How to decide what to change in your Meta ads (the decision framework)

Every day a media buyer faces the same question: what should I actually change in this account? Scale something, kill something, refresh a creative, test a new idea, or leave it alone. The winning answer is almost never a gut call. It is a short, repeatable loop: trust the number, diagnose the cause, decide the action against the ad's own baseline, and change one thing at a time. This page is the map, and each step links to the deep dive that runs it.

The mistake that costs the most is jumping from a number straight to an action. ROAS dipped, so you kill the ad. Frequency looks high, so you cut the budget. Both skip the two steps in the middle, and both are usually wrong.

Step 1: Trust the number before you act on it

You cannot make a good decision on a number you do not trust. Three things distort the read most often.

First, which ROAS. Platform ROAS is a channel's self-report and is systematically inflated, so it is useful for comparing ad sets inside one account but dangerous as business truth. Read ROAS vs MER vs blended CAC to pick the right number for the question, and why nCAC beats blended CAC before any scaling call.

Second, the attribution window. The same ad shows a different ROAS under 7-day click versus 1-day, and mixing them silently double counts or hides sales. See 7-day vs 1-day attribution.

Third, is the drop even real. A one-day dip is noise and attribution fills in late. When results fall, run the why did my ROAS drop diagnostic before you touch anything.

Step 2: Diagnose the cause, not the symptom

A falling ROAS is an outcome, not a cause. Find the actual leak.

Walk the funnel as a chain of ratios and find the first step that falls off a cliff: reading your Meta funnel leak. If the whole ad is aging, check the leading fatigue signals of frequency, CTR, and CPM moving together: creative fatigue signals. High frequency on its own is not a verdict: when frequency is too high. And if you just changed something and delivery went noisy, you may have reset the learning phase.

Step 3: Decide the action, judged against the ad's own baseline

Now, and only now, choose the move. Universal thresholds lie, so compare each ad to what it normally does.

Use when to kill a Meta ad vs give it more time, a three-question test of whether the ad is judgeable, truly losing, and failing as an ad rather than an ad set. Use when to scale without breaking it, the roughly 20% step that does not reset learning. Set cadence with how often to refresh creative, by format, spend, and account age. And keep the pipeline full with a real creative testing framework, testing the hook first because it moves click-through rate more than any other single variable.

Step 4: Structure the budget and change one lever at a time

How your budget is structured decides how all these moves behave: CBO vs ABO budget strategy. Whatever you change, change one thing, then wait for delivery to settle before you read the result. The same discipline applies on Google, where the impression-share report tells you whether you are capped by budget or by rank, two problems that need opposite fixes.

The decision at a glance

| The signal you see | The step it belongs to | Where to go |

|---|---|---|

| ROAS looks off | Trust the number | ROAS vs MER, attribution, why ROAS dropped |

| Cost rising, results falling | Diagnose the cause | funnel leak, fatigue, frequency, learning phase |

| An ad clearly winning or losing | Decide the action | scale, kill, refresh, test, hooks |

| Deciding how to hold budget | Structure | CBO vs ABO |

| Ads not showing on Google | Diagnose and decide | impression share: budget vs rank |

The loop is boring on purpose. Trust, diagnose, decide, change one thing. Do it against each ad's own history, not a number you read in a blog, and the account gets clearer every week. This is exactly the read AdScale runs for you daily: it judges each ad on its own objective and history, then tells you what to scale, refresh, or kill, with the reason attached. It never changes your account, and every call is a draft you action yourself.