What does this skill do, and when should you use it?
This skill reconciles Meta, Google, and other paid-platform conversion reports against a GA4 or ecommerce order-ID truth set. It normalizes attribution windows and currencies, identifies cross-platform duplicate credit, and compares attribution models. When geo or holdout data exists, it reads incremental performance; otherwise, incrementality is marked not applicable. It does not calculate ROAS, CPA, or ROI, and it does not perform the point-in-time ad-account audit.
It reads a GA4 or ecommerce order-ID export, each platform's conversion export, the stated attribution window and currency, and optional geo or holdout data. It normalizes windows and monetary values before matching conversions by order ID, with timestamp-plus-value matching as a fallback. It labels claims as matched, double-counted, or unmatched; reports deduplicated conversions and the double-count rate; compares credit allocation under at least two attribution models; and reads test-versus-control incrementality when a holdout is available. It can save a reconciliation workbook to the specified memory paths only after the user agrees.
- A marketing analyst sees Meta and Google reporting more sales than the Shopify order export and needs to identify duplicated credit.
- A paid-media lead needs a monthly reconciliation that puts Meta's 7-day click window and Google's 30-day window on a common basis.
- A growth team has test-region and control-region orders plus platform spend and wants to compare incremental orders with last-click attribution.
- An ecommerce team wants clean, deduplicated conversion counts before handing the data to an ROI or ROAS analysis.
How do you install this skill?
- This is a static file review only; the skill, scripts, and reconciliation workflow were not executed against real data.
- Manually validate timestamp-plus-value fallback matching, exchange-rate selection, attribution-window normalization, and geo/holdout incrementality denominators.
- Order IDs, values, and exports may contain sensitive commercial or personal data; redaction, retention, and runtime-encryption requirements are not explicit in the skill.
- Referenced shared materials and downstream skills were not all supplied, so handoff behavior remains unverified.
- Local filesystem
This skill is located at ad/scale/attribution-reconciler/SKILL.md in the GitHub repository aaron-he-zhu/aaron-marketing-skills. In Claude Code, register the marketplace with “/plugin marketplace add aaron-he-zhu/aaron-marketing-skills” and install the collection with “/plugin install aaron-marketing@aaron”. Agent Skills-compatible hosts can use “npx skills add aaron-he-zhu/aaron-marketing-skills”. The repository can also be obtained with “git clone https://github.com/aaron-he-zhu/aaron-marketing-skills”. The source does not document a repository-level command for installing only this skill.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/aaron-he-zhu/aaron-marketing-skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/ad/scale/attribution-reconciler" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- Reconcile my paid conversions for May. Truth set is this GA4 order-ID export. Here are the Meta and Google conversion exports. Find the double-counting.
Provide the GA4 or ecommerce order-ID export, each platform conversion export, the attribution window and currency for each platform, and any available test/control export. A direct trigger is: “Reconcile my paid conversions for May. Truth set is this GA4 order-ID export. Here are the Meta and Google conversion exports. Find the double-counting.” If the order-ID truth set is missing, the skill should return NEEDS_INPUT rather than reconcile against a platform-reported count. Send deduplicated real-order counts to roi-calculator for ROAS, CPA, or ROI math.
What are this skill's strengths and limitations?
- Uses GA4 or ecommerce order IDs as the sole truth set instead of trusting reported platform totals.
- Covers window and currency normalization, order matching, cross-platform deduplication, model comparison, and holdout-based incrementality.
- Keeps matched, double-counted, and unmatched claims separate, which suits recurring monthly reconciliation.
- Works from manually exported account data and does not require paid-ad APIs or MCP.
- Cannot reconcile without the order-ID export.
- Does not calculate ROAS, CPA, ROI, or EMV; those are delegated to roi-calculator.
- Cannot infer incrementality without a geo or holdout test.
- The supplied source does not document a dedicated test suite or concrete platform-export examples for this individual skill.
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Paid Attribution Reconciler this page | 59 · Recommended | ★ 3k | 12d ago | Apache-2.0 |
| Paid Ads Account Auditor | 59 · Recommended | ★ 3k | 12d ago | Apache-2.0 |
| Ad Experiment Designer | 59 · Recommended | ★ 3k | 12d ago | Apache-2.0 |
| Paid Ad Budget Pacing Monitor | 61 · Recommended | ★ 3k | 12d ago | Apache-2.0 |
| TAO DAFT Dataset Converter ✓ NVIDIA · Official | 54 · Use with care | ★ 3.6k | 12d ago | Apache-2.0 |
Unlike ad-account-auditor, this skill performs recurring order-level reconciliation and deduplication, while ad-account-auditor owns the point-in-time R2 veto and RQS gate. Unlike roi-calculator, it cleans and standardizes real order counts; roi-calculator owns ROAS, CPA, and ROI calculations. Unlike dark-social-attributor, it covers paid channels only and does not decompose GA4 direct traffic or organic dark-social attribution.
How did FollowSkills review this skill?
The skill limits inputs to manually supplied exports from the user's own accounts, avoids required APIs, treats exports as untrusted data, and requires confirmation before saving memory. Points are deducted because sensitive order data, unencrypted runtime memory, deletion/rollback boundaries, and skill-level data-flow disclosure remain incomplete, with no clear skill-specific provenance attribution.
The procedure, NEEDS_INPUT behavior, reconciliation labels, completion criteria, and abnormal-input feedback are reasonably explicit. Points are deducted because static review cannot reproduce the key paths; timestamp-plus-value fallback matching is ambiguous, incrementality denominators and malformed-data handling are underspecified, and several handoffs depend on referenced skills not supplied here.
Triggers, inputs, outputs, exclusions, and the argument hint are clear, with Chinese labeling and a manual-data path. Points are deducted because mainland-China reachability is not evidenced, and use requires multiple platform exports, currency/window metadata, and optionally a valid holdout dataset, limiting applicability.
The skill has layered documentation, Quick Start examples, a shared contract, data-source notes, numbered instructions, handoffs, Apache-2.0 licensing, versioning, and host compatibility. Points are deducted because maintenance ownership, changelog details, and update path are not clear for this skill; the repository security policy supports 17.x while the skill declares 18.0.0, creating governance inconsistency.
The claimed deliverable includes a row-level match table, deduplicated counts, normalized windows and currency, model comparison, and holdout incrementality, which is plausibly actionable. Static calibration limits this to 7 because no executed representative output or third-party result verifies the key paths, and fuzzy matching and incrementality interpretation may require substantial review.
The skill points to repository-level ROAS, measurement, security, and handoff materials, while the repository includes CI and test files. Points are deducted because those tests do not cover this skill's core reconciliation workflow, and there is no independent reproduction record, third-party execution evidence, or external-source corroboration; static calibration caps this at 5.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →