Creative testing that doesn't burn the budget
Three questions, three days, one decision.
2 min read
Key takeaways
- Test one variable at a time, and name the variable before you spend.
- A test needs a stopping rule written down in advance, or it never ends.
- Most 'winning' creatives at 20 conversions are noise; at 100 they are signal.
- Decide on cost per result, not on CTR. CTR misleads you about the wrong audience.
On this page5 sections
Creative testing goes wrong in a predictable way. Six new assets go live on Monday. By Wednesday one of them has a CPA thirty percent below the others on eighteen conversions, so it gets the budget. By the following Monday it has regressed to the mean and the cycle restarts. Nothing was learned, and a week of spend produced one anecdote.
Name the variable first
Write down, before anything goes live, which single thing changes between variants: hook, format, offer framing, proof or call to action. If two things change at once, the result tells you nothing about either.
Variable
Hook (first 3 seconds)
- What you learn
- Whether the audience stops
- Typical spend to a decision
- 1–2× target CPA per variant × 50
Variable
Angle / motivation
- What you learn
- Which problem people buy on
- Typical spend to a decision
- 2–3× target CPA per variant × 50
Variable
Format (UGC, static, motion)
- What you learn
- Which container fits the placement
- Typical spend to a decision
- 1× target CPA per variant × 50
Variable
Offer framing
- What you learn
- Price sensitivity and objections
- Typical spend to a decision
- 3× target CPA per variant × 50
| Variable | What you learn | Typical spend to a decision |
|---|---|---|
| Hook (first 3 seconds) | Whether the audience stops | 1–2× target CPA per variant × 50 |
| Angle / motivation | Which problem people buy on | 2–3× target CPA per variant × 50 |
| Format (UGC, static, motion) | Which container fits the placement | 1× target CPA per variant × 50 |
| Offer framing | Price sensitivity and objections | 3× target CPA per variant × 50 |
The arithmetic nobody wants to do
A difference in conversion rate needs enough events to be distinguishable from chance. As a working rule for media buying: below thirty conversions per variant, treat differences under fifty percent as noise. At a hundred conversions, a twenty percent difference starts to mean something. Below ten, you are reading tea leaves.
This is why testing six variants at once is usually a mistake for accounts below a certain budget. Three variants that each reach a hundred conversions beat six that each reach thirty, every time.
Write the stopping rule down
A test without a stopping rule runs until somebody gets bored. Before launch, write one sentence: 'We stop when each variant has 100 conversions or when we have spent $3,000, whichever comes first, and we pick on cost per purchase, not CTR.' Put it in the brief. It ends the argument before it starts.
A three-by-three that works
- 1Pick three angles that address genuinely different motivations.
- 2Produce three executions per angle: one static, one UGC-style, one motion.
- 3Launch as one ad set per angle, with three ads inside.
- 4Let it run to your stopping rule without edits.
- 5Take the winning angle forward and test executions inside it next round.
The point of round one is the angle, not the asset. Angles are strategy; assets are production. Most teams invert this and spend their budget discovering which shade of blue converts.
What to do with losers
Do not delete them. Archive each one with the note that matters: what the angle claimed, which audience saw it, what the cost per result was, and your one-line reading of why. Six months of those notes is the most valuable asset an advertiser accumulates, and it belongs to the business, not to any one campaign.
The output of a creative test is a sentence you can act on, not a file you can upload.
Related guides and tools
Written by
Krög editorial team
Account managers, compliance and billing
Written by the people at Krög who open, fund and support advertising accounts for verified businesses every day. Platform rules change often. If something here is out of date, tell us on Telegram.
Meet the team