Why This Approach Works
Media buying has a dirty secret that nobody talks about:
Most of the work is the same thing over and over again.
Pull data. Look for patterns. Spot what's broken. Decide what to change. Execute. Wait. Repeat.
The strategic thinking matters. The creative direction matters. The big-picture decisions matter.
But the actual analysis? That's pattern recognition. And pattern recognition is exactly what AI does well.
The core idea: Instead of using AI as a generic assistant, set up specialized agents — each focused on one specific part of media buying. They handle the analytical work. You handle the decisions and execution.
This isn't about replacing media buyers. It's about not doing the same analysis manually every single week.
What Changes
| Before | After |
|---|---|
| Pull data manually every Monday | Still pull data (agents need input) |
| Spend 2-4 hours analyzing campaigns | 15-30 minutes reviewing agent outputs |
| Build your own pivot tables and calculations | Get structured recommendations instantly |
| Sometimes miss things when you're tired | Consistent analysis every time |
| Hard to document your analysis process | Prompts serve as documentation |
Key insight: The goal isn't to automate everything. It's to automate the parts that don't require human judgment — so you can spend more time on the parts that do.
The 4 Agents
Each agent has one job. One focus. One type of analysis.
This matters because generic "analyze my ads" prompts give generic answers. Specialized prompts give specific, actionable outputs.
Bid & Budget Manager
Where should I spend money, and how much?
What It Analyzes
- Spend efficiency across campaigns
- Budget pacing — too fast, too slow, or on track
- Bid performance optimization opportunities
- Wasted spend identification
What You Get
- Specific bid adjustment recommendations
- Budget reallocation suggestions with reasoning
- "Pause immediately" list
- Pacing alerts for over/under delivery
Data Needed
Campaign name, Spend, Conversions, CPA, ROAS, Impressions, CTR — last 7-14 days from Google or Meta Ads Manager
Creative Analyst
Why are my ads working or not working?
What It Analyzes
- Creative performance patterns
- Hook and angle effectiveness
- Format performance (static vs. video vs. carousel)
- Creative fatigue signals
What You Get
- Top performers with WHY explanations
- Failing ads analysis with specific reasons
- Fatigue alerts with refresh recommendations
- Kill/scale/iterate decisions per creative
Data Needed
Ad name, Spend, Impressions, CTR, CPC, Conversions, CPA, Frequency — plus ad copy/descriptions if available
Audience Architect
Who should I target, and where can I expand?
What It Analyzes
- Audience efficiency by segment
- Saturation signals (rising frequency, CPMs)
- Overlap issues between audiences
- Scaling opportunities
What You Get
- Audience efficiency ranking
- Saturation alerts before performance tanks
- LAL percentage recommendations
- New audience ideas based on winners
Data Needed
Ad set name, Audience, Spend, Reach, Frequency, CPM, Conversions, CPA, ROAS
Performance Auditor
What's broken and what am I missing?
What It Analyzes
- Overall account health
- Anomalies and concerning trends
- Hidden losers and hidden winners
- Attribution gaps
What You Get
- Account health assessment
- Anomaly detection with potential causes
- Platform vs. backend discrepancy report
- Prioritized action list
Data Needed
Campaign name, Status, Spend, Clicks, Conversions, CPA, ROAS — plus backend/CRM data if available
How to Set This Up
The setup takes about 10 minutes. After that, it's just a matter of running the prompts with fresh data each week.
Create a Claude Project
Create a dedicated project for your ad account analysis. Call it "Media Buying Squad" or whatever makes sense for your workflow.
This keeps all your prompts and conversations organized in one place. You can also add account-specific context to the project instructions — things like your target CPA, main KPIs, or business context that stays consistent across analyses.
Set Up Your Data Exports
Each agent needs specific data. The easiest approach is to create saved report templates in Google Ads or Meta Ads Manager that you can export with one click.
For Google Ads: Create custom reports at the campaign, ad group, and ad level with the metrics each agent needs. Save them for quick access.
For Meta Ads: Same idea — create saved breakdowns at the campaign, ad set, and ad level. Export as CSV or just copy-paste the table directly.
