Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Framework scale ads khi da co winner: kiem tra dieu kien du scale (CPL dat, frequency thap, volume tep con), vertical scaling +20-30%/lan, horizontal scaling (dup ad set, tep moi, kenh moi), quan ly rui ro reset learning phase va lich scale. Quy tac cung: khong tang budget khi CPL dang xau. Kich hoat khi user nhac 'scale ads', 'tang budget', 'nhan winner', 'scale campaign'.
.claude/skills/minhnv0807-55-scaling-ads/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-08 | ✗→✓ | ▲ Improved | 153% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 129% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 216% | 0% |
| case-05 | ✓→✗ | ▼ Worse | 64% | 0% |
| case-13 | ✓→✗ | ▼ Worse | 137% | 0% |
> Scale = tang spend ma giu nguyen (hoac cai thien) efficiency. Scale sai = CPL/CPA tang vot, dot budget. QUY TAC CUNG: khong tang budget khi CPL dang xau — fix van de truoc (goi 21-audit-ads-performance), scale sau.
Hoi toi da 4 cau:
Thieu bat ky muc nao → chua scale; quay ve toi uu hoac chuan bi them.
| Muc do | Muc tang | Tan suat | Dieu kien | |--------|---------|---------|-----------| | Conservative | +20% | Moi 72h | CPL on dinh o target | | Moderate | +30% | Moi 2-3 ngay | CPL tot hon baseline ro rang | | Aggressive | +50% | 1 lan duy nhat | Campaign rat stable, chap nhan rui ro | | KHONG LAM | x2 ngay | — | CPL se tang vot, reset learning |
Sau moi lan tang, monitor 24-48h. Day la nguong DELTA — do muc tang so voi chinh no truoc khi scale, khac voi thang CPA tuyet doi trong Gate 1:
> Hai nguong chay song song, khong thay the nhau. Neu CPL sau khi scale vuot 3x CPA muc tieu thi Gate 1 thang: tat, du muc tang chi 15%.
| Hanh dong | Co reset learning? | Cach lam an toan | |-----------|--------------------|------------------| | Tang budget <= 20-30% | Thuong khong | Tang tung buoc, cho 72h | | Tang budget > 30-50% | Co the | Chia nho thanh nhieu lan | | Doi audience / placement / bid | Co | Duplicate ad set moi thay vi sua truc tiep | | Them/xoa creative trong ad set | Co the | Them o ad set moi hoac campaign testing | | Pause roi bat lai lau | Co the | Han che pause winner; dieu tiet bang budget |
| Tuan | Budget/ngay (vi du) | Action | KPI checkpoint | |------|--------------------|--------|----------------| | 1 (Testing) | 300-500K | Test 3-5 ad set | CPL < target | | 2 | 500-800K | +20% winner, dung loser | CPL on dinh 5/7 ngay | | 3 | 800K-1.5M | +20-30% tiep, duplicate winner sang tep moi | Volume tang, CPL giu | | 4 | 1.5M-3M | Horizontal: LAL/broad/kenh moi | CPL giu trong nguong | | Thang 2+ | 3M+ | Full scale + creative moi lien tuc | ROAS >= target |
→ Giam ve budget gan nhat con hieu qua, cho on dinh 48-72h, chay 21-audit-ads-performance neu khong hoi phuc.
