The creative performance loop
Set the learning agenda by funnel stage, tag every variation, trace results back to the creative elements behind them, and feed what was learned into the next brief. Pipeline and quality sit behind the loop.
Set the learning agenda
Each test names the question it answers and the KPI for its funnel stage.
14 tests across 7 platforms · 3 running
Build and tag variations
Testing dimensions are built into the concepts, and every asset is tagged to a shared ad taxonomy.
71% of assets tagged · 7 of 10 accounts tagged
Trace results to creative
Results are read against control, by element, so a win says which hook, talent or format did the work.
64% win rate · +19% average winning lift
Feed the next brief
AI synthesizes the read into learnings and brief inputs, so the next round starts smarter.
6 learnings · reused 11 times · AI brief refinement in pilot
Learning agenda by funnel stage
Each test sits in the stage its primary KPI measures. Creative and success measures differ by stage.
| Stage | KPIs in use | Read | Won | Win rate | Running |
|---|---|---|---|---|---|
| Awareness | Brand lift | 1 | 1 | 100% | 0 |
| Consideration | Cost per SQL | 2 | 1 | 50% | 0 |
| Conversion | CPA, IPM, CPI, CAC, ROAS | 8 | 5 | 63% | 3 |
Gap to close: the agenda leans toward conversion. Upper-funnel creative needs its own tests and measures, like brand lift and view-through, not conversion KPIs.
Hit rate by platform
Q3 · the mark is the 30% target
- Meta · 41 concepts34%
- Pinterest · 13 concepts31%
- TikTok · 34 concepts29%
- AppLovin · 22 concepts27%
- LinkedIn · 17 concepts24%
- YouTube · 18 concepts22%
- Snap · 11 concepts18%
Wearing out
Signal down 15%+ at frequency above 3
- Northpeak Outdoor · Trail-Tested Holidayrefresh by Nov 13Meta: CTR −34% in 10 days · frequency 4.8 · 41% of spend
- Lumora Beauty · Glow in 7 Daysrefresh by Oct 30TikTok: Hook rate −22% in 21 days · frequency 3.9 · 33% of spend
- Pulse Games · Fail Compilationrefresh by Nov 6TikTok: IPM −15% in 9 days · frequency 3.1 · 18% of spend
Feeds the next brief
Learnings synthesized from the read, to reuse across accounts and pitches
- UGCReused on 3 accounts
Creator-led room makeovers beat brand edits on CPA by 31%.
Source: Atlas Home - Mobile appReused on 2 accounts
Playable end cards lift IPM about 24% on AppLovin. Reuse the template across titles.
Source: Pulse Games - B2BReused on 1 account
A practitioner talking head beats polished brand video for LinkedIn cost per SQL (−22%).
Source: Vantix Cloud - TalentNot reused yet
Expert talent wins on Meta ROAS for beauty, but creator talent still wins on TikTok.
Source: Lumora Beauty - FormatReused on 1 account
Price-forward statics beat 6s video for subscription acquisition at Kinfolk.
Source: Kinfolk Pet - NativeReused on 4 accounts
Platform-native captions lift TikTok CPI about 18% over branded supers.
Source: Pulse Games
Test log
Running first, then newest. Sample results. Lift is the change in the test’s primary KPI against the control; a win or loss needs 90%+ confidence, and anything below that reads as inconclusive.
| Concept | Account | Platform | Variable | KPI | Result | Confidence |
|---|---|---|---|---|---|---|
| Money Mythsfrom Oct 5 | Helio Money | TikTok | Series vs. single | CPI | Running · day 12 | – |
| Fail Compilationfrom Sep 21 | Pulse Games | TikTok | Native caption style | CPI | Won +18% | 95% |
| Glow in 7 Daysfrom Sep 15 | Lumora Beauty | TikTok | Creator vs. studio talent | CPA | Lost −12% | 94% |