Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when the user needs a retargeting and lookalike plan: warm audience tiers, a different message per tier, frequency caps, attribution windows, lookalike seeds, exclusions, and a first-party data base that survives iOS ATT and cookie deprecation. Trigger on 'retargeting', 'remarketing', 'warm audience', 'custom audience', 'lookalike audience', 'retarget cart abandoners', or 'my retargeting pool is shrinking'.
.claude/skills/minhnv0807-56-retargeting-plan-global/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | -5% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 77% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 91% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 108% | 0% |
| case-20 | ✗→✓ | ▲ Improved | 103% | 0% |
Retargeting is a conversation with people who already know you. The message must differ completely from cold ads: handle objections, do not reintroduce the brand. Never serve the same ad to cold and warm audiences.
Ask up to 4 questions:
53-tracking-setup-global first and let data accumulate.53-tracking-setup-global.Before designing pixel tiers, confirm these exist:
| Asset | Why it matters | Minimum action | |-------|----------------|----------------| | Email list with consent | Survives cookie and ATT changes; usable on every platform | Capture email before or during checkout, and on the lead magnet | | SMS list with consent | High-intent channel in US markets | Explicit opt-in with a documented double confirmation where required | | CRM with source fields | Connects retargeting to actual revenue | Source, campaign, stage, revenue, opt-out flag | | Logged-in or account users | Stable identifier across devices | Encourage account creation post-purchase | | Server-side event stream | Recovers events browsers drop | Conversions API / Events API (53-tracking-setup-global) |
If pixel pools are too small to run, the first-party list is what keeps retargeting alive. Upload it as a Custom Audience and Customer Match list, and run email and SMS flows in parallel through 14-email-marketing-global.
| Tier | Audience | Signal | Window | Message direction | |------|----------|--------|--------|-------------------| | 1 (hottest) | Add to cart / checkout started | Pixel + server event | 7-14 days | Strong offer, urgency, guarantee | | 2 | Lead form opened but not submitted, open sales conversation | Pixel, CRM | 3-7 days | Reminder, remove the next-step friction | | 3 | Landing page visitors who did not convert | Pixel + server event | 14-30 days | Trust, social proof, objection handling | | 4 | Video viewers above 50% | Platform engagement audience | 14-30 days | Deepen interest, educate, prove | | 5 | Page and post engagers | Platform engagement audience | 30-60 days | Soft offer, value add | | 6 | Email / SMS / CRM list | First-party upload | Continuous | Nurture through owned channels, ads as reinforcement | | 7 | Past customers | Purchase event + CRM | 180 days | Upsell, cross-sell, replenishment, loyalty |
Note on durability: tiers 1-5 are pixel-dependent and will be smaller than the platform-reported estimate suggests. Tiers 6 and 7 are the ones you control. If tiers 1-3 are too small to spend the planned budget, merge them into a single hot tier rather than running several starved ad sets.
18-referral-program-global).Message-to-audience map to hand to 05-ad-copy-global:
| Audience | Core message | |----------|--------------| | Video viewers 75% | Education plus proof | | Site visitors, last 3 days | Strong offer plus urgency | | Page visitors who did not convert | Address the objection plus scarcity | | Open sales conversation | Personal, last chance | | Email subscribers, never purchased | First-purchase incentive | | Past customers | Upsell, cross-sell, replenishment |
| Platform | Recommended cap | |----------|-----------------| | Meta | 2-3 impressions per 7 days (hot tiers may run slightly higher) | | TikTok | 3-4 impressions per 7 days | | YouTube | 2 impressions per day | | Any platform, absolute ceiling | 2-3 impressions per person per day |
| Seed | Minimum seed size | Lookalike % | Platform | Purpose | |------|------------------:|-------------|----------|---------| | Purchasers | >= 100 | 1% | Meta | Highest-quality prospecting | | High-value purchasers (top revenue decile) | >= 100 | 1-2% | Meta | Value-based lookalike | | Qualified leads | >= 500 | 1-3% | Meta | Scale lead generation | | Video viewers 75% | >= 1000 | Broad | TikTok | Scale awareness | | Email / CRM list | >= 300 | 1-5% | Meta, Google Customer Match | Extend from owned data |
Lookalikes are cold campaigns and sit in the 5% lookalike slice of the 30/50/15/5 split, not in the retargeting budget — but their seeds are produced by this system. Build value-based lookalikes wherever revenue data is available; they consistently beat flat purchaser seeds.
