agentleFS
vidual-labsvidual-labs/digital-marketing-skills/skills/marketing/meta-ads-diagnostics/SKILL.md

meta-ads-diagnostics

Skill3 starsChanged 45 days agoMITView on GitHub

Use when analyzing Meta Ads (Facebook/Instagram) performance data and proposing concrete campaign changes — structure, budget allocation, Advantage+ vs manual setup, audiences, learning phase, creative fatigue, bid strategy, placements and attribution. Also use when the user pastes an Ads Manager export or screenshot and asks why CPA rose, ROAS fell, or how to scale without breaking delivery. Don't use for writing ad copy (meta-ads-creative), pixel/CAPI/consent bugs (gtm-debugging) or landing page fixes (landing-page-funnel).

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 3 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Don't use for: ad copy and creative concepts (use meta-ads-creative), pixel, Conversions API or consent problems (use gtm-debugging), …

    line 30
  2. Distinctive:

    Start with Advantage+ placements.

    line 141
  3. Distinctive:

    Editing daily: Every structural edit restarts learning.

    line 189
  4. Distinctive:

    Reading platform ROAS as truth: Check against actual orders.

    line 199

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Meta Ads Performance Diagnostics

Overview

Translate Meta Ads performance data into specific configuration changes with expected impact and a safe implementation order. Covers campaign structure, Advantage+ versus manual setups, budget allocation and scaling, audience strategy, learning phase, creative fatigue, bid strategies, placements and attribution settings. The output is a prioritized change list, not a description of the dashboard.

When to Use

  • Reviewing campaign, ad set or ad level performance
  • Diagnosing rising CPA/CPM, falling ROAS or stalled delivery
  • Deciding how to scale winners or when to kill losers
  • Fixing "Learning limited" ad sets
  • Choosing between Advantage+ sales/leads/app campaigns and manual campaigns
  • Choosing a bid strategy or attribution setting

Don't use for: ad copy and creative concepts (use meta-ads-creative), pixel, Conversions API or consent problems (use gtm-debugging), landing page conversion rate (use landing-page-funnel), or competitor analysis.

Inputs

Request at least 7 days of data (14–28 preferred) at these levels, with the attribution setting shown:

LevelMetricsWhy
AccountSpend, results, cost per result, ROAS, attribution setting (e.g. 7-day click / 1-day view), CAPI status, Event Match QualityBaseline and data trust
CampaignType (Advantage+ sales / leads / app, manual sales, traffic, engagement), budget type (campaign vs ad set), spend, results, CPA/ROAS, frequencyStructure and budget
Ad setAudience type (broad, Advantage+ audience, interests, lookalike, custom), audience size, placements, optimization event, bid strategy, learning status, spend, CPM, CTR (link), CPC, results, CPA, frequencyTargeting and learning health
AdCreative type, spend, impressions, CPM, link CTR, hook rate (3-second plays ÷ impressions), hold rate (ThruPlays ÷ 3-second plays), results, CPA, first run dateCreative fatigue
FunnelLink clicks → landing page views → add to cart / lead → purchase; rates between stepsWhere drop-off happens

Data rules: work only from the user's numbers and name the attribution window when quoting results. Benchmarks are starting points — the account's own history wins. Use the account's currency. Meta renames things often (Advantage+ shopping became Advantage+ sales; "lowest cost" is now "highest volume"); map labels to the nearest concept and say so.

Funnel benchmarks (starting points)

MetricGoodWarningCritical
Link CTR> 1.0%0.5–1.0%< 0.5%
Hook rate (video)> 30%20–30%< 20%
Hold rate (video)> 15%8–15%< 8%
Landing page views ÷ link clicks> 75%55–75%< 55% (slow page or bot clicks)
Frequency (7-day, prospecting)< 2.52.5–4> 4
Frequency (7-day, retargeting)< 66–10> 10
CPA vs target≤ 1.0×1.0–1.5×> 1.5×

Diagnostic Framework

Phase 0: Can the data be trusted?

  • Is the Conversions API sending the same events as the pixel with matching event_id for deduplication? Double counting inflates results; missing CAPI under-reports after consent denial.
  • Is Event Match Quality ≥ 6 on the optimization event? Low quality means weaker optimization and attribution.
  • Did the attribution setting change? 7-day click/1-day view vs 1-day click changes reported CPA by 20–50%.
  • Compare Ads Manager purchases against the shop's actual orders for the same window. A gap over ~20% needs gtm-debugging before optimization.

