A creative testing framework that finds winners faster

On modern ad platforms, targeting is increasingly automated. What you still control — and what decides most of your results — is the creative. The teams that win are not the ones with the best single idea, but the ones with the best system for testing ideas. Here is that system, step by step, with a planner for your own budget.

The system on one line

1Concepts

2Hooks

3Test & kill

4Scale winners

5Log & repeat

Most creatives fail — and that is fine. The system exists to find the few that work quickly and cheaply. A typical month for an account testing steadily might look like this:

20new concepts tested

7beat the account’s average CTR

2hit the target CPA — winners

10hook variations of the winners

4graduate to scaling campaigns

Illustrative numbers. A 10% hit rate on new concepts is normal; the multiplier comes from iterating on winners.

1. Test concepts, not colours

Changing a button colour or a headline word rarely moves the needle. Start one level higher: the concept — the angle that makes someone stop scrolling. Here are five concept families, shown for one example product, a language-learning app:

Concept	What it shows	Example opening line
Problem → solution	The pain, then the relief	“Froze when the waiter spoke Spanish? Here’s how I fixed it in 10 minutes a day.”
Social proof	Reviews, user-generated content, numbers	“Real users share their first conversation after 30 days.”
Demonstration	The product working: screen recording, before/after	“Watch a full lesson in 20 seconds.”
Offer-led	The deal itself: trial, discount, bonus	“7 days free. Cancel in two taps.”
Expert or founder voice	A person explaining why it matters	“I taught languages for 12 years. Here’s what actually works.”

Find the concept that works first. Iterate on hooks, formats and visuals second.

2. The first three seconds are the ad

For video, the hook decides whether anything else is seen. When a concept shows promise, produce three to five versions with different opening frames and first lines while keeping the rest identical. It is the cheapest way to multiply a winner.

0 s3 s30 s

“Froze when the waiter spoke Spanish?”Same body: product demo, review, offer

Close-up: phone alarm, “Lesson time”Same body

“10 minutes a day. 30 days. Watch.”Same body

Street interview: “Say hello in Spanish”Same body

Text on screen: “Stop scrolling, start talking”Same body

Only the first three seconds change, so any difference in results comes from the hook.

3. Give each test a fair budget and a clear metric

Decide the kill rule before you launch. A practical default: let each creative spend about one to two times your target CPA. Judge early on a leading metric (click-through rate, cost per landing-page view) and confirm winners on the metric you actually pay for.

Spend so far	What you see	Decision
0.5 × target CPA	CTR less than half the account average	Kill — the hook isn’t stopping anyone
1 × target CPA	Good CTR, no conversions yet	Keep running to 2 × CPA
2 × target CPA	Still no conversions	Kill and log why
2 × target CPA	CPA within 20% of target	Make hook variations, keep testing
3–5 × target CPA	CPA at or below target, stable for several days	Winner — graduate to scaling

Plan your testing budget

How many new creatives can you afford to test each month? Enter your numbers:

Creatives you can test per month 0

Testing budget: $0
Kill threshold per creative: $0
New creatives per week: 0
Expected winners per month: 0

Estimate only, based on the inputs above. Real results depend on your product, market and traffic.

4. Separate testing from scaling

Mixing new ideas into a campaign that already performs disturbs the delivery of your winners. Keep a dedicated testing campaign with a fixed share of budget — often 10–20% — and graduate proven creatives into your scaling campaigns.

15%85% · scaling campaigns

Testing campaign: new concepts and hooks, strict kill rulesScaling campaigns: proven winners only

Winners move right; nothing untested enters the scaling campaigns.

5. Keep a creative log

Every test should leave a trace (here with a $30 target CPA). After a couple of months this log becomes your most valuable asset — it tells you which angles resonate in which markets. A simple spreadsheet is enough:

Concept	Hook	Format	Spend	CPA	Result and why
Problem → solution	Waiter scene	UGC video	$72	$24	Winner. Relatable moment in the first second.
Offer-led	“7 days free”	Static	$60	—	Killed. Offer without context; low CTR.
Expert voice	Teacher to camera	Video	$82	$41	Iterate. Good CTR, weak landing match.

6. Refresh before fatigue, not after

Winning ads decay as the same people see them repeatedly. Watch frequency and click-through rate together: when CTR has dropped about 20% from its peak while frequency keeps climbing, the replacement should already be tested and ready.

Illustrative example. Start testing replacements while the winner is still at its peak — they need a week or two to prove themselves.

7. Localise, don’t just translate

A creative that wins in one country often needs more than subtitles elsewhere. Budget for local versions of your top concepts when you expand.

Element	Just translated	Localised
People	Same actors, dubbed voice	Local faces and native speakers
Setting	Original streets and interiors	Recognisable local places
Offer	Price converted at exchange rate	Local price points and popular payment methods
Humour and pace	Original jokes and editing speed	References and rhythm that feel native

The weekly rhythm

MonReview last week’s tests, update the creative log

TueProduce variations of winners

WedProduce 1–2 new concepts

ThuLaunch tests

Fri–SunLet tests collect data

Every day apply kill rules and move graduates to scaling

Boring? A little. But a steady rhythm beats occasional brilliance — and it is exactly how pay-for-performance teams keep CPA stable while volume grows. Not sure what your target CPA should be? Our max CPA guide shows how to calculate it.