Reviewed by Muze AI · 22 October 2026
Pause clear delivery faults first; investigate uncertain performance
Start with ads that lead to a broken page, unavailable offer or a market you cannot serve. Verify the issue and pause only the affected entities if needed while it is repaired. A recent low-ROAS row on its own is a weaker reason to stop delivery.
ROAS is attributed revenue divided by ad spend. It can fall because spending rose, conversion rate fell, order values changed, returns were recorded or attribution changed. Identify which part moved before choosing a response.
1. Make the comparison meaningful
Use equivalent time periods, currency, conversion actions and attribution settings. Separate prospecting from retargeting and brand from non-brand search. Inspect change history for budget, bidding, creative, tracking and website changes.
Then account for conversions that have not arrived yet. Google documents how recent spend can be fully visible while later conversions are still missing. Use your own account history and business sales cycle; a universal 72-hour hold or seven-day cutoff can be wrong.
2. Use a triage list instead of a blanket rule
| Evidence | Likely next step | Before applying it |
|---|---|---|
| Broken destination or unavailable product | Repair or pause affected ads | Reproduce the fault and confirm scope. |
| Clearly irrelevant Google query | Consider a narrowly scoped negative | Check actual intent and negative match type. |
| Recent spend, few purchases, normal delay not elapsed | Hold a scheduled review unless another fault exists | Compare mature cohorts and business cash constraints. |
| Persistent poor contribution after a mature window | Consider a limited pause or budget change | Verify tracking, margin and comparable evidence. |
| Store sales stable, platform attribution falls | Investigate measurement and allocation | Do not infer that all campaigns are healthy. |
3. Check profitability, not an arbitrary ROAS target
Use contribution margin before advertising to estimate a first-order break-even threshold. A 40% contribution margin gives an illustrative break-even ROAS of 2.5x. That margin must reflect the variable costs you actually pay.
For example, $400 attributed revenue on $200 spend is 2x ROAS. At a 40% contribution margin, that revenue contributes $160 before advertising, leaving a $40 shortfall after the $200 ad cost. Repeat purchases, fixed costs and attribution uncertainty require a broader model.
4. Do not confuse a small sample with a stable result
An ad with no purchases after one target CPA of spend can still be a useful ad. Purchase counts vary, and an average acquisition cost is not a promise that every spend window produces one customer. A fixed “pause at two times CPA” rule ignores uncertainty, reporting delay and your tolerance for loss.
Review the volume of mature data, prior results and the cost of being wrong. If evidence remains weak, define an affordable test limit and a review date. A cash constraint can require a business decision even while the performance estimate is uncertain; record that distinction.
5. Make one decision you can review
Write the evidence, affected ads, proposed action, approver and review date. Keep a record of changes so a later improvement or decline can be interpreted. Revisit the business outcome after enough data have matured.
Muze can help investigate connected Meta and Google data and assemble questions for that review. It does not guarantee that an ad is wasteful or validate your checkout. In Muze MCP, write tools ask for your confirmation first; Autopilot is a separate setting you turn on.
Investigate your account with Muze
See the connected Meta and Google analysis workflow, its limits and the plan before signing up.
Explore ad analysis