Ad creative testing: change the version before you turn off the format
When a video doesn't convert, the next test is another video, not the usual static ad. See how to frame the hypothesis, what to change between versions and how to read results in the CRM.
When an ad format converts below expectations, the next step is to test another version of the same format instead of going back to what you already knew. The video that didn't convert may have lost on the hook, the first seconds or the offer, without the format being the problem. Ad creative testing with one hypothesis per version and one UTM per creative shows the difference, with the reading done in the CRM, where leads become opportunities.
Before turning off the format, change the version
Turning off a format because of one weak version throws away the learning and, sometimes, the account's best channel. In a campaign we run with Franq, part of the work described in the Franq case study, the video was converting below its potential. The recommendation was to test a new video version instead of turning the format off.
In the 20 days with the new video, compared with the previous 20 days:
| Metric | Previous 20 days | 20 days with the new video |
|---|---|---|
| Leads | 63 | 105 |
| Cost per lead | BRL 22.43 | BRL 18.25 |
| Spend | baseline | 35.6% higher |
The new video became the creative that brought in the most leads in the campaign. Note that spend also went up; even so, cost per lead fell.
Ad creative testing with one hypothesis per version
Each new version exists to test one idea, written down before the ad goes live. The format that works is a single sentence: “if we change this, we expect that, because”. For example:
- “if the video opens with the customer's pain point, more people watch the first three seconds, because today the opening only shows the brand”;
- “if the offer appears in the first third of the video, cost per lead goes down, because today it arrives after most people have left”;
- “if we swap the lead magnet offer for an assessment, leads arrive more qualified”.
The best hypotheses come from the numbers the account already has and from sales feedback on the leads from each ad; sales and marketing alignment is what makes that feedback arrive. Without a hypothesis, the test becomes a matter of taste. With one, the result teaches you something even when the version loses.
What changes from one version to the next
Change one thing at a time, starting with the one that has the biggest effect on conversion. The table helps you choose:
| What to change | When to test it | Sign it's the problem |
|---|---|---|
| Hook (first line or image) | before anything else | many impressions, low retention in the first seconds |
| First seconds of the video | once the hook holds attention | viewers leave before the offer |
| Offer (what the person gets) | when the video is watched but doesn't convert | good clicks, low conversion on the page |
| Call to action | fine-tuning | people reach the page without knowing the next step |
The offer matters more than it seems. The creative that won with one lead magnet can lose with another, as we saw when promoting a lead generation quiz.
A monthly testing calendar
One ad creative test at a time per campaign, planned at the start of the month, keeps everything from changing at once. For each test, define:
- the campaign and the hypothesis;
- who produces the new version, and by when;
- the date it goes live and the date it gets read;
- the reserved budget, which needs to be in the plan instead of coming from whatever is left over.
When test budget is already in the year's budget, as we suggest in marketing planning 2027, nobody needs to ask for approval for every new version.
One UTM per creative: reading it in the CRM
Each creative carries its own UTM, so the lead reaches the CRM saying which ad it came from. Google describes the utm_content parameter precisely as a way to differentiate creatives within the same campaign. In practice:
- agree on a naming pattern before launch (format, version and date, for example
video-v2-oct26); - use the same name in the ad, the UTM and the report;
- carry the UTM into the CRM with hidden form fields;
- compare versions by qualified lead and opportunity, not only by cost per lead.
That way the test stops depending only on the platform dashboard, which counts leads its own way. It's the same logic as reading cost per lead by campaign, now going down to the creative.
Reading window
Agree before launch on how long the test runs and how many conversions it needs before it can be read. Three rules prevent hasty conclusions:
- compare periods of the same length, like the 20 days against 20 days in the case above;
- wait for a minimum number of conversions per version before deciding, set by the account's volume;
- note any budget or audience changes during the period, because they move results as much as the creative does.
When to turn it off
Turn off the version when it loses its hypothesis within the agreed window; the format stays in testing. The format leaves the account when several versions, with different hypotheses, lose for the same reason, or when the leads it brings don't become opportunities in the CRM.
The log of each test, with hypothesis, period, result and decision, goes into the month's paid media report. In a few months, it shows which hooks and offers work for your audience. This is the work we do in paid media management, with tests planned at the start of each month.
Talk to usWant help planning this at your company?Tell us what you need. We'll reply by email or WhatsApp and, if it makes sense, put together a quote.Contact us
Frequently asked questions
How many creatives should you test at once?
A few, with different hypotheses. With a limited budget, two or three versions per ad set leave each one with enough volume to be read.
How long does ad creative testing need to run?
Long enough to gather the conversions needed to compare versions, over a period of the same length for all of them. Low-volume accounts need longer windows.
Video or static?
It depends on the offer and the audience, and the answer changes over time. Test both, but change the version before concluding that a format doesn't work.
