In the real world of cloud systems, you may have dozens of loosely coupled databases. The updates typically happen in a few seconds, but under certain conditions--such as weekend system refreshes or quarter-end reporting--transactions may be held up for an hour or even longer. That's the way it's supposed to work.

So in loosely coupled cloud systems, the first thing to do is make sure that any transactions have time stamps on all end-point systems, and make sure that your business logic understands what to do (or how to reconcile things) in case you have a "new" update that is actually several hours late.

OK, that's fine for the transactional tables (such as deals, payments, or transfers). But what about the core of the CRM system, which typically doesn't have much accounting content?

CRM systems are different from other enterprise software. The standards of data quality are different, both because of the data input sources (your customers or your sales reps) and the frequency of data update (some individual records may be hit several times in a day).

The highest-risk tables in a CRM are the Opportunity (particularly if its stage moves to "closed-won") and the Account (which is typically at the top of the information pyramid). So updates to these tables should be very carefully controlled, both with access control systems ("a sales rep may not modify these fields when this record is in a particular state") and with validation rules or coded mechanisms that limit the conditions under which data may be changed.