For the past several years our QA team has been very active in manual and automated testing of our products which led to a very small amount of real-user complaints about the functioning and usability of our major products.
One of the things that has been working for us, is "start testing early" - while a feature is either partially implemented, or just released in a very immature state (in a "dev" environment). The idea was to give developers feedback as quickly as possible. This is motivated by the Barry Boehm's law: the earlier a problem is found, cheaper it is to solve.
While it proved to work for us, there are related issues. Mostly, it is "Signal to noise" ratio.
On one hand, testing a feature that is not yet ready often helped to reveal critical problems, issues in designing/architecting a feature.
On the other hand, a lot of issues were not really bugs or problems, but rather things that were just "work-in-progress" or temporarily broken. This led to wasting time of both our developers and testers.
What would you recommend to improve on "signal to noise" ratio?