Wouldn't use them for something mission critical but for statistics where you don't mind missing a few minutes like in this case they're ok.
Also I'm using a Pi to do some tests for a project that will run on custom hardware (not even Broadcom) eventually. Saves time when you don't have to deal with Yocto and editing dts files just to talk to something on spi and see if it can do your job.
(edit for clarity)
There are cheaper alternatives, but there are not better supported alternatives. The most expensive rPi is a rounding error in terms of engineering time. Why bother saving ~$30 if you have to spend $1k worth of engineering time to diagnose a compatibility issue on some less well documented board.
You can also reduce your write wear with putting stuff like temporal logs in ram etc.
* depending on volume.
For a couple hundred/thousand devices doing industrial automation yes. For a million and up, it makes sense to pay an engineer 3 months to reduce the BOM by 10 cents.
Our volume would be well below 1 million at market saturation. More important for us to be able to iterate and customize.