Get the Prompts
Grab the prompts from the GitHub repo. Each one is designed to work standalone — you don't need to run all four every time. Pick the one that matches what you need to analyze.
Run Your First Analysis
Start with the Performance Auditor. It gives you the big picture of account health and helps you identify which other agents to run based on what it finds.
Copy the prompt, paste your data where indicated, fill in your targets and context, and run it.
Review and Execute
Review the recommendations. Validate that they make sense for your specific situation. Then execute the changes in-platform.
The agents give you recommendations — you make the final call on what to implement.
Important: These agents analyze data you provide. They don't connect directly to your ad accounts, can't see historical trends beyond what you share, and can't make changes automatically. This is intentional — you stay in control.
Weekly Workflow
Here's what a typical Monday morning looks like with this system:
| Time | Action |
|---|---|
| 5 min | Export fresh data from Google Ads and/or Meta Ads Manager |
| 5 min | Run Performance Auditor for overall health check — this tells you where to focus |
| 5 min | Run Bid & Budget Manager if auditor flagged spend issues |
| 5 min | Run Creative Analyst if performance is dropping or you're scaling |
| 5 min | Run Audience Architect if you see saturation signals or need expansion ideas |
| 10 min | Review all outputs, make decisions, execute changes in platform |
Total time: 30-35 minutes
Compare that to manually building pivot tables, calculating efficiency metrics, and writing up recommendations — which typically takes 2-4 hours.
When to Run Each Agent
You don't need to run all four agents every week. Here's when each one is most useful:
- Performance Auditor: Every week. This is your starting point — it shows overall health and flags where to dig deeper.
- Bid & Budget Manager: Weekly, or whenever you need to reallocate spend. Essential before scaling or after performance changes.
- Creative Analyst: When performance is dropping, when you're testing new creatives, or when planning a creative refresh.
- Audience Architect: When you see saturation signals (rising CPMs, frequency above 3), when planning to scale, or monthly as a targeting review.
Tips for Better Results
1. More Data = Better Analysis
Include at least 7 days of data. 14-30 days is better for spotting trends. Less than 7 days often doesn't have enough signal for meaningful patterns.
2. Include Your Targets
Always fill in your target CPA, ROAS, or other KPIs. The agents use these to calibrate what "good" and "bad" performance looks like for your specific situation.
3. Ask Follow-Up Questions
After the initial analysis, dig deeper. Ask things like:
- "Why do you think Campaign X is underperforming compared to Campaign Y?"
- "What would you recommend if I need to cut budget by 30%?"
- "Which of these recommendations should I prioritize if I can only make 3 changes this week?"
4. Combine Agents
Use agents together for deeper analysis. For example:
- After Bid & Budget Manager recommends scaling certain campaigns, ask Creative Analyst which creatives to prioritize for those campaigns
- After Audience Architect identifies saturated segments, ask it to suggest alternatives for those specific audiences
5. Include Backend Data When Possible
Platform-reported conversions vs. actual backend conversions is where the real insights live. If you have CRM data or actual revenue numbers, include them — especially for the Performance Auditor.
6. Run Consistently
A quick weekly analysis catches issues before they become expensive. Consistency matters more than depth.
Limitations
This system is useful, but it's not magic. Here's what it won't do:
- No real-time monitoring. These agents analyze data you provide — they can't watch your accounts continuously or alert you to sudden changes.
- No automatic execution. Recommendations require you to implement them manually. This is by design — you should review before making changes.
- Limited historical context. The agents only know what you share in that session. They don't remember last week's analysis unless you include it.
- Platform-specific nuances. The prompts work for both Google and Meta, but each platform has quirks. You may need to adjust recommendations based on platform-specific context.
- No competitive intelligence. These agents analyze YOUR data. They don't know what competitors are doing or industry benchmarks unless you provide that context.
For most media buyers, these limitations are fine. The goal is saving time on routine analysis — not replacing human judgment entirely.
Summary
4 specialized agents, each with one job:
- Bid & Budget Manager — where to spend and how much
- Creative Analyst — why ads work or fail
- Audience Architect — who to target and where to expand
- Performance Auditor — what's broken and what to fix first
Run them weekly. Review the outputs. Execute in platform.
You stay in control — Claude handles the repetitive analysis.