Phase 1: Campaign structure

IssueSymptomFix
Too many ad setsResults spread thin, several "Learning limited"Consolidate to 1–3 ad sets per campaign; let the algorithm allocate
Audience overlapRising frequency and CPM across ad setsUse the Audience Overlap tool; merge or exclude; prefer one broad ad set
Manual campaign where Advantage+ would winMature account, ≥ 50 purchases/week, stable creative pipelineTest an Advantage+ sales campaign with the same creative; cap the existing-customer budget share
Advantage+ where control is neededNew brand, strict geo/age rules, B2B lead quality problemsManual campaign with broad or Advantage+ audience and lead-quality filters
Prospecting and retargeting mixedRetargeting share drives CPA down while prospecting starvesSeparate by intent or use the existing-customer budget cap in Advantage+
Budget too low per ad setDaily budget < 5× target CPAFewer ad sets, higher budget each, or a cheaper optimization event
Wrong optimization eventOptimizing for add to cart because purchases are fewOnly step down the funnel while purchases < ~30/week; step back up as volume grows

Phase 2: Creative fatigue

SignalMeaningAction
Frequency > 3 (7 days) on prospectingSame people seeing the ad repeatedlyNew creative within 7 days
Link CTR down > 20% week over week with stable CPMAttention is fadingNew hook or angle
CPM up > 30% with stable CTRAuction pressure, seasonality or audience saturationBroaden, add creative diversity, check the calendar
Hook rate down, hold rate stableThumbnail/first second is tiredRe-cut the first 2 seconds only
First run date > 4–6 weeks and CPA creepingNatural decayRotate in 2–4 new concepts; keep the winner running until it is beaten

Plan creative refresh on a schedule (new concepts every 1–2 weeks for accounts spending meaningful budgets). Diversity across formats (static, video, UGC, carousel) extends the life of a campaign more than more variants of one format.

Phase 3: Learning phase

An ad set needs about 50 optimization events within 7 days to exit learning. Below that it shows "Learning limited" and delivery is less stable.

StatusReadAction
LearningNormal for new or recently edited ad setsDo not edit; wait for 50 events or 7 days
Learning limitedBudget, audience or event too smallRaise budget, broaden audience, consolidate ad sets, or optimize for a higher-volume event
ActiveStableEdit sparingly; scale ≤ 20% per day

Edits that reset learning: changing targeting, placements, optimization event, bid strategy or amount; budget changes above ~20%; pausing for 7+ days then resuming; adding a new ad to the ad set (resets that ad set). Edits that do not: pausing an individual ad, budget nudges under ~20%. Editing an existing ad's creative or copy effectively creates a new ad (and drops its accumulated social proof), so treat creative changes as new ads rather than edits.

Phase 4: Budget allocation and scaling

  • Campaign budget (Advantage+ campaign budget) lets Meta shift spend between ad sets; use ad set budgets only when you must guarantee spend per audience.
  • Tiering: top ad sets by CPA/ROAS get 50–60% of budget, middle 30–40%, test budget 10–15%.
  • Scale winners by ≤ 20% per day or duplicate at a higher budget and let the duplicate learn.
  • Kill rule: an ad set that has spent 2× target CPA with zero results, or 3× with results far above target, is paused.
  • Minimum viable daily budget per ad set ≈ 5× target CPA, otherwise it cannot exit learning.

Phase 5: Bid strategy (current names)

StrategyFormerlyUse when
Highest volumeLowest costDefault. Learning, prospecting, most accounts
Cost per result goalCost capKnown acceptable CPA; set at 1.1–1.3× target; expect lower volume
Highest valueValue optimizationPurchase values vary and ≥ 30 value events/week
ROAS goalMinimum ROASStable e-commerce with value data; set slightly below the true target
Bid capBid capRare: experienced buyers controlling auction bids directly

A goal set below what the account has achieved makes the ad set under-deliver. Start at reality and tighten by ≤ 10–20% per change.

