# Evaluate Paid Campaign Changes with Controls

Paid campaign changes can look successful when attribution, learning phase, or seasonal effects distort results. This skill compares a fixed candidate window with a control and gives a documented decision.

## Install

```bash
npx skillstore add aaron-he-zhu/paid-measurement-loop
```

## Metadata

- Status: approved
- Slug: aaron-he-zhu-paid-measurement-loop
- Version: 19.0.0
- Author version: 19.0.0
- Skillstore revision: r2
- Version status: valid
- Tree hash: a3df3392cee1eba81a4b650cfc4354c0514ce5469e8c4c238bc376bfdd34da78
- Author: aaron-he-zhu
- GitHub username: aaron-he-zhu
- License: Apache-2.0
- Repository: https://github.com/aaron-he-zhu/aaron-marketing-skills/tree/main/ad/scale/paid-measurement-loop
- Ref: 0715a6e09ea875c8e28cb705ce87cc83045e69c1
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: allowed
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands, network, filesystem
- Quality score: 72
- Public page: https://skillstore.pages.dev/skills/aaron-he-zhu-paid-measurement-loop
- Manifest: https://skillstore.pages.dev/api/skills/aaron-he-zhu-paid-measurement-loop/manifest

## Capabilities

- Defines a fixed readback window after a campaign exits learning phase.
- Compares ROAS or CPA changes against an unchanged control.
- Normalizes attribution windows and currency before cross-platform comparisons.
- Classifies a change as Promote, Keep-testing, Rollback, or Unproven.
- Flags broken tracking and duplicate conversion attribution as reasons to defer a decision.
- Produces a reusable readback summary and next-readback date.

## Use Cases

- Review a budget increase: Compare performance after a budget change with an unchanged campaign over the same fixed window.
- Assess new creative: Decide whether new creative should be promoted, tested longer, or rolled back using control-based evidence.
- Prepare a stakeholder readback: Create a clear decision record with windows, controls, attribution notes, and measurement limitations.

## Prompt Templates

### Review one campaign change

```
Read back the budget increase for Campaign X. Compare ROAS with its unchanged control over the last 14 days.
```

### Check creative performance

```
Evaluate the new prospecting creative against the prior creative. Use CPA as the primary metric and state the readback decision.
```

### Compare platform results

```
Compare the attached Meta and Google exports. Normalize currency and attribution windows before assessing ROAS against each control.
```

### Handle uncertain tracking

```
Review this campaign change and the conversion exports. Mark the result Unproven if duplicate orders or broken conversion tracking affects the evidence.
```

## Limitations

- Requires campaign, analytics, or ecommerce exports supplied by the user.
- Does not access advertising platforms or estimate results from dashboard screenshots alone.
- Delegates ROAS and CPA arithmetic to the related roi-calculator skill.
- Does not make RQS scores or final measurement-integrity veto decisions.

## Best Practices

- Fix the readback window and control before interpreting performance data.
- Provide exports with dates, spend, conversions, revenue, attribution windows, and currency.
- Label every figure as measured, user-provided, or estimated.

## Anti Patterns

- Do not declare success from a raw before-and-after comparison without a control.
- Do not compare platform ROAS before normalizing attribution windows and currency.
- Do not make a decision while the campaign remains in learning phase or tracking is unreliable.

## Security Audit

- Audited at: 2026-07-27T11:36:05.711\+00:00
- Summary: Most static alerts are false positives caused by Markdown code formatting and static repository links. One instructed ledger command interpolates a campaign placeholder into a shell command, and the result filename convention lacks a campaign-name sanitization rule. No prompt injection, credential collection, or unauthorized network behavior was found.

## Stats

- Views: 0
- Downloads: 6
- Favorites: 0
- Popularity score: 0