You can't predict which message will book jobs in your area. Testing is how you find out — but only if the tests are designed to teach you something.
Test angles, not tweaks
Changing a button from blue to green won't tell you much. Changing the angle will: a problem-led hook versus an offer-led one; a checklist versus a project showcase; an emergency message versus a maintenance message.
Run enough, but not too many
For a small local budget, three to five active creatives per campaign is plenty. More than that and none of them gets enough impressions to judge.
Set minimum data before deciding
Decide in advance how much evidence you need: at least a few days of delivery and a minimum number of inquiries. Turning ads off after one quiet afternoon just resets the learning.
Judge on outcomes, not clicks
Click-through rate tells you whether the ad is interesting. Booked estimates tell you whether it's useful. Use the CRM outcome when you have it; fall back to network-reported conversions when you don't, and label which is which.
Refresh before fatigue
When click-through falls steadily or the same people see the ad too often, swap in a fresh creative in the winning angle. Keep a few approved variations ready so refreshing is a five-minute job.
Keep a log
Write down what you tested and what you learned. Next season, you'll start from the winning angle instead of from scratch.