I Built an Adv Ad Automation System with Claude Code in 3 Days

Last year I wrote a post called “Why You Should Get into Media Buying Early”.

That post was mainly about how to run Advertizer (Adv for short) accounts, plus some hands-on tips.

Starting this year, I’ll be writing more about AI + ad automation.

Especially this: how to actually put vibe coding to work inside your business and turn it into a sustainable closed loop.

Freeing your time from repetitive operations, so you can spend it thinking about business models and studying cases.

Judging by the feedback I get, plenty of people have run Adv, but not many have actually stuck with it.

The pain points are clear too. It’s not that people don’t know how to run ads; they get stuck on these things:

1️⃣ Accounts are hard to get, and QS is hard to keep steadily above 80% for long

2️⃣ You don’t dare create ads often, GEOs get shut off, and the entry points keep shrinking

3️⃣ In serious cases the CPA feature gets suspended outright

But in my experience, Adv ads aren’t something you need to create fresh every day. Sometimes one ad runs steadily for 1-2 months, and the longest-lived ad I’ve ever had survived close to 1 year.

There’s only one precondition, of course: you understand well enough how this platform works. And that’s exactly why I spent 3 days using Claude Code to write myself an ad automation system.

Below is a quick rundown of what the system does. I hope it also gives you an idea of where AI can genuinely make an ad business more efficient.

The goals I set for this system are very clear

1️⃣ QS below 80%: the system steps in directly

The system monitors QS continuously. As soon as it detects a drop below 80%, it triggers the intervention logic and does its best to keep the whole account’s performance score above 80%.

There’s only one purpose behind this: making sure the account always stays eligible to keep creating ads and getting traffic entry points.

2️⃣ Automatic analysis of platform trends + the Submission Report

Every so often, the system pulls the platform’s trend report, so it never misses a region where traffic is ramping up.

At the same time it combines that with Adv’s Submission Report and recommends whether a new ad is worth creating.

3️⃣ Auto-generating Bemob ad links

When the platform conditions are right for a new ad, the system automatically generates the Bemob ad link for the corresponding region, avoiding the omissions and errors that come with creating them by hand.

4️⃣ Auto-mapping to Adv with full validation

The generated ad link is automatically mapped onto the Adv platform, and once every parameter and setting has been automatically confirmed correct, the ad gets created.

5️⃣ The system “learns” the whole creation process

The system isn’t set in stone. It keeps learning:

For example: which creation recommendations ended up running better, how QS trends at different stages, how stability differs across GEOs, and plugging in AI (Gemini 3 Flash Preview, for instance) to help come up with new creation recommendations.

So what problems does this system ultimately solve?

• It steps in automatically when QS slips, extending the ad’s lifespan
• It gives creation recommendations based on the account Report
• It creates the ad links and publishes the ads automatically
• After publishing, it keeps monitoring QS dynamically, reducing manual intervention

A more interesting feature: dynamic ad arbitrage

Based on Bemob’s Send Postback % and Revshare %

You preset a target ROI

And the system automatically adjusts the cost-to-revenue ratio of the QS interventions

Down the line it can also take on offer-mix adjustment, handing things that used to rely on experience and gut judgment over to a system that has learned from the data.

Honestly, in those 3 days I lost count of how many versions of this system I went through.

But the biggest takeaway this time really comes down to one sentence:

Vibe coding can massively improve business efficiency, provided you know the business itself well enough.

If the process and its key points aren’t clear to begin with, automation only amplifies the chaos instead of solving problems and improving efficiency.

Only when you think the requirements through from a product manager’s perspective can Claude Code become a tech team with serious execution power.

In the end you’ll find:

Judgment matters more than raw coding skills.

Next, I’ll keep using AI to build more automation tools and automate the ads I’m running.

For example, the “Build Your Survey Lander: the arbitrage play” I mentioned earlier can be run with AI automation tools, but you need a clear idea of how to play it before you can build it. For now, let’s call it the SaaS tracking arbitrage edition.