How to use the Ad Campaign Analyzer agent skill

What the Ad Campaign Analyzer skill makes the agent do

Once installed, the skill changes how the agent responds when you share ad campaign performance data. Instead of summarizing numbers, it works through a diagnostic process: it normalizes data from Google Ads, Meta Ads Manager, or LinkedIn Campaign Manager into a common format, checks campaigns against health metrics such as CTR, CPC, conversion rate, CPA and ROAS, and flags budget waste and winners.

It also checks whether A/B test differences are statistically meaningful, and when you have several channels it ranks their efficiency and can propose a budget split across them. The skill's own minimums are 100 clicks per variant for CTR tests and 30 conversions per variant for CPA tests; below those, the skill treats results as possibly just noise.

• Flags zero-conversion spend, high-CPA outliers, low-CTR ads and broad-match bleed
• Identifies top keywords, winning ads, best audiences and best times
• Recommends what to pause, scale or test next
• Works only from the data you provide, with no connection to your ad accounts

Install the skill from the ZIP download

Download the ZIP for the skill from this page. Then unzip it into one of two locations. For a personal install available across your projects, use ~/.claude/skills/. For a skill that applies only to one project, put it in that project's .claude/skills/ folder.

If you use the Claude apps instead, you can upload the skill under Settings → Capabilities → Skills rather than placing files by hand.

• Personal: unzip into ~/.claude/skills/
• Single project: unzip into the project's .claude/skills/ folder
• Claude apps: Settings → Capabilities → Skills, then upload

Trigger the skill with a natural request

You do not need a special command. The skill is meant to activate when you share performance data and ask what to cut, scale or test. Share a CSV export, a pasted table, or a screenshot, then ask in plain language. Phrases the skill is written to respond to include "analyze my ad campaigns", "where am I wasting ad spend", "reallocate my ad budget", "which ads are actually working" and "ROAS analysis".

The skill is designed not to trigger without results. Asking for campaign planning or new creative ideas with no performance data will not invoke it, so include the numbers.

• Give the time period, monthly budget and your main goal (for example leads or purchases) if you have them
• Say if you changed creative, budget or targeting during the period, since the skill asks about this
• If you have a target CPA or ROAS, state it; otherwise the skill benchmarks against generic figures

Example requests and what to expect

A useful request names the platform, attaches the data and states the decision you face. For example: "Analyze my Google Ads performance for March, here is the CSV. Which keywords should I pause?" Or, for a split across channels: "I have $5,000 a month for ads across Google and LinkedIn. Here are last month's exports. How should I divide it?"

Expect a structured answer: a normalized table, campaign health statuses, waste and winner lists, significance verdicts for any tests, and concrete next actions. Check the numbers against your own dashboard before you act on them.

Limits and safety notes

The skill runs no scripts and makes no network calls. It does not connect to Google, Meta or LinkedIn and needs no API keys, so it can only analyze what you paste, upload or screenshot. Its benchmarks and thresholds are generic rules of thumb, not your industry's real figures.

Because it sees only your exported data, it cannot verify tracking, attribution or account settings. Small samples are a common source of false conclusions, so treat early winners with caution.

• Runs scripts: no
• Uses network: no
• Benchmarks are generic; verify the math yourself for high-stakes budget decisions

Tips for better results

Export full, recent data that includes the columns the skill expects for your platform, such as Campaign, Impressions, Clicks, CTR, CPC, Conversions and Cost. Include funnel figures such as lead-to-MQL or close rate if you have them, because the skill can then estimate a funnel-adjusted acquisition cost per channel.

Ask one decision at a time. A question like "which ads should I kill?" gets sharper answers than a broad request for everything at once.

• Use the same date range across all channels
• Include the time period and monthly budget in your request
• Re-run the analysis after a fair test period rather than after a few days