File name: retargeting-plan-[product]-[YYYYMMDD].md
markdown# Retargeting Plan — [Product/Campaign] Platforms: [x] · Budget: [amount] ([%] of total) · Primary window: [days] ## 1. First-party base [lists owned, size, consent status, growth plan] ## 2. Audience tiers [tier, audience, signal, window, estimated size] ## 3. Message per tier [tier -> message -> CTA -> format] ## 4. Creative brief [what to produce; hand to 05-ad-copy-global] ## 5. Frequency caps [per platform] ## 6. Lookalike brief [seed, %, purpose, refresh cadence] ## 7. Exclusion rules [converters, opt-outs, prospecting overlap, cross-tier] ## 8. Expected KPIs [retargeting CPA and ROAS, typically 2-3x better than prospecting]
51-audience-research-global: defines the original cold/warm/hot tiering.53-tracking-setup-global: pixel, server-side events, and consent are the raw material for every custom audience.05-ad-copy-global: writes the copy for each tier's message.52-account-structure-global: retargeting sits in its own campaign with cross-exclusions.55-scaling-ads-global: scaling cold grows the warm pool — raise retargeting budget in step.14-email-marketing-global: the owned-channel half of tiers 6 and 7.16-marketing-psychology-global: objection handling, social proof, and honest scarcity.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | fail→pass | 52,131 | 35,550 | -32% | 1 | 1 | 0% | 8,330 | 7,943 | -5% | 0 | 0 | — |
case-01 | fail→fail | 43,475 | 40,552 | -7% | 1 | 1 | 0% | 6,863 | 8,922 | +30% | 0 | 0 | — |
case-03 | fail→fail | 45,149 | 42,490 | -6% | 1 | 1 | 0% | 8,321 | 10,727 | +29% | 0 | 0 | — |
case-04 | pass→pass | 13,670 | 11,684 | -15% | 1 | 1 | 0% | 2,328 | 4,514 | +94% | 0 | 0 | — |
case-05 | pass→pass | 21,823 | 20,286 | -7% | 1 | 1 | 0% | 3,764 | 6,361 | +69% | 0 | 0 | — |
case-06 | pass→pass | 17,027 | 13,917 | -18% | 1 | 1 | 0% | 2,742 | 4,849 | +77% | 0 | 0 | — |
case-07 | pass→pass | 13,240 | 12,721 | -4% | 1 | 1 | 0% | 2,473 | 4,981 | +101% | 0 | 0 | — |
case-08 | fail→pass | 17,172 | 13,228 | -23% | 1 | 1 | 0% | 2,737 | 4,853 | +77% | 0 | 0 | — |
case-09 | fail→pass | 14,386 | 13,685 | -5% | 1 | 1 | 0% | 2,475 | 4,719 | +91% | 0 | 0 | — |
case-10 | fail→pass | 14,065 | 14,225 | +1% | 1 | 1 | 0% | 2,386 | 4,967 | +108% | 0 | 0 | — |
case-11 | pass→pass | 11,263 | 8,645 | -23% | 1 | 1 | 0% | 1,881 | 3,979 | +112% | 0 | 0 | — |
case-12 | pass→pass | 11,383 | 11,140 | -2% | 1 | 1 | 0% | 2,002 | 4,325 | +116% | 0 | 0 | — |
case-13 | pass→pass | 15,584 | 15,451 | -1% | 1 | 1 | 0% | 2,513 | 5,085 | +102% | 0 | 0 | — |
case-14 | pass→pass | 18,525 | 19,243 | +4% | 1 | 1 | 0% | 2,939 | 5,841 | +99% | 0 | 0 | — |
case-15 | pass→pass | 15,212 | 11,963 | -21% | 1 | 1 | 0% | 2,484 | 4,531 | +82% | 0 | 0 | — |
case-16 | pass→pass | 14,662 | 10,777 | -26% | 1 | 1 | 0% | 2,301 | 4,186 | +82% | 0 | 0 | — |
case-17 | pass→pass | 15,004 | 13,066 | -13% | 1 | 1 | 0% | 2,515 | 4,743 | +89% | 0 | 0 | — |
case-18 | fail→fail | 9,110 | 3,365 | -63% | 1 | 1 | 0% | 1,802 | 3,189 | +77% | 0 | 0 | — |
case-19 | pass→pass | 10,507 | 11,601 | +10% | 1 | 1 | 0% | 1,854 | 4,418 | +138% | 0 | 0 | — |
case-20 | fail→pass | 12,084 | 9,055 | -25% | 1 | 1 | 0% | 2,006 | 4,068 | +103% | 0 | 0 | — |
case-21 | fail→pass | 15,550 | 13,620 | -12% | 1 | 1 | 0% | 2,437 | 4,763 | +95% | 0 | 0 | — |
case-22 | pass→pass | 15,428 | 12,596 | -18% | 1 | 1 | 0% | 2,472 | 4,593 | +86% | 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 +27 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.