Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Lap plan ads ky sau TU DATA ky truoc: doc report/audit, giu va scale winner, thay loser, dat gia thuyet test moi, budget ky sau theo 3 kich ban. Luon quet data ky truoc truoc khi viet plan. Kich hoat khi user nhac 'plan ads thang toi', 'next ads plan', 'ke hoach ads ky sau', 'plan tu report ads'.
.claude/skills/minhnv0807-57-next-ads-plan/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 39% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 53% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 108% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 129% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 85% | 0% |
> Plan khong co data cu = doan mo = dot tien. Thu tu bat bien: Quet data → Nhan winner → Fix/dung loser → Test hypothesis moi. Day la skill cuoi chuoi performance va la skill dau cua chu ky moi.
Truoc khi viet bat ky dong plan nao, phai co:
07-bao-cao-marketing hoac dashboard tracking) — spend, CPL, ROAS, leads, don theo kenh/campaign/creative.21-audit-ads-performance / 03-danh-gia-hieu-suat).10-tinh-kpi-nguoc)Thieu report → dung lai, yeu cau data hoac chay 07-bao-cao-marketing truoc. Neu chi thieu 1 phan, ghi ro gia dinh trong plan.
| Da work (GIU) | Khong work (DUNG/FIX) | Chua test (THEM VAO) | |---------------|----------------------|---------------------| | winning creative/audience/kenh + so lieu] | losing ad set + so lieu + ly do] | hypothesis moi tu quan sat] |
Rut winning patterns:
| Ky plan | Budget tong | Objective | KPI target | Campaign/Offer moi | |---------|------------|-----------|-----------|--------------------| | thang/quy] | so] | lead/conv] | CPL so] / ROAS so]x / Leads so] | neu co] |
| Winner goc | CPL/ROAS ky truoc | Nhan ban thanh | Thay doi gi | Budget moi | Kenh | |-----------|-------------------|----------------|-------------|-----------|------| | Creative A] | so] | Giu nguyen + tang budget +20-30% | Khong doi | so] | x] | | Audience B] | so] | Duplicate + creative moi | Them 2 variant | so] | x] | | Hook C] | so] | Adapt sang kenh moi | Doi format | so] | x] |
Ap quy tac 55-scaling-ads: tang +20-30%/lan, khong x2; giu nguyen winning ad set.
| Campaign / Ad set | Van de (so lieu) | Quyet dinh | Neu fix: thay gi | |-------------------|------------------|-----------|------------------| | ten] | CPL > 2x target 3 ngay | DUNG | — | | ten] | CTR < 0.5% | FIX | Doi creative/hook, giu tep | | ten] | Frequency > 3.5, CTR giam 30% | FIX | Refresh creative, mo tep |
Moi loser dung phai ghi ly do vao log — de ky sau khong test lai y het.
| Campaign | Muc tieu | Budget/ngay | % Tong | |----------|---------|------------|--------| | Scale winner | Maintain + grow | so] | 50% | | Testing (hypothesis moi) | Tim winner moi | so] | 30% | | Retargeting | Convert warm | so] | 15% | | Lookalike | Scale tu seed moi | so] | 5% | | Tong | | | 100% |
| Kich ban | Gia dinh | Budget | Leads du kien | Hanh dong neu xay ra | |----------|---------|--------|---------------|---------------------| | Xau | CPL +30% vs ky truoc (mua cao diem, fatigue) | so] | so] | Cat testing con 20%, don ve winner + retarget | | Co so | CPL bang ky truoc | so] | so] | Chay dung plan nay | | Tot | CPL -20% (winner moi an) | so] | so] | Day them budget scale +20%/72h, mo kenh moi |
Cam ket voi stakeholder theo kich ban Co so; 2 kich ban con lai co san action de khong bi dong.
| # | Hypothesis ("Neu X thi Y vi Z") | Bien test | Platform | KPI do | Timeline | |---|--------------------------------|-----------|----------|--------|---------| | 1 | vd: Neu dung UGC testimonial thi CPL giam vi trust cao hon] | Creative format | x] | CPL | 7 ngay | | 2 | vd: Neu mo LAL 3% thi volume tang ma CPL giu] | Audience | x] | CPL | 7 ngay |
Setup va tieu chi ket luan theo 19-ab-test-setup (1 bien/lan, >=7 ngay, du conversions).
| Creative can lam | Type | Hook | Kenh | Deadline | Dung cho | |-----------------|------|------|------|---------|---------| | ten] | Video | Pain | Meta | ngay] | Testing | | ten] | Static | Social proof | Meta | ngay] | MOFU/Retarget | | ten] | UGC | Testimonial | TikTok | ngay] | Scale |
Luon co 2-3 creative du tru truoc khi ky moi bat dau — khong doi fatigue moi brief.
| Tuan | Focus | Action | % KPI thang can dat (luy ke) | |------|-------|--------|------------------------------| | 1 | Launch winner clones | Setup + launch, verify tracking | 20% | | 2 | Launch test moi | Chay hypothesis, doc so ngay 3 | 45% | | 3 | Optimize | Scale winner, pause loser | 70% | | 4 | Review + chuan bi ky sau | Pull data, viet report, draft plan moi | 100% |
Cuoi moi tuan doi chieu luy ke vs nguong: hut > 15% → kich hoat hanh dong kich ban Xau ngay, khong doi het thang.
Ten file: next-ads-plan-[ten-san-pham]-[YYYYMMDD].md — gom 9 phan theo dung 9 buoc tren (Review · Dinh huong · Nhan winner · Dung loser · Budget split · 3 kich ban · Hypothesis · Creative pipeline · Lich tuan).
07-bao-cao-marketing: nguon data chinh — chay truoc skill nay.21-audit-ads-performance / 03-danh-gia-hieu-suat: neu ky truoc co van de, audit truoc khi plan.10-tinh-kpi-nguoc: tinh lai KPI khi doi target/budget/offer.54-media-plan: khi ky moi thay doi lon (kenh moi, offer moi) → lap lai media plan day du.19-ab-test-setup: chi tiet setup hypothesis test.52-account-structure: setup campaign moi theo plan nay va launch.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 33,731 | 40,346 | +20% | 1 | 1 | 0% | 6,327 | 8,821 | +39% | 0 | 0 | — |
case-02 | fail→fail | 27,532 | 32,033 | +16% | 1 | 1 | 0% | 4,974 | 8,325 | +67% | 0 | 0 | — |
case-03 | fail→pass | 33,749 | 35,326 | +5% | 1 | 1 | 0% | 5,770 | 8,853 | +53% | 0 | 0 | — |
case-04 | fail→fail | 19,980 | 30,170 | +51% | 1 | 1 | 0% | 3,371 | 7,650 | +127% | 0 | 0 | — |
case-05 | fail→pass | 21,927 | 29,149 | +33% | 1 | 1 | 0% | 3,688 | 7,689 | +108% | 0 | 0 | — |
case-06 | pass→pass | 16,494 | 20,684 | +25% | 1 | 1 | 0% | 2,829 | 5,810 | +105% | 0 | 0 | — |
case-07 | fail→pass | 16,345 | 23,845 | +46% | 1 | 1 | 0% | 2,766 | 6,332 | +129% | 0 | 0 | — |
case-08 | fail→fail | 21,602 | 22,644 | +5% | 1 | 1 | 0% | 3,805 | 6,278 | +65% | 0 | 0 | — |
case-09 | fail→fail | 20,401 | 24,320 | +19% | 1 | 1 | 0% | 3,537 | 6,962 | +97% | 0 | 0 | — |
case-10 | fail→pass | 17,258 | 18,321 | +6% | 1 | 1 | 0% | 2,851 | 5,272 | +85% | 0 | 0 | — |
case-11 | pass→pass | 16,840 | 17,625 | +5% | 1 | 1 | 0% | 2,729 | 5,159 | +89% | 0 | 0 | — |
case-12 | fail→pass | 16,856 | 18,114 | +7% | 1 | 1 | 0% | 3,084 | 5,487 | +78% | 0 | 0 | — |
case-13 | fail→pass | 21,235 | 20,668 | -3% | 1 | 1 | 0% | 2,973 | 5,684 | +91% | 0 | 0 | — |
case-14 | fail→pass | 25,098 | 8,070 | -68% | 1 | 1 | 0% | 4,264 | 3,710 | -13% | 0 | 0 | — |
case-15 | pass→pass | 15,655 | 14,940 | -5% | 1 | 1 | 0% | 2,429 | 4,843 | +99% | 0 | 0 | — |
case-16 | fail→pass | 27,438 | 32,173 | +17% | 1 | 1 | 0% | 3,906 | 8,542 | +119% | 0 | 0 | — |
case-17 | fail→pass | 21,435 | 18,162 | -15% | 1 | 1 | 0% | 3,453 | 5,304 | +54% | 0 | 0 | — |
case-18 | fail→fail | 16,125 | 18,472 | +15% | 1 | 1 | 0% | 2,709 | 5,334 | +97% | 0 | 0 | — |
case-19 | fail→pass | 19,863 | 26,670 | +34% | 1 | 1 | 0% | 2,964 | 6,779 | +129% | 0 | 0 | — |
case-20 | fail→fail | 14,648 | 20,015 | +37% | 1 | 1 | 0% | 2,415 | 5,574 | +131% | 0 | 0 | — |
case-21 | fail→fail | 21,617 | 26,013 | +20% | 1 | 1 | 0% | 3,278 | 6,410 | +96% | 0 | 0 | — |
case-22 | fail→fail | 19,238 | 25,396 | +32% | 1 | 1 | 0% | 3,821 | 7,621 | +99% | 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 +50 percentage points is the difference between those two pass rates over the 22 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.