This is just as valid for mobile app and website development.
When all you use for testing is Browserstack, local emulators and whatnot and only the latest iPhone and Samsung S-series flagship, your Thing will be unusable for wide parts of the population.
Always, always use at the very least the oldest iPhone Apple still supports, the cheapest and oldest (!) Samsung A-series models still being sold in retail stores as "new", and at least one Huawei and Xiaomi device. And then, don't test your Thing only on wifi backed by your Gbit Wifi 7 router and uplink. Disable wifi and limit mobile data to 2G or whatever is the lowest your phone provider supports.
And then, have someone from QA visit the countryside with long stretches of no service at all or serious degradation (think packet loss rates of 60% or more, latencies of 2 seconds+). If your app survives this with minimal loss of functionality, you did good.
A bunch of issues will only crop up in real world testing. Stuff like instead of keeping a single socket to the mothership open, using fresh from scratch SSL connections for each interactions is the main bummer... latency really really eats such bottlenecks alive. Forgotten async handling leading to non-responsiveness of the main application. You won't catch that, not even with Chrome's network inspector - you won't feel the sheer rage of the end user having a pressing need and be let down by your Thing - even if you're not responsible for their shitty phone service, they will associate the bad service with your app.
Oh, and also test out getting interrupted while using your Thing on the cheap-ass phones. Whatsapp and FB Messenger calls, for example - these gobble so much RAM that your app or browser will get killed by OOM or battery saver, and when the user has their interruption finished, if you didn't do it right your Thing's local state will have gotten corrupted or removed, leading the user having to start from scratch!