skills/ads/references/meta-decision-system.md
A quantified kill/keep/scale engine for Meta ads. Every threshold derives from one anchor number, so decisions become arithmetic instead of vibes. Pairs with the strategy-level Meta playbook in SKILL.md (creative-as-targeting, creative volume) — this file is the operating layer.
TCPL = Target Cost Per Qualified Lead (qualified = meets your ICP bar, not just a form-fill). Set it one of three ways:
Every rule below is expressed in multiples of TCPL. Review TCPL monthly.
More active ads than your budget can feed = every ad starves and nothing gets a fair read.
Ceiling = (daily budget × 14) / (2 × TCPL) — i.e., over a 14-day evaluation window, each ad needs at least 2× TCPL of spend to be judged.
$1,000/day at $500 TCPL → ceiling of 14 ads; run 6–10 (winners + 2–3 test slots). At the ceiling, launching a new test requires killing something first.
Run two CBO campaigns over the same audience:
Why: inside a single CBO, proven ads always starve new ads — tests never get enough spend to be judged. Why not ABO for testing: equal forced distribution keeps spending on ads Meta has already deprioritized. The separation is budget protection, not audience segmentation.
Image-first validation: launch new concepts as statics first; only produce the video/carousel/UGC version after the image passes the checks below. Exception: concepts that are inherently video (testimonial, demo, UGC).
A complementary structure for when the lander, not the creative, is the biggest unknown: one CBO per persona; inside it, one ad set per destination type — PDP, listicle/advertorial, quiz, demo page — with the same creatives in every ad set. Holding creative constant makes the read clean: any CPM or performance divergence between ad sets is the destination.
Why it works: the destination is a test axis of the same rank as creative — a losing funnel can hide winning creative, and different personas convert through different funnel shapes. CBO allocates budget across destinations the way it allocates across ads, and practitioners running this report wide CPM/performance spreads between destinations plus meaningful new-reach gains (~30%) from the added variety.
Fit with the two-campaign structure: treat a destination test like a concept test — run it in the Testing campaign with a protected budget, judge each ad set against TCPL at the usual spend gates, then graduate the winning creative × destination pair. Practitioner-reported pattern (Alexander Pauwelyn, 2026), not a platform-documented mechanic — validate against your own account data.
CBO's spend allocation is itself a signal — Meta pre-screens your ads. At day 7 for each test ad:
When iterating on a delivery-killed ad, change the hook/visual/format only — the audience never got far enough for copy or CTA to matter.
Run in order; stop at the first triggered action:
Graduate only when all are true: ≥5 qualified leads · qualified rate ≥60% · cost per QL ≤ TCPL · running ≥14 days · ≥1 QL in the last 7 days.
Frequency bands by campaign type (safe / warning / critical):
| Campaign type | Safe | Warning | Critical |
|---|---|---|---|
| Cold prospecting | 1.0–2.5 | 2.5–4.0 | >4.0 |
| Retargeting | 2.0–4.0 | 4.0–6.0 | >6.0 |
| ABM (small audiences) | 2.0–5.0 | 5.0–8.0 | >8.0 |
Other signals, in urgency order: CTR down 20%+ from baseline over 7 days; CPM up 30%+ over 2 weeks (leading indicator — moves before CTR); ad relevance rankings "below average"; CPA up with stable targeting.
For scaling-campaign ads, apply a deliberately stricter bar than the general bands — these ads carry ~80% of spend, so fatigue there costs the most: warning at frequency 3.0–3.5 or cost +20% → start 2 iterations now (they take ~14 days to be ready); swap at >3.5, cost +40%, or >1.5× TCPL for 2 weeks.
Lifespan expectations (B2B): statics 14–28 days; short video and carousels 21–35; UGC/testimonial 28–42. Small B2B audiences build frequency fast — plan refresh every 14–21 days.
Retire (don't iterate) when CTR drops 30%+ from peak or frequency crosses the campaign type's critical band above — the concept is exhausted, not the execution.
Rotation without resetting learning: never edit creative inside a performing ad — that resets the learning phase. Launch new ads alongside existing ones, or spin up a new ad set with the same targeting. Pausing doesn't reset; editing does.
Never pause without a replacement. Keep 2–3 iterations staged; replacement live within 7 days, immediately for critical fatigue. If the pipeline is empty, redirect the budget to proven ads rather than leaving a zombie running. What to change depends on why it died: delivery kill → hook/visual; quality kill → angle and ICP language; cost kill → offer and audience; fatigue → fresh execution of the same proven concept.
Scale only when all: proven-ad count meets the next budget level's minimum; account frequency <3.0; cost per QL ≤ TCPL for 2+ consecutive weeks; 3+ replacements staged.
The #1 B2B Meta lead-quality problem: frictionless auto-filled forms produce leads who don't remember converting ("social amnesia"). Intentional friction = awareness = quality:
Lead form vs. landing page: LP converting ≥5% → use the LP; LP under ~2% → lead form; demo/trial offers → LP; content/webinar → form.
Manual is where you learn; Advantage+ is where you earn. Transition a campaign to Advantage+ only after: a proven offer, a validated audience, and ~50 conversions/week on the optimization event (the learning-phase exit bar — budget needed ≈ target CPA × 50 ÷ 7 per day). If you can't hit 50/week on the target event, optimize a higher-volume event up-funnel and retarget converters. Advantage+ conflicts with strict ABM (you can't lock it to a list) — see the ABM playbook. Watch Campaign Score directionally (70+ healthy, <50 = fighting the algorithm) but never trade lead quality for score.
Everything above optimizes conversion inside an audience Meta already reaches you. Partnership ads are how you reach a net-new one. Andromeda targets by persona, not interest lists — and a creator's own following is a pre-assembled persona. Running an ad as a partnership (branded content from the creator's handle) inherits that seed audience, so the algorithm expands from people who already trust the fronting creator. This is the single highest-leverage lever on Meta right now; a serious account without partnership ads is bringing a butter knife to a gunfight.
Where it fits the decision system: partnership ads are a scaling move, not a testing gimmick. When the account hits the wall (frequency >3.5, rolling reach flattening — see below), the "add new seed audiences" step in the scaling protocol is largely this. Judge them against TCPL like any other ad, but expect a different failure mode: a weak partnership ad is usually the wrong creator, not the wrong hook.
Partnership-ads playbook:
Format-level guidance on which creator-fronted formats to run (founder content, yapper, authority, amateur-investigation, creator low-fi statics, etc.) lives in the ad-creative format taxonomy: meta-creative-formats.md (sibling addition — forward link).
Rolling month-over-month reach (unique people reached, MoM) is the account's net-new-audience gauge — the thing conversion metrics can't tell you. CPL and ROAS can look fine while you quietly recycle the same shrinking pool; the tell is reach going flat or declining month over month even as spend holds.
B2B SaaS Meta ranges (practitioner-reported; recalibrate on your own first 30 days): CTR 1.0–1.5% (red flag <0.8%); CPM $10–20 (red flag >$25); CPL (form) $20–50 (red flag >$75); landing page CVR 8–12%. Seasonality: Q1 CPMs are the year's lowest (scale aggressively); Q4 runs +60–80% (consider reducing B2B spend and banking budget for January).
Framework lineage: this decision system is adapted (re-expressed, reconciled, and restructured) from practitioner operating systems, notably Ivan Falco's ads-skills. All thresholds are starting points — recalibrate against your own account.