rampstackco/claude-skills / skills/ads-creative-development
ads-creative-development
VerifiedHow to produce ad creative that converts at performance scale. Hook patterns, format selection, video pacing, variation systems, sequential testing methodology, fatigue detection, brand-voice alignment without conversion dilution, and platform-specific creative norms. Triggers on ad creative, ad design, hook patterns, ad video pacing, creative testing, ad variations, creative refresh, creative fatigue, refresh ad creative, video ads for Meta, TikTok creative, LinkedIn ad creative, ad asset library. Also triggers when a team is producing creative at scale, planning a creative test cycle, or auditing why creative is not converting.
| File | Kind | Size |
|---|---|---|
| README.md | markdown | 365 B |
| SKILL.md | skill md | 20.8 KB |
| references/brand-voice-performance-balance.md | markdown | 7.2 KB |
| references/creative-variation-templates.md | markdown | 5.5 KB |
| references/fatigue-detection-checklist.md | markdown | 6.3 KB |
| references/format-decision-matrix.md | markdown | 4.7 KB |
| references/hook-pattern-library.md | markdown | 6.9 KB |
| references/platform-creative-norms.md | markdown | 6.7 KB |
| references/testing-cadence-playbook.md | markdown | 5.9 KB |
references/testing-cadence-playbook.md
View on GitHub# Testing cadence playbook
The sequential testing waterfall. How to find the winning creative without testing 50 things at once.
The principle. Variance compounds. Testing all variables at once means you cannot tell whether the hook, the body, the CTA, or the format made the difference. Sequential testing isolates the variable that matters at each step.
---
## The waterfall
Five stages. Each stage isolates one variable and advances the winners.
### Stage 1: Hook test
**Goal.** Find which hooks stop the scroll.
**Setup.** 5 to 10 hook variants, same body, same CTA, same format. Each variant gets equal budget initially.
**Duration.** 7 to 14 days, or until each variant has 50+ conversions or 1M+ impressions, whichever comes first.
**Winner criteria.** Top 2 to 3 hooks by CTR (the metric that isolates hook performance) plus conversion rate above the floor. Eliminate hooks with high CTR but low conversion (the click is generated by the hook but the ad does not deliver).
**Output.** 2 to 3 winning hooks advance to Stage 2.
### Stage 2: Body test
**Goal.** Find which body copy or middle-section visuals convert best given the winning hooks.
**Setup.** Winning hooks times 5 to 10 body variants. Same CTA, same format. If two hooks advanced, that is 2 hooks times 5 bodies equals 10 ad sets.
**Duration.** 7 to 14 days, or until each combination has 50+ conversions.
**Winner criteria.** Top 2 to 3 hook plus body combinations by conversion rate. CTR is no longer the primary metric; the body is what holds attention after the hook.
**Output.** 2 to 3 hook plus body combinations advance to Stage 3.
### Stage 3: CTA test
**Goal.** Find which CTA produces the strongest action given the winning hook plus body.
**Setup.** Winning hook plus body combinations times 3 to 5 CTAs. Same format. Differences at this stage are usually marginal; do not over-engineer.
**Duration.** 5 to 10 days, or until each combination has 30+ conversions.
**Winner criteria.** Top hook plus body plus CTA combinations by conversion rate plus AOV (if value matters).
**Output.** 2 to 4 winning combinations advance to Stage 4.
### Stage 4: Format test
**Goal.** Find which format works best for the winning combinations.
**Setup.** Winning combinations rendered in 2 to 3 formats (vertical video, carousel, static).
**Duration.** 7 to 14 days.
**Winner criteria.** Top combinations by conversion rate within each format. Different formats may have different winners; that is fine. Run the winning format per surface.
**Output.** 2 to 4 production-ready creatives ready for evergreen rotation.
### Stage 5: Evergreen rotation
**Goal.** Run winners at scale; refresh on cadence.
**Setup.** 2 to 4 winners running in rotation. Frequency capping per audience.
**Duration.** Ongoing. Refresh creative on the cadence in `fatigue-detection-checklist.md`.
---
## Statistical thresholds
Two ways to call winners. Pick one and standardize across the team.
**Statistical significance.** Run until the winner has 95% confidence (or whatever threshold the team agrees on). Some platforms expose a "significance" indicator; others do not.
**Pre-commit duration plus minimum volume.** Decide before launching: "we will run for 14 days, each variant gets at least 50 conversions, top performer wins." This is the simpler, more honest approach for most performance work. The platform's algorithm will not run a clean test under your nose anyway, so trying for high-confidence statistical significance is often overkill.
The trap. Calling winners on small samples. A variant with 5 conversions and 10% lift over the other looks impressive but is within margin of platform noise. Wait for volume before declaring.
---
## Common mistakes
**Testing too many variables at once.** The most common mistake. Six hooks times five bodies times three CTAs times two formats equals 180 ad sets. The platform cannot optimize against that many variants; the team cannot interpret the results. Sequential is slower but produces real learnings.
**Killing variants too early.** A variant in the learning phase has not yet converged. The platform's first 3 days of delivery are not representative. Wait at least 7 days before killing.
**Calling winners on CTR alone.** A hook can drive high CTR with no conversion follow-through. The hook is doing its job (stopping the scroll) but the body is failing. Use conversion rate as the primary winner criterion; CTR is a hook-quality signal, not a creative-quality signal.
**No exclusion of recent converters from the test audience.** Recent converters click ads at higher rates because they are warm; mixing them into the test audience produces false-positive results. Exclude.
**Ignoring fatigue during the test.** A variant that wins the test in week 1 because it is new can underperform in week 4 once frequency builds. Test results are read against frequency; high-frequency winners are unreliable.
---
## Per-stage budget guidance
Rough percentages for the testing budget within a campaign's total spend.
| Stage | Share of testing budget |
|---|---|
| Hook test | 40% |
| Body test | 25% |
| CTA test | 10% |
| Format test | 15% |
| Buffer (re-tests, lift checks) | 10% |
The hook test gets the biggest share because it filters the most variants. CTA and format tests are smaller because the differences are usually marginal.
---
## When to break the waterfall
Two situations.
**Time pressure.** A peak moment (Black Friday, product launch) does not allow 6 weeks of sequential testing. Compress: run a hook test concurrently with format test using your best-guess body and CTA. Accept lower confidence in exchange for speed.
**Discovery of an outlier.** A hook variant that is winning by 5x at the end of week 1 should not wait for the full 14-day stage. Promote early, but track the long-tail performance to confirm.
The default is the waterfall. The exceptions are real but rare; do not let "we are on a tight timeline" become the default reason to skip the discipline.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |