agentleFS
minhnv0807minhnv0807/ai-business-skills/skills/en/03-performance-eval-global/SKILL.md

03-performance-eval-global

Skill644 starsChanged 6 months agoMITView on GitHub

Use when marketing numbers look bad and the user needs to know WHY — layered root-cause diagnosis across tracking, delivery, creative, landing page, offer, and audience quality, 5-Whys, regional benchmarks for US, EU, SEA, and LATAM, and a 48-hour action plan. Covers dropshipping ROAS and BE-ROAS. Trigger on 'why is CPA so high', 'my ads are not converting', 'ROAS dropped', 'we are burning budget', 'performance is bad this month', 'diagnose the funnel'. Also use when the user pastes a screenshot of numbers with no question attached. Not for — full account configuration audit, see `21-ads-audit-global`; writing a report someone else reads, see `07-marketing-report-global`; turning a raw export into insight, see `13-data-analysis-global`.

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

8 rules. About 0 are common, 7 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈7

    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. Distinctive:

    If you've never done a performance review:

    line 33
  2. Distinctive:

    One metric at a time: Don't try to fix CPM, CTR, CPL, ROAS simultaneously — pick the bottleneck and fix that first.

    line 39
  3. Distinctive:

    The 5-Whys method works: Ask "why?" five times until you hit a root cause that's actionable (process, system, or skill gap — not a symptom).

    line 40
  4. Distinctive:

    Auto-pull data: if MCP integrations are connected (Meta Official MCP, Google Ads MCP, TikTok Ads MCP, Pipeboard).

    line 55
  5. Distinctive:

    Ask user up to 4 questions:

    line 59
  6. Very specific:

    Ask "Why?" 5 times until you reach a process/system/skill root cause:

    line 169
  7. Distinctive:

    Use the variant that matches your ad account billing region, NOT just where the user is located:

    line 207
  8. Distinctive:

    If running multi-region: load multiple variants, compare benchmarks side-by-side, set per-region targets.

    line 216

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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

Performance Evaluation (Global)

Operating rules: skills/en/references/operating-rules-global.md (esp. rules 2, 3, 7).

Diagnose marketing performance — paid ads, organic, funnel — find root cause, propose optimization with 48h action plan and weekly checklist. Region variants (US/EU/SEA/LATAM) provide localized benchmarks.


For newbies — Read this first

If you've never done a performance review:

  1. Performance evaluation = a structured health check of your marketing. You collect numbers, compare against benchmarks, find the gap, and propose fixes.
  2. You don't need everything perfect to start. Even rough data (last 30 days spend + leads + revenue) is enough for a first pass.
  3. The order matters: measure first, diagnose second, fix third. Don't jump to "let's redo creative" before checking why CTR dropped.
  4. Benchmarks are regional. A $10 CPM in the US is normal; in SEA it's expensive. Always pick the right region variant.
  5. One metric at a time. Don't try to fix CPM, CTR, CPL, ROAS simultaneously — pick the bottleneck and fix that first.
  6. The 5-Whys method works. Ask "why?" five times until you hit a root cause that's actionable (process, system, or skill gap — not a symptom).
  7. Health Score 0-100 gives you a single number to track. Below 60 = stop scaling, fix first. Above 75 = safe to optimize and grow.

Step 0 — Read context + select region variant

Before diagnosis:

  1. Read .agents/product-marketing-context-global.md — get product, audience, region, currency, current channels.
  2. Pick the region variant:
    • variants/01-us.md — North America (USD)
    • variants/02-eu.md — Europe (EUR/GBP, GDPR-aware)
    • variants/03-sea.md — Southeast Asia (USD/local, low CPM)
    • variants/04-latam.md — Latin America (USD/BRL/MXN, WhatsApp-first)
  3. Auto-pull data if MCP integrations are connected (Meta Official MCP, Google Ads MCP, TikTok Ads MCP, Pipeboard).