Phase 6: Schedule and attribution

  • Ad scheduling (dayparting) is available only with a lifetime budget. Use it when the data shows CPA 2–3× worse in specific hours (e.g. a call-based business overnight) — not on a hunch.
  • Attribution settings can be set per ad set: 7-day click / 1-day view (default), 1-day click, or engaged-view for video. Report what the setting is; compare periods only with the same setting.
  • Incrementality: for large accounts, run a conversion lift or geo holdout before trusting platform-reported ROAS for budget decisions.

Phase 7: Placements

PlacementTypical behaviourKeep when
Facebook & Instagram feedsCore volume and conversionsAlways
Instagram & Facebook ReelsCheap reach, strong for vertical videoYou have 9:16 creative
StoriesMobile-native, time-limitedYou have 9:16 creative with safe zones
ThreadsNew inventory, limited dataTesting with Advantage+ placements
Audience NetworkLow CPM, weak quality, accidental clicksOnly if CPA on it is at or below campaign CPA
Marketplace, right column, searchSmall volumeLeave on under Advantage+ unless the breakdown shows waste
Messenger inbox / sponsored messagesMessaging objectivesConversation goals only

Start with Advantage+ placements; after ~30 results, break down by placement and remove any placement whose CPA is more than 50% above the campaign average.

Output Format

META ADS AUDIT: [Account / Campaign]
Date range: [Start] – [End] | Attribution: [setting] | Currency: [XXX]
Spend: [X] | Results: [Y] ([event]) | CPA: [Z] (target [T]) | ROAS: [R]
Data confidence: [High / Medium / Low — CAPI, EMQ, attribution, gaps]

--- HEALTH SCORE: X/10 ---
[One-line rationale]

0. TRACKING & ATTRIBUTION
   [CAPI / EMQ / attribution issues] → [Fix]

1. STRUCTURE
   [Issue] → [Change] → [Expected effect]

2. BUDGET & SCALING
   [Current allocation] → [Change, in ≤ 20%/day steps] → [Expected effect]

3. AUDIENCES
   [Overlap, size, broad vs targeted, exclusions] → [Change]

4. CREATIVE FATIGUE
   [Ad] — [signal: frequency / CTR / hook rate] → [Refresh plan and date]

5. LEARNING PHASE
   [Ad sets in learning / limited] → [Action or hold] → [Reasoning]

6. BID STRATEGY
   [Current] → [Recommended] → [Expected effect]

7. PLACEMENTS
   [Keep / remove, with CPA evidence]

PRIORITY CHANGES (implementation order, grouped so learning resets happen once)
1. [...]
2. [...]
3. [...]

ESTIMATED IMPACT: [Range with assumptions]
RE-CHECK DATE: [Usually 7 days]

Common Pitfalls

  1. Editing daily. Every structural edit restarts learning. Batch changes and let 7 days pass.
  2. Judging on 1–2 days. Delivery is noisy and attribution lags. Decide on 7 days minimum, 14 for small budgets.
  3. Scaling by 100% overnight. The ad set re-enters learning and CPA usually jumps. Scale ≤ 20% per day or duplicate.
  4. Treating "Learning limited" as a bug. It is a volume problem. Consolidate, raise budget or move up the funnel event.
  5. Comparing periods with different attribution settings. A switch from 7-day click to 1-day click "drops" results without anything changing.
  6. Reading platform ROAS as truth. Check against actual orders; run lift tests at scale.
  7. Ignoring creative as the cause. Most "targeting" problems in broad or Advantage+ campaigns are creative fatigue. Check Phase 2 before touching audiences.
  8. Over-segmenting audiences. Ten interest ad sets split budget and compete. One broad ad set with strong creative usually wins.

Verification Checklist

  • Tracking trust checked first (CAPI dedup, Event Match Quality, attribution window, order reconciliation)
  • Analysis uses ≥ 7 days of data or states the caveat
  • Attribution setting named whenever results are quoted
  • All seven phases evaluated; missing data called out
  • Funnel drop-off point identified (impression → click → landing page → event)
  • Creative fatigue assessed with frequency, CTR trend and hook/hold rates before structural changes
  • Learning-phase status of each ad set considered and resets batched
  • Budget changes stay within ~20%/day; kill rules applied with numbers
  • Bid strategy named with current (not legacy) terminology and a realistic goal
  • Placement recommendations backed by placement-level CPA
  • Priority changes in implementation order with expected impact range
  • Specific numbers cited from the user's data, not generic advice

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.