Built on Claude · in production since Aug 2026
How I run your account — monitored daily, reviewed weekly with Claude
Behind every account I manage is an operations system I built on Anthropic’s Claude. It watches spend and conversions every day, flags problems early, and drafts a weekly review — which I read, question and act on myself.
13 accounts · Google Ads + Meta
An ops layer that never sleeps.
Daily anomaly monitoring
Hourly budget guard
Midday pacing check
Weekly review with Claude
Onboarding with Claude Code
Read-only by design
From raw data to a decision — in four steps
Data. Every morning the system pulls recent performance data from the Google Ads and Meta APIs — only for the accounts I manage, nothing more.
Rules. Each account is compared with its own history, not generic benchmarks: stopped delivery, unusual spend, conversion drops, budget overruns. Problems reach me the same morning.
Claude. Once a week, Claude reads the numbers together with each client’s goals and context, and drafts a structured review: what changed, why it likely changed, and what to do next.
Me. I review every draft against the account itself and decide what to change. Nothing is sent to you or changed in your account automatically.
Inside the system — real screens, client data blurred.
These are live screens from the system. Interface labels are shown in English for this page; client names, account IDs, figures and Claude’s written analysis are replaced with random characters and blurred before capture, so no client data appears here.
Daily monitoring — every account against its own 7-day median, colour-coded. Anything critical reaches me by Telegram at 8 am.
Weekly triage — thirteen accounts ranked by urgency, each with flags, week-over-week metrics and a written summary.
Claude’s weekly review for one account (shortened) — red alerts, evaluation, action plan, what went well and badly, and a client message draft I edit before anything is sent. Generated by claude-opus-5 through the API.
Week-over-week comparison, pulled from the Google Ads API with read-only access.
Under the hood — excerpts from the codebase.
Every weekly review comes back in one fixed structure — enforced as a schema through the Claude API, not just requested in the prompt.
Alerts compare each account with its own history (median, not average). A spike explained by a budget change is labelled as info, not an alarm.
Client onboarding runs as a Claude Code skill: it researches a new client and writes the context files Claude reads in every weekly review.
01
Does AI make changes to my campaigns?
No. The system only reads your account data. Every change in your account is made by me, after I’ve looked at the numbers myself.
02
What does Claude actually do?
It drafts the weekly analysis: how performance moved against your own baseline, what likely caused it, and which next steps are worth considering. I treat it as a sharp first draft, not a decision-maker.
03
Is my data safe?
The system only reads the accounts I manage for you, and your data is never shared with other clients or used in anyone else’s reports. The analysis itself is done by Anthropic’s Claude.
04
Why build this instead of using an off-the-shelf tool?
Generic tools alert on generic thresholds. My rules compare your account with its own history, and the weekly review uses your specific goals, margins and tone — context a generic tool doesn’t have.
05
Why are the screenshots blurred?
They come from live client accounts. Names, IDs, figures and written analysis are replaced and blurred before the screenshot is taken, so nothing on this page can identify a client. The interface itself runs in Turkish; labels are shown in English here.