Information gathering

Ask user up to 4 questions:

  1. Which channel(s) to audit? Meta / Google / TikTok / LinkedIn / email / organic / all?
  2. Current numbers? Spend, impressions, clicks, CTR, CPM, CPC, leads, CPL, conversions, CPA, ROAS, time period.
  3. What's the issue? CPM rising / ROAS dropping / lead quality poor / no orders / creative fatigue?
  4. Original target KPI? What was the goal — CPL, CPA, ROAS, monthly leads/orders?

Auto-pull via MCP (if connected)

PlatformMCP recommendedData pulled
Meta AdsMeta Official MCP (mcp.facebook.com/ads)List tools at connect time — performance, anomalies, benchmarks
Meta Ads (alt)Pipeboard, brijr/meta-mcpTargeting research, ad library
Google AdsGoogle Official MCPGAQL — cost, clicks, conv, impression share
TikTok AdsAdsMCP/tiktok-ads-mcp-serverCampaign performance reports
Cross-platformAdspirer ads-mcp175+ tools, unified metrics

The official Meta MCP can write, not only read, so any change recommended from MCP data follows rule 7 (draft-first, named approver) — never edit campaigns directly.

Quick MCP diagnostic flow:

1. ads_insights_performance_trend → trending down?
2. ads_insights_anomaly_signal → which KPI is abnormal?
3. ads_insights_industry_benchmark → vs. industry?
4. ads_get_opportunity_score → what does Meta suggest?
→ Feed into Benchmark Table + Diagnostic Tree

Universal diagnosis framework

Part 0 — Ads Health Score (0-100)

Quick screening score only. Follow rule 3: mark each check pass / fail / unknown / n_a; under 60% coverage give no score and list the unknowns. The full platform-checklist score is 21-ads-audit-global.

Health Score = Σ(Check_pass × W_severity × W_category) / Σ(Check_total × W_severity × W_category) × 100
SeverityWeightExample checkpoints
Critical5×Pixel not firing, CAPI not setup, CPA > 3× target
High3×near-duplicate concepts (differ on < 2 axes), budget < 5× CPA/day
Medium1.5×No retargeting audience, headline > 40 chars
Low0.5×Naming convention not followed
ScoreGradeAction
90–100AExcellent — focus on scaling
75–89BGood — fix Medium, scale gently
60–74CAverage — fix High before scaling
40–59DPoor — fix Critical + High first
<40FDanger — pause, full audit, rebuild

Part 1 — Diagnostic decision tree

[CPM HIGH]
  |-- CTR low (<1%)?
  |     |-- Weak hook → rewrite first 3s
  |     |-- Wrong targeting → narrow/expand audience
  |     |-- Stale creative (>7d) → make new creative
  |
  |-- CTR normal (1–3%) but CPM still high?
  |     |-- Auction peak → reduce budget, shift schedule
  |     |-- Audience too small → expand, test new LAL
  |     |-- Bid too high → switch to lowest cost
  |
  |-- CTR high (>3%) but CPM high?
        |-- Landing page/inbox not converting → fix LP
        |-- Unclear CTA → rewrite CTA
[ROAS LOW (<2x)]
  |-- CPM ok but few conversions?
  |     |-- Funnel leak → check each step
  |     |-- Slow sales response → improve speed, script
  |     |-- Wrong offer/price → review offer
  |
  |-- High CPM + few conversions?
  |     |-- Loop back to CPM tree
  |
  |-- Many orders but low AOV?
        |-- Weak cross-sell/upsell → bundles, larger packages
        |-- Bargain hunters → tighten targeting
[POOR LEAD QUALITY]
  |-- Many messages but few qualified leads?
  |     |-- Targeting too broad → narrow with LAL of best customers
  |     |-- Wrong-fit content → adjust hook + CTA
  |     |-- Value prop unclear → educate before CTA
  |
  |-- Qualified leads but no booking?
  |     |-- Slow follow-up (>2h) → respond within 15 min
  |     |-- Weak close script → rewrite, train sales
  |     |-- Price too high → bundle, financing, trial
  |
  |-- Booking but no-show?
        |-- No reminder → auto-remind 24h + 2h before
        |-- Trust gap → nurture more before booking
        |-- Decision fatigue → simplify booking flow

Part 2 — Root Cause Analysis (5 Whys)

Ask "Why?" 5 times until you reach a process/system/skill root cause:

RoundWhy?Answer
1Why did CPM rise?CTR dropped from 2.5% → 1.1%
2Why did CTR drop?Creative ran 12 days, audience saturation
3Why did creative run so long?No creative refresh schedule
4Why no schedule?No SOP for periodic creative review
5Why no SOP?No creative rotation process defined

Root cause: Missing SOP for creative rotation Solution: Lock in 3-5 new creatives/week, review every 3-5 days

Part 3 — Creative fatigue indicators

IndicatorWarningDangerAction
Frequency>2.5>4Refresh creative, expand audience
CTR decline-20% vs first 3 days-40%New creative
CPA increase+25% vs first 3 days+50%Diagnostic signal: check frequency and creative, prepare new creative. Any pause decision follows the house ladder below, not this metric alone
Run time>7 days>14 daysMandatory refresh
Engagement decline-30% comments/shares-50%Switch angle
Negative feedback>3%>5%Pause immediately

Part 4 — Audience saturation indicators

MetricHealthyWarningAction
Audience overlap (between ad sets)<20%>30%Merge or exclude
Frequency<2.5>4Expand audience, new creative
Reach vs audience size<50%>70%Audience too small, expand
Cost per incremental resultStable+30%Saturated — find new audience
% new users in clicks>60%<40%Re-showing to existing — refresh

Decision tree — which variant to use

Use the variant that matches your ad account billing region, NOT just where the user is located:

RegionVariantWhen to use
United States, Canada01-us.mdAccount billed in USD, US/CA traffic
UK, EU, EEA02-eu.mdAccount billed in EUR/GBP, GDPR consent stack
SEA (Vietnam, Thailand, Indonesia, Philippines, Malaysia, Singapore)03-sea.mdLower CPM, mobile-first, super-app context
LATAM (Brazil, Mexico, Colombia, Argentina, Chile)04-latam.mdWhatsApp-dominant, low CPM, FX volatility

If running multi-region: load multiple variants, compare benchmarks side-by-side, set per-region targets.


Dropshipping KPI section

Dropshipping has different success criteria than typical service/B2B marketing. Use this layer when product is sourced from suppliers (AliExpress, CJ, Zendrop) and shipped directly to customer.

Core KPIs

KPIFormulaHealthy target
ROASRevenue / Ad spend2.5–3x minimum, 4x+ ideal
BE-ROAS (Break-even ROAS)1 / Contribution margin (before ads)If contribution margin = 30%, BE-ROAS = 3.33x
Contribution margin (before ads)(Revenue − COGS − Shipping − Payment/platform fees − Returns) / RevenueBase for BE-ROAS and the CAC target
Profit margin (net, includes ad spend)(Revenue − COGS − Shipping − Fees − Ad spend) / Revenue20%+ healthy, 25%+ ideal. Never use it as the BE-ROAS base
CAC (Customer Acquisition Cost)Ad spend / Customers acquired< AOV × contribution margin before ads
CPA (Cost Per Add-to-cart / Purchase)Ad spend / Conversions<30% of AOV
AOV (Average Order Value)Revenue / OrdersHigher AOV = lower BE-ROAS pressure

BE-ROAS formula

BE-ROAS = 1 / Contribution margin (before ads)

Examples (Margin = contribution margin before ads):
  Margin 20% → BE-ROAS = 5.0x   (you need 5x ROAS just to break even)
  Margin 25% → BE-ROAS = 4.0x
  Margin 30% → BE-ROAS = 3.33x
  Margin 35% → BE-ROAS = 2.86x
  Margin 40% → BE-ROAS = 2.5x
  Margin 50% → BE-ROAS = 2.0x

Below BE-ROAS = losing money on every order. 1.2× BE-ROAS = healthy growth zone. 2× BE-ROAS = scale aggressively.

Dropshipping diagnostic flags

FlagThresholdAction
ROAS < BE-ROAS for 7 daysAlways losingOnce spend >= 3x target CPA: propose a pause through rule 7 (draft block, named approver), then redo product/creative
Profit margin < 20%Too thinRaise price OR find cheaper supplier
Refund rate > 5%Too highAudit product quality + ad claims
Chargeback rate > 1%CriticalReview CB reasons, may risk Stripe ban
Shipping time > 18 daysLongSwitch to faster supplier or warehouse stock
Cart abandon > 75%Too highAudit checkout (shipping cost reveal, slow load)

When to scale dropshipping

SignalThresholdDecision
ROAS sustained > 2× BE-ROAS7+ consecutive daysScale +20% per cycle
30+ purchases at target CPAWithin ad setCBO scaling allowed
Profit margin stable > 25%14+ daysOpen new market
Email/SMS list > 1000ReachedAdd retention campaigns

See skill 29-dropshipping-mastery-global for full dropshipping playbook.


48-hour action plan template

Apply when CRITICAL issue detected:

#TimeActionSeverityExpected outcome
10–2hApply the house ladder (below) and propose pause for creatives/ad sets at CPA > 3.0x target (rule 7 draft block)CRITICALStop bleeding spend
22–4hDeep-dive data: which creative, audience, time slotCRITICALIdentify root cause
34–8hDuplicate winning ad set, test new creativesHIGHNew creative live
48–12hAdjust audience: exclude overlap, test new LALHIGHReduce frequency
512–24hA/B test 3 new hooks for top creativeHIGHFind better hook
612–24hReview landing page / inbox flowMEDIUMFix funnel leak
724–36hCompare new vs. old performanceMEDIUMDecide next step
836–48hReport + propose week-2 planMEDIUMConcrete plan

House heuristic ladder (no EN gate file; verify against your own account data): spend < 3x target CPA = not enough data; <= 1.0x = winner; 1.0-1.5x = watch; > 1.5x for 7 days = change creative or audience; > 3.0x = propose pause (rule 7 draft block); 0 conversions at >= 3x spend = the concept is wrong, go back to customer insight (09-customer-insight-global).


Weekly optimization checklist

DayFocus
MonCompare WoW KPIs, identify top/bottom 3 creatives, check frequency + audience overlap
TueMake 3-5 new creatives, test new hook, review UGC pipeline + organic
WedOptimize bid/budget on winners, propose pause per the house ladder (CPA > 3.0x target, rule 7 draft block), test new audiences
ThuCheck Mess→Lead, Lead→Booking, response time, sales script
FriWoW table, top 3 priorities for next week, content calendar update

Cross-reference

NeedSkill
Full audit with 84 checkpoints21-ads-audit-global
Monthly report07-marketing-report-global
Recompute budget from revenue target10-reverse-kpi-global
New ad copy with frameworks05-ad-copy-global
New video script04-script-video-global
Replan from scratch00-marketing-plan-global
Dropshipping-specific playbook29-dropshipping-mastery-global

Quality checklist

Before delivering evaluation:

  • Region variant loaded — benchmarks match user's market
  • Health Score estimated (if data sufficient) — <60 means fix first
  • User data filled into benchmark tables
  • Diagnostic decision tree run — root cause named
  • 5 Whys done — process/system root cause, not symptom
  • Creative fatigue + audience saturation checked
  • WoW + MoM trend analysis included
  • 48h action plan has time, owner, expected outcome
  • Top 3 priorities with deadlines
  • If dropshipping: BE-ROAS calculated, profit margin verified
  • All numbers verifiable — no vague estimates
  • Recommendations realistic for user's resources

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.