Is the creative tired, or did the audience change?

A banner’s click-through rate falls from 0.8% to 0.5%. The creative has been running for two weeks, so the team calls it fatigue and requests a replacement.

That may be the right action. It is not yet a diagnosis.

Performance can change because the message is less effective, the same audience has seen it repeatedly, or delivery has shifted to people and placements that respond differently. Those explanations call for different decisions.

Check whether the comparison changed

Use equivalent date ranges and confirm that both periods contain complete data. Compare device, placement, audience, geography, and campaign objective before evaluating the creative as a whole.

Here is a deliberately simplified example with two placement groups:

Period	Higher-response placement	Lower-response placement	Combined CTR
Week one	80,000 impressions at 1%	20,000 impressions at 0.2%	0.84%
Week two	20,000 impressions at 1%	80,000 impressions at 0.2%	0.36%

Neither placement’s click-through rate changed. The mix did. Replacing the banner might still be useful, but the combined decline does not show that the existing creative became less effective.

The CTR calculator can check the totals. Compare total clicks divided by total impressions rather than averaging the two placement percentages.

Look beyond clicks

For a sales campaign, a more curious audience can generate more clicks without generating better customers. For an awareness campaign, a click may not be the primary intended response.

Review the agreed outcome, landing-page behavior, and conversion delay. Check whether the offer, price, availability, or destination changed while the creative stayed the same.

Look at exposure concentration where suitable reporting exists. The frequency-cap guide explains why the average alone cannot show whether a small group is seeing most of the ads.

Give the replacement a specific hypothesis

“Try something fresh” is difficult to evaluate. A stronger proposal is: “The current image does not show the product in use; test an image that does, while keeping the offer and call to action consistent.”

Decide which outcome determines success, how much data is needed for a useful decision, and when the result will be reviewed. Avoid declaring a winner after the first favorable day.

Where the platform supports a controlled experiment, understand how it allocates traffic and budget. Google’s custom experiment documentation describes a control and trial setup; use the appropriate method for the platform and campaign type rather than assuming ordinary creative rotation is a randomized test.

Keep the execution interpretable

Give each variation a stable identifier. Record the actual start time, eligible placements, delivery settings, and any changes during the test.

If the new creative receives different inventory or a different audience, say so in the analysis. Automated delivery may favor one variation unevenly; a result from that setup answers a different question from a balanced controlled comparison.

Check the new asset and destination before release using the creative QA guide. A test with a broken click path says little about the message.

Make a decision you can explain

Replace a losing creative when the evidence and campaign constraints justify it. Keep a useful variation when it still serves a purpose. If the result is inconclusive, record the uncertainty and the next reasonable test.

The valuable record is not “new creative won.” It is what changed, who saw each version, what outcome improved, and which other explanations were checked.