Automate your entire Meta Ads workflow with Claude Code: less time, more control

The problem with managing Meta Ads campaigns

Over the past few months, we’ve spoken with several entrepreneurs and employees at SMEs who’ve reported the same issue. All of them run Meta Ads campaigns and manage them in-house (without any external support from agencies or consultants); they know how to navigate the platform but lack true technical expertise or a dedicated team focused solely on this task; thus, the main problem is the enormous amount of time wasted on campaign management, as well as the anxiety caused by uncertainty: “Am I spending my budget wisely? Could I spend less and more effectively? How much room do I have to improve performance (e.g., cost per lead)?”

Our solution

We therefore decided to develop a pilot project, which we are now replicating with other clients, by integrating in-depth expertise in performance marketing, analytics, and automation to automate the entire Meta campaign management process:

• weekly data and performance analysis
• technical optimization suggestions
• new creative concepts and copywriting
• graphic production
• uploading directly to Ads Manager

The real competitive advantage: using AI agentically

All of this has been implemented and runs on Claude Code, used agentically. Without getting into overly technical details in this article (Cesare has written this very detailed guide in case you’d like to dive deeper and build the system yourself), the key point, the potential that’s revolutionizing the way we work, is HOW artificial intelligence is used and applied (regardless of the model used, whether it’s Claude, ChatGPT, or another).

Most people use AI models like this: they open the chat, ask a question or type a prompt, and the model answers. The answers are useful, no doubt, and a real help in solving many problems, but the point is that this process has huge limitations: it starts and ends there. You still have to do the task yourself, which has often become even more complex in the meantime because the AI has added elements, given you ideas, and opened new doors. This is called cognitive overload: more information to process, not less.

The result: you don't save time, quite the opposite. And often, a sense of incompleteness sets in because the AI always points out possible improvements, and you never truly reach a conclusion.

Using AI agentically, on the other hand, means shifting paradigms and changing how we work with it: instead of using AI for one-off questions, you train it: you give it the context about your business, your goals, your processes, what you want it to do and how.

You do this just once. The AI learns and will continue to learn over time (always under your control: you decide how much freedom it has and what we want to maintain full control over), and it autonomously performs all the tasks you assign to it, with a level of precision that is becoming truly high.

Imagine: it’s effectively like having an employee or a team of employees, and just as when you hire someone, you need to invest quality time to onboarding the employee: you explain the company, the strategy, provide the full context, tell them exactly what their responsibilities and areas of expertise will be, what you expect from them, how things are done. In short, you set them up to perform at their best. The better trained an employee is at the start, the better and faster they’ll do their job.

So, when using AI agentically, the context you provide and the quality of the information you feed it are what matter most. Like a good manager, you need a clear picture of your processes and goals, and you need to pass it on. It's work you do once, and it pays you back for years.

What our system does

What have we achieved? In our case, the system running on Claude Code does the following:

1. Once a week it runs a deep analysis of our campaigns: ad groups, creative, key metrics. It’s exactly as if it were our team's analyst.
2. It hands us a full report: what's working, what isn't, what we can improve and why, along with suggested changes. Here it's acting as our performance marketing manager.
3. It brainstorms new creatives and copy; designs the creatives as if it were our in-house designer, and produces them directly (we used Higgsfield for graphic production) after receiving our approval and incorporating any changes we want to make.
4. Our agent then uploads the new creative and applies all the approved changes to our campaigns.

There you have it: our marketing team, or marketing assistant. We've essentially delegated all the execution; what's left on our side is oversight, approval, and requesting changes or a deeper look. Whenever we're not convinced or want to explore a suggestion, we brainstorm directly with Claude in chat to understand it better and stay on top of the decisions.

A few important disclaimers:

• You're the one who knows your context and your business better than anyone. You're the one who sets the boundaries the system operates within, and never steps outside.
• In our case, we trained the system with our own knowledge of marketing, performance, and analytics, meaning we gave the agent the skills that make it a genuine expert at what it does. Claude's baseline is already very strong, but it's the combination of its capabilities and your specific instructions that makes a huge difference.
• You ALWAYS keep control of what's happening: nothing goes live without your approval.
• The system keeps learning over time as it works, so you can expect performance to keep improving.

This way you've: delegated the execution that drains your time, freeing it up for other things; leveled up your capabilities (Claude's knowledge base is enormous, and can always be combined with your own expertise or external skills you teach it); gotten better performance; and cut management costs.

A few numbers from our first tests:

• Operational hours saved per week: 5–10
• Key campaign metrics improved (e.g. CPL −5–15%, CTR +10–15%)
• Optimization cycle: from monthly to weekly - consistent, not sporadic

Want to try building the system yourself? Here's our complete step-by-step guide: the GitHub template.

Want to tell us about your business and see if we can help? Write to us and book your free call with us.