Creative testing
Test with guardrails. Scale only what clears them.
Every test carries a budget cap and an optional stop-loss and reports a winner with a stated confidence level. Nothing scales until you confirm it.
01 How a winner is called
A winner is a number that cleared its own bar, not just the higher one.
Every experiment runs a control against one or more variants, with guardrails checked against real delivery. Evaluating one computes a winner with a two tailed two-proportion z-test and a confidence level: never a bare score.
Guardrail band: p < 0.05 on either side, or medium confidence just past it
- Transformation openerMedium confidence
- Shop NowMedium confidence
High confidence. p < 0.05 and at least 30 conversions on both sides.
Medium confidence. At least 10 conversions on each side, where “Transformation opener” and “Shop Now” landed.
Low confidence. Otherwise, with a note stating exactly what count would be enough.
Fig. 2 · The two-proportion z-test’s guardrail bands · Northwind Sleep demo workspace, fictional data
02 Seven kinds of test, one result panel
In the demo workspace, “Learn More vs Shop Now” found Shop Now ahead at medium confidence, and “Lead-in hook test: cooling layer angle” found Transformation opener ahead at medium confidence.
2
Completed · demo workspace
2
Winners found · demo workspace
Fig. 3 · Experiments, the results table · Northwind Sleep demo workspace, fictional data
Fig. 3 · Experiments · Northwind Sleep demo workspace, fictional data
CTA test
One control ad against a CTA button variant. Everything else in the ad stays the same.
CTA test
One control ad against a CTA button variant. Everything else in the ad stays the same.
Lead-in test
A new opening line from Lead-In Lab against the original, grouped by hook type.
Lead-in test
A new opening line from Lead-In Lab against the original, grouped by hook type.
Identity test
The same creative run from a different Page or founder identity, compared on efficiency.
Identity test
The same creative run from a different Page or founder identity, compared on efficiency.
Custom
Any other controlled comparison a media buyer sets up by hand, with the same guardrails and result panel.
Custom
Any other controlled comparison a media buyer sets up by hand, with the same guardrails and result panel.
Creative matrix
A ranked hook, image, headline and CTA combination, built in Creative Lab and tracked here.
Creative matrix
A ranked hook, image, headline and CTA combination, built in Creative Lab and tracked here.
Second chance
A Rescue Lab candidate, launched with a capped budget and a stop-loss.
Second chance
A Rescue Lab candidate, launched with a capped budget and a stop-loss.
Winner revival
A staged relaunch of a fatiguing past winner, tracked stage by stage.
Winner revival
A staged relaunch of a fatiguing past winner, tracked stage by stage.
03 What a passing test makes
Scale Winners checks every campaign, ad set and ad with delivery in the current 7 day window against a run of rules: enough data to judge, beating its target or the account by 15% or more, not declining against the prior 7 days, no fatiguing ad dominating the spend, frequency stable. Every rule’s pass or fail is shown, never a single score.
A passing raise is capped at a conventional 20% step, 10% at medium confidence and none at low. Before offering it, SocialBuster checks Meta’s own edit-rate limit, a cooldown since the last edit, and whether the ad set or campaign is even active.
Nothing here changes on Meta until you confirm it.
Scale Winners
Fig. 4 · Scale Winners, the candidate table · Northwind Sleep demo workspace, fictional data
Fig. 4 · Scale Winners · Northwind Sleep demo workspace, fictional data
04 Where a test starts
A CTA test starts from a new experiment, set up directly above. A lead-in test starts a step earlier, from an opening line generated and classified by hook type.
Lead-In Lab →Prove a change worked before you scale it.
Connect a Meta ad account and SocialBuster syncs its campaigns, ads and daily results. Or load the Northwind Sleep demo workspace and look around first.