Observation mode is the biggest time-waster in paid search if you're not careful. Everyone parrots this "let the data speak" nonsense, but most people sit there for months collecting audience insights they never actually action. By the time you've got statistically meaningful conversion data on a segment, the market's already moved.
The only scenario where observation makes sense is if you're testing broad audiences with zero prior signal. Otherwise, you're better off running multiple ad groups with segmented targeting from day one, even if it means smaller volume initially. At least you'll get forced learning velocity rather than a vague dashboard of "interesting."
And that bit about keyword overlap-that's spot on but most people ignore it. I've seen campaigns where an audience looks like garbage because all the high-intent queries were already cannibalised by a separate exact match campaign. So you pause the audience, lose the incremental volume, and then wonder why CPA went up. The data is never clean, which is why I'd rather make aggressive assumptions and iterate than pretend observation gives you crystal clear truth.
Here's the contrarian take: observation mode is just analysis paralysis. Give me a PMAX audience signal and a strict CPC floor any day. At least that forces action.
