Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Thiet ke cau truc tai khoan ads: hierarchy campaign/ad set/ad theo objective, naming convention chuan de doc data nhanh, chon CBO/ABO theo giai doan, so luong ad set va creative toi uu, ngan sach toi thieu moi ad set. Kich hoat khi user nhac 'cau truc campaign', 'account structure', 'naming convention', 'setup campaign', 'len camp'.
.claude/skills/minhnv0807-52-account-structure/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 22% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 68% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 120% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 104% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 79% | 0% |
> Cau truc tot = doc data nhanh, toi uu dung cho, scale khong roi. Chay sau 53-tracking-setup (tracking xanh truoc) va sau khi co 51-audience-research + 54-media-plan.
Doc media plan va audience profile neu da co. Neu thieu, hoi toi da 4 cau:
Account
├── Campaign 1 — [Objective] — [San pham/Offer] (Cold Testing)
│ ├── Ad Set 1.1 — [Broad] ├── Ad A (hook 1) / Ad B (hook 2)
│ ├── Ad Set 1.2 — [Interest nhom A] ├── Ad A / Ad B
│ ├── Ad Set 1.3 — [Interest nhom B] ├── Ad A / Ad B
│ └── Ad Set 1.4 — [LAL 1-3%] └── Ad A / Ad B
├── Campaign 2 — [Objective] — Scale winner (tach rieng khi co winner)
└── Campaign 3 — Retargeting
└── Ad Set 3.1 — [Warm 7-30 ngay] — creative xu ly objection| Cap | Format | Vi du | |-----|--------|-------| | Campaign | [Platform]_[Objective]_[Product]_[AudienceType]_[MMYY] | META_LEAD_COURSE-A_COLD_0726 | | Ad Set | [Audience]_[Age]_[Gender]_[Placement]_[Budget] | INT-BUSINESS_25-45_ALL_FEED_200K | | Ad | [CreativeType]_[HookType]_[Version] | VID_PAIN_v1, IMG_SOCIAL-PROOF_v2 |
Quy tac: viet HOA, phan cach _, khong dau, khong khoang trang. Ten ad khop voi utm_content (xem 53-tracking-setup) de doi chieu data Ads Manager ↔ GA4.
| | ABO (budget o ad set) | CBO (budget o campaign) | |---|----------------------|-------------------------| | Dung khi | Testing — can kiem soat spend deu cho tung tep | Scaling — da co winner, de thuat toan tu phan bo | | Uu diem | Moi ad set chac chan co du data | Toi uu tu dong ve ad set hieu qua nhat | | Rui ro | Phai tu tay tat loser | CBO don tien vao 1 ad set, tep test khac khong duoc chi | | Quy tac | Testing phase mac dinh ABO | Chi chuyen CBO khi ad set da chung minh on dinh |
TESTING (tuan 1-2):
SCALING:
55-scaling-ads).MAINTAINING:
| Loai campaign | % Budget | Muc tieu | |---------------|---------|---------| | Cold — Testing | 30% | Tim winner moi | | Cold — Scaling | 50% | Scale winner | | Retargeting | 15% | Convert warm | | Lookalike | 5% | Mo tep moi tu seed |
| Platform | Toi thieu | Ly do | |----------|-----------|-------| | Meta Ads | 100-200K VND | Thoat learning phase (~50 conv/tuan) | | TikTok Ads | 150-300K VND | Minimum delivery | | Google Ads | 200-500K VND | Du clicks de toi uu | | Zalo Ads | 50-100K VND | CPM thap hon |
Quy tac kiem tra: budget/ngay cua ad set >= 5x CPL (hoac CPMess) muc tieu. Neu khong du → gop ad set lai, dung chia nho.
