Ironically the smaller banks like Monzo are the only ones that actually do a decent job.
By serious stuff I meant stuff that requires perfect accuracy and integrity - credit checking wouldn't care if you're missing one transaction here or there, or if the timestamps are a little off.
But for accountancy, perfect accuracy is required and OB with legacy banks is far from perfect in that regard - the data quality is bad, you get missing/duplicate transactions, timestamps are off (due to TZ issues). Modern banks are the only ones where the data is any good. For accountancy services, this ends up causing significant support overhead to companies where customers complain their numbers are off and the company can't do anything because the bank sends incorrect data, not to mention engineering overhead where you have to try and clean up the data in-house (the problems are different for each bank, so you have to essentially reverse-engineer how each bank mangles the data and implement bank-specific workarounds).
> Uptime and performance seems to be monitored pretty closely since 2020
Uptime, sure. I too can make a 100% uptime API that returns random data.
When it comes to data quality, there didn't seem to be any authority (in terms of real, practical outcomes - not theoretical powers that never end up being used like the GDPR for example) to complain to - which I find to be a fatal mistake in a situation where banks otherwise have zero incentive to provide a functional & usable OB API (in fact, OB is detrimental to their business as it would allow customers to use a third-party's - often better - service over the bank's own one). Even the OB "gateways" like Plaid, TrueLayer, etc have their hands tied when it comes to this - once the issue is confirmed to be with the bank, the lead time on getting any kind of resolution is often months, during which you have a disgruntled customer breathing down your neck and blaming you for the problem.