When isnt an event-driven setup the best choice, even for really complicated systems?

Everyone talks about how awesome event-driven architectures are for making systems bigger, less connected, and snappier, especially with microservices. But I've been wondering, when does all that extra work or complication from an EDA just not make sense? Like, what if your system needs super strict transaction consistency across a bunch of 'services' that would normally be separate event listeners? Or if the exact order of events is totally vital, and getting them out of order would be a disaster. That makes handling duplicates or complex event sorting a real headache. I'm really interested in what people have actually done. When have you skipped an EDA, even when it felt like the cool, up-to-date, or scalable thing to do, and why did you decide against it?