 12 yards central on a player's strong foot will have the same xG regardless of whether they had to curl it top bins around 3 defenders or just pass it past a poorly positioned goalkeeper one-on-one.
Over a sufficient number of shots (i.e. a long series of games), xG will eventually start to resemble the training datasets (law of large numbers, or as most people understand it in a sense: "luck evens out over a long period of time"). Only then does comparing a player or a team's actual performance to their expected performance make sense from a logically/mathematically/inferentially valid perspective.
Nothing I'm saying is controversial. The people who work on xG models would be the first to agree with me, along with any statistician you can find who understands xG conceptually. The problem is that the eggheads who develop the models and program the algorithms work for large football stats businesses whose sales departments will sell whatever stats a client wants. Salespeople don't care if they're selling data to be misused out of context: it's not their job to care, and that's for the few of them who even understand what it is they're really selling.
xG is a terrible blight on football -- at least as it's currently used since seeping into the mainstream. Don't get me wrong, it's a beautiful achievement from a mathematical/algorithmic perspective (and it would've been an intractable problem up until the last, maybe, 20 years), but the way it's being used is definitely not.
If you lack a strong grasp on university-level statistics (which few people can be reasonably expected to acquire), the only "safe" way to use xG without probably being wrong is to apply it to a long series of games (e.g. a player's performance at the end of a whole season). Otherwise, for an individual game, trust your own eyes! They're probably more accurate than xG!
I thought that from Norwich's shots, overall, they were unlucky not to score at least one from open play (how many times did they blaze it just past the post?). Spurs' xG, on the other hand, illustrates major problems with the metric: Son's goal alone will have had an xG of about 0.75 (I don't have access to my usual database to get the exact number right now but I've seen enough shots and their xGs). The important thing about Son's goal, though, is that the chance of a deflection like that occurring and falling to his feet the way it did was so low as to give the goal a somewhat farcical quality. As usual, the xG from this game can't be used to justify or defend its outcome -- and, for me at least, the game was much, much closer than its xG (just imagine Kane on the end of all of Norwich's chances..).
Sorry for the long post about a tangential matter but I see people becoming more and more reliant on these E(x) stats. Trust your eyes!