| Tin hieu | Muc do | Action | |---------|--------|--------| | Frequency > 2.5 | Canh bao | Chuan bi creative moi (brief ngay) | | Frequency > 3.5 | Khan cap | Launch creative moi ngay | | CTR giam > 30% | Tuan 2-3 | Refresh hook, giu body | | Dinh ky | Moi 2-3 tuan | Luon co 2-3 creative moi de swap |
Ten file: scaling-plan-[ten-san-pham]-[YYYYMMDD].md
markdown# Scaling Plan — [Campaign/San pham] Baseline: CPL [so] · Budget hien tai [so]/ngay · Target [so]/ngay ## 1. Checklist dieu kien scale [7 muc — trang thai tung muc] ## 2. Vertical plan [lich tang %, moc kiem tra 24-48h] ## 3. Horizontal plan [tep moi / creative moi / kenh moi + exclusions] ## 4. Roadmap theo tuan [budget, action, checkpoint] ## 5. Nguong dung scale [cac dau hieu + hanh dong] ## 6. Creative pipeline [creative du tru + deadline]
19-ab-test-setup: winner phai qua test dung chuan truoc khi scale.21-audit-ads-performance: audit truoc khi scale neu CPL chua on; audit khi scale drop.03-danh-gia-hieu-suat: daily/weekly check va decision rules.56-retargeting-plan: scale cold luon di kem tang retarget (them volume warm).52-account-structure: tach winner ra campaign scale (CBO) dung cau truc.57-next-ads-plan: ket qua scale ky nay la input plan ky sau.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 29,787 | 32,607 | +9% | 1 | 1 | 0% | 5,415 | 7,436 | +37% | 0 | 0 | — |
case-02 | fail→fail | 8,353 | 5,711 | -32% | 1 | 1 | 0% | 1,469 | 3,220 | +119% | 0 | 0 | — |
case-03 | fail→fail | 30,589 | 21,268 | -30% | 1 | 1 | 0% | 5,097 | 6,085 | +19% | 0 | 0 | — |
case-04 | pass→pass | 13,131 | 9,259 | -29% | 1 | 1 | 0% | 2,053 | 3,743 | +82% | 0 | 0 | — |
case-05 | pass→fail | 20,816 | 21,262 | +2% | 1 | 1 | 0% | 3,637 | 5,956 | +64% | 0 | 0 | — |
case-06 | pass→pass | 13,610 | 13,823 | +2% | 1 | 1 | 0% | 2,516 | 4,756 | +89% | 0 | 0 | — |
case-07 | fail→fail | 16,752 | 17,757 | +6% | 1 | 1 | 0% | 2,830 | 5,290 | +87% | 0 | 0 | — |
case-08 | fail→pass | 9,055 | 8,998 | -1% | 1 | 1 | 0% | 1,466 | 3,705 | +153% | 0 | 0 | — |
case-09 | pass→pass | 11,367 | 8,286 | -27% | 1 | 1 | 0% | 1,852 | 3,607 | +95% | 0 | 0 | — |
case-10 | fail→fail | 11,410 | 12,323 | +8% | 1 | 1 | 0% | 1,907 | 4,336 | +127% | 0 | 0 | — |
case-11 | pass→pass | 12,368 | 13,377 | +8% | 1 | 1 | 0% | 2,038 | 4,657 | +129% | 0 | 0 | — |
case-12 | fail→pass | 10,811 | 11,187 | +3% | 1 | 1 | 0% | 1,797 | 4,112 | +129% | 0 | 0 | — |
case-13 | pass→fail | 11,933 | 16,026 | +34% | 1 | 1 | 0% | 2,119 | 5,022 | +137% | 0 | 0 | — |
case-14 | pass→pass | 12,738 | 10,359 | -19% | 1 | 1 | 0% | 2,096 | 3,780 | +80% | 0 | 0 | — |
case-15 | fail→pass | 7,763 | 11,450 | +47% | 1 | 1 | 0% | 1,355 | 4,278 | +216% | 0 | 0 | — |
case-16 | pass→pass | 17,047 | 23,362 | +37% | 1 | 1 | 0% | 2,686 | 5,795 | +116% | 0 | 0 | — |
case-17 | pass→pass | 11,298 | 10,119 | -10% | 1 | 1 | 0% | 1,707 | 3,836 | +125% | 0 | 0 | — |
case-18 | pass→pass | 17,060 | 16,617 | -3% | 1 | 1 | 0% | 2,883 | 5,152 | +79% | 0 | 0 | — |
case-19 | fail→fail | 16,518 | 20,073 | +22% | 1 | 1 | 0% | 3,185 | 6,100 | +92% | 0 | 0 | — |
case-20 | pass→pass | 11,510 | 12,700 | +10% | 1 | 1 | 0% | 1,939 | 4,403 | +127% | 0 | 0 | — |
case-21 | pass→pass | 12,494 | 12,296 | -2% | 1 | 1 | 0% | 2,073 | 4,398 | +112% | 0 | 0 | — |
case-22 | pass→pass | 14,559 | 16,181 | +11% | 1 | 1 | 0% | 2,159 | 4,598 | +113% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +5 percentage points is the difference between those two pass rates over the 22 comparable cases. 3 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.