This is a production problem before it is a media problem
Two ads to the same audience with the same budget routinely differ in result by more than any targeting change would produce. That is the central fact about this channel and it decides how the engagement is shaped.
So the work is weighted toward making things: video shot inside your actual business, offers written to be understood in two seconds, and enough distinct concepts to test rather than three crops of one photo.
Testing one creative tells you whether that creative works. It does not tell you whether the channel works, which is the question you are actually paying to answer.
Fatigue is planned for rather than diagnosed. A local audience sees the same ad repeatedly, so rotation is scheduled from the start instead of being a reaction to a decline.
Reporting against a platform that flatters itself
Meta reports conversions using its own attribution model, and that model is inclined to take credit. The dashboard number is usually higher than the number in your bank account.
This is not a reason to avoid the channel, it is a reason to keep a second source of truth. We compare platform-reported conversions against customers recorded on your side and manage the gap explicitly rather than pretending it is not there.
For a local business the least sophisticated measurement is often the most accurate: ask new customers how they found you and write it down. We set that up as part of the engagement because it is what makes the rest of the reporting checkable.
When Meta is not the channel
When people already search for what you sell. If demand exists, capturing it on Google is a shorter path than creating it on Meta.
When you have nothing to show. This channel runs on images and video. A business that cannot produce or permit filming will underperform here regardless of budget.
When your category is restricted. Health, finance, housing and employment carry targeting and copy rules that remove most of what makes the channel effective, and ads that address the reader's personal attributes get rejected.
When you need attributable results this month for a board meeting. The attribution gap described above makes this the hardest channel to defend with a single clean number.