Ten file: account-structure-[ten-san-pham]-[YYYYMMDD].md
markdown# Account Structure — [San pham] — [Platform] Giai doan: [Testing/Scaling/Maintaining] · Budget thang: [so] ## 1. So do campaign [Cay hierarchy day du voi ten theo naming convention] ## 2. Bang ad set | Campaign | Ad Set (ten chuan) | Audience | Budget/ngay | ABO/CBO | Ads ben trong | ## 3. Naming convention ap dung | Cap | Format | Vi du thuc te cua project | ## 4. Budget allocation | Loai campaign | % | So tien/ngay | So tien/thang | ## 5. Checklist truoc launch - [ ] Pixel fire dung tren landing + thank you page (xem 53-tracking-setup) - [ ] Conversion event dung (Lead / Purchase), khong duplicate - [ ] UTM day du tren tat ca link, utm_content khop ten ad - [ ] Test Event Tool xanh het — CHUA XANH = CHUA CHAY - [ ] Naming dung format toan bo campaign/ad set/ad - [ ] Budget/ad set >= 5x CPL target/ngay - [ ] Exclusion: cold da loai khach mua + warm audience
| Loi | Hau qua | Fix | |-----|---------|-----| | Qua nhieu ad set (>5/campaign khi test) | Budget loang, khong du data | Gop ve 3-5 ad set | | Sua ad set dang learning | Reset learning, CPL nhay loan | Cho thoat learning roi moi sua | | Naming tuy hung | Khong doc duoc data, khong biet camp nao test gi | Ap format chuan, doi ten ngay tu dau | | Tron retarget vao cold | Message sai tep, CPL ao thap | Tach campaign, exclude cheo | | CBO khi dang test | Tien don vao 1 tep, tep khac khong co data | ABO khi test, CBO khi scale |
51-audience-research: nguon targeting cho tung ad set.54-media-plan: tong budget va phan bo kenh truoc khi chia campaign.53-tracking-setup: bat buoc xanh truoc khi launch.05-copy-quang-cao: viet copy/creative cho tung ad.19-ab-test-setup: quy trinh test 1 bien trong cau truc nay.55-scaling-ads: quy tac tach winner va tang budget.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 34,119 | 26,918 | -21% | 1 | 1 | 0% | 5,492 | 6,691 | +22% | 0 | 0 | — |
case-11 | pass→pass | 13,045 | 9,708 | -26% | 1 | 1 | 0% | 2,291 | 3,782 | +65% | 0 | 0 | — |
case-02 | fail→pass | 26,501 | 27,540 | +4% | 1 | 1 | 0% | 4,611 | 7,769 | +68% | 0 | 0 | — |
case-03 | fail→fail | 33,071 | 23,359 | -29% | 1 | 1 | 0% | 6,226 | 7,245 | +16% | 0 | 0 | — |
case-04 | pass→pass | 14,113 | 17,831 | +26% | 1 | 1 | 0% | 2,362 | 5,391 | +128% | 0 | 0 | — |
case-05 | fail→pass | 15,123 | 19,206 | +27% | 1 | 1 | 0% | 2,525 | 5,565 | +120% | 0 | 0 | — |
case-06 | pass→pass | 11,962 | 13,875 | +16% | 1 | 1 | 0% | 1,931 | 4,659 | +141% | 0 | 0 | — |
case-07 | pass→pass | 16,242 | 17,852 | +10% | 1 | 1 | 0% | 2,760 | 5,688 | +106% | 0 | 0 | — |
case-08 | fail→pass | 12,582 | 13,835 | +10% | 1 | 1 | 0% | 2,390 | 4,873 | +104% | 0 | 0 | — |
case-09 | pass→pass | 12,408 | 13,745 | +11% | 1 | 1 | 0% | 2,020 | 4,513 | +123% | 0 | 0 | — |
case-10 | pass→pass | 14,712 | 14,131 | -4% | 1 | 1 | 0% | 2,555 | 4,630 | +81% | 0 | 0 | — |
case-12 | pass→pass | 13,410 | 13,639 | +2% | 1 | 1 | 0% | 2,126 | 4,074 | +92% | 0 | 0 | — |
case-13 | fail→fail | 14,150 | 14,296 | +1% | 1 | 1 | 0% | 2,441 | 4,694 | +92% | 0 | 0 | — |
case-14 | fail→pass | 8,794 | 3,678 | -58% | 1 | 1 | 0% | 1,592 | 2,855 | +79% | 0 | 0 | — |
case-15 | fail→pass | 9,318 | 6,729 | -28% | 1 | 1 | 0% | 1,730 | 3,494 | +102% | 0 | 0 | — |
case-16 | fail→pass | 18,982 | 17,804 | -6% | 1 | 1 | 0% | 3,624 | 5,503 | +52% | 0 | 0 | — |
case-17 | fail→pass | 6,027 | 7,385 | +23% | 1 | 1 | 0% | 880 | 3,440 | +291% | 0 | 0 | — |
case-18 | fail→pass | 11,162 | 10,198 | -9% | 1 | 1 | 0% | 2,113 | 3,918 | +85% | 0 | 0 | — |
case-19 | fail→pass | 10,043 | 9,442 | -6% | 1 | 1 | 0% | 1,837 | 3,867 | +111% | 0 | 0 | — |
case-20 | pass→pass | 14,658 | 11,719 | -20% | 1 | 1 | 0% | 2,341 | 4,136 | +77% | 0 | 0 | — |
case-21 | fail→pass | 6,423 | 2,266 | -65% | 1 | 1 | 0% | 1,083 | 2,565 | +137% | 0 | 0 | — |
case-22 | fail→pass | 12,894 | 11,152 | -14% | 1 | 1 | 0% | 2,098 | 4,074 | +94% | 0 | 0 | — |
case-23 | fail→pass | 10,749 | 11,494 | +7% | 1 | 1 | 0% | 1,716 | 4,132 | +141% | 0 | 0 | — |
case-24 | fail→fail | 27,955 | 31,390 | +12% | 1 | 1 | 0% | 4,652 | 7,483 | +61% | 0 | 0 | — |
case-25 | fail→fail | 13,006 | 14,985 | +15% | 1 | 1 | 0% | 2,468 | 5,002 | +103% | 0 | 0 | — |
case-26 | fail→fail | 37,655 | 28,383 | -25% | 1 | 1 | 0% | 8,249 | 7,560 | -8% | 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. 26 cases were attempted. The headline lift of +50 percentage points is the difference between those two pass rates over the 26 comparable cases.
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.