How I run 100+ Google Ads accounts by myself: the full system

Monday, 7am. I open one screen and see the health of every account I run. Over 100 of them, anywhere from $300k to $2.5M moving through them in a given month. Three years ago that took a team of six people and a standing Tuesday call. Now it takes me, a laptop, and a stack of Claude skills that read spend reports before I've had coffee.

This edition is the real system, not the highlight reel. What a "skill" is in plain terms, the weekly rhythm that keeps 100-plus accounts from drifting, what got automated, and the one job I will never hand off no matter how good the model gets.

**What a skill is**

A skill isn't a plugin and it isn't magic. It's a folder of instructions, the same folder every time, that tells Claude exactly how I want a job done. What "wasted spend" means in my accounts. What a real audit checks, tier by tier. What format a client report takes before it goes out. Once I write it, I never explain it again.

Think of it as onboarding one new hire, one time, except that hire never forgets the process and runs the same checklist on account four as it does on account ninety.

[IMAGE-1: Hyper-realistic screenshot of the Claude desktop app mid-conversation, dark mode UI, left sidebar showing a list of named "skills" (google-ads-audit, budget-pacing, creative-scoring, weekly-report), main chat pane showing Claude running a structured audit against a Google Ads account with a checklist of pass/fail line items streaming in, monospace terminal-style output visible in a code block within the chat]

**Monday: the audit**

Every account gets a 223-point check across 17 tiers before the week starts. Wasted spend by campaign type, broken conversion tracking, budget caps throttling a winning campaign without anyone noticing, negative keyword gaps, Performance Max cannibalizing branded search. All of it scored on conversions and ROAS by conversion time, never click time. Click-time numbers lie to you about what happened.

This used to be a Monday I dreaded. Six people, six spreadsheets, and half the findings buried in someone's notes by Wednesday. Now the audit runs before I'm at my desk and flags exactly which accounts need me, not which accounts need a rubber stamp.

**Tuesday and Wednesday: pacing and creative**

I check budget pacing against a simple band. An account drifting more than a set threshold off its monthly pace gets flagged, whether it's overspending toward month-end or sitting on unspent budget. Creative testing runs on a fixed cadence, a handful of new variations per account per week, and I score every variation on ROAS, not click-through rate. CTR looks great and pays for nothing.

One real example: an 8-figure supplement brand came in running about $180k a month through Performance Max, sitting at a 3.1x ROAS the account owner thought was fine. The audit flagged that a big chunk of that "PMax performance" was branded search demand the brand already owned organically, relabeled as a PMax win. We reallocated the branded spend out, redirected the freed budget into prospecting, and the account settled at 4.4x within a few weeks on the same total spend. Nobody worked harder. The audit refused to let a vanity number hide the real one.

[IMAGE-2: Hyper-realistic screenshot of the Google Ads web UI, real Google Ads logo top-left, left navigation showing Campaigns, Ad groups, Performance Max, Recommendations, main panel showing a Performance Max campaign overview with a conversion value chart trending upward, a small callout box highlighting "branded search overlap" in a warning-yellow color, account name and customer ID visibly blurred out with a gaussian blur bar]

**Thursday: reporting**

The system generates a report for every account, in the client's own format, built off the week's real numbers, not a template with the blanks filled in. I still read every one before it goes out. That's not optional. A model can write a clean paragraph about a number it doesn't understand the context behind. I know why an account dipped on a Tuesday because a client paused their own site for maintenance. The report needs that context or it reads as either alarmist or dishonest.

**What stays human**

The skill runs the checklist. It does not decide the strategy. Which accounts get my direct attention this week, when to override a flag because I know something the data doesn't, what to say to a client whose account had a rough month, whether to push a client toward a bigger swing or hold steady through a slow season. I don't automate any of that, and I'm not trying to. The system's entire job is buying back the hours I used to spend on checklists so I can spend them on the decisions that move an account.

[IMAGE-3: Dense hyper-realistic system architecture diagram on a dark navy canvas, a single hub card at the top labeled "Operator (Stephan)", curved color-coded connector lines branching into four vertical columns labeled Audits, Budget Pacing, Creative Testing, Reporting, each column stacking four to five small step cards naming exact tools (Claude Skills, Google Ads API, Supabase, scheduled scripts), clean sans-serif labels, icon badges at the top of each column, off-white text on dark background]

**Friday**

Friday is exceptions and real conversations. I give every account the week's system flagged a real look, and every client who needs to hear something gets a call, not an auto-generated email. That's the whole point of building the system in the first place: make sure the operator spends time on the parts of the job that need a person, not on the checklist.

Next edition, I'm breaking down exactly how the audit catches wasted Performance Max spend line by line, the checks most agencies skip entirely. If there's a piece of this system you want opened up first, reply and tell me. I read every one.

The part that makes 100 accounts possible

The honest answer is not a tool. It is that every account has the same shape, and anything that is not the same shape gets made the same shape before it joins the system.

One brief file per account, in the same format every time: margin per order, the brand terms list, conversion actions and which is primary, the things nobody is allowed to touch, and the client's actual goal in one sentence. Without that file, automation is a very confident stranger looking at a dashboard.

One naming convention across every account. Not because it is tidy, but because every downstream check depends on being able to find things by pattern rather than by memory.

One reporting definition. Conversion time, margin-aware, same date logic everywhere. The moment two accounts define revenue differently you cannot compare them, and comparison is the whole point of running many.

What runs nightly, exactly

Conversion health first, because everything downstream is worthless if this is broken. Any action that stopped recording, any duplicate primary, any import that changed.

Spend anomalies second. Three times daily budget with zero conversions in 48 hours, flagged with the campaign name, the spend and the last conversion date.

Feed and disapprovals third. Percentage of catalogue out of the auction, grouped by reason code so the fix is one action rather than two hundred.

Search terms fourth, on a rolling 90 days, sorted by cost with zero conversions.

Then the draft. Not a summary of what happened. A list of what changed, what it costs, and what it recommends doing about it.

Why this scales when headcount does not

A person doing this well takes about forty minutes per account per week and does it inconsistently by month three, not through laziness but because attention is finite and clients are loud. The system does it at the same depth on account 100 as on account 1, at four in the morning, whether or not anyone feels like it.

The output is not speed. The output is that the floor comes up. My worst week now looks like my average week used to.

The honest limit

This does not scale forever and I would not claim it does. Somewhere past a certain account count the constraint stops being hours and becomes judgement: the number of genuinely novel situations arriving each week that no rule covers. That number grows with clients and it is the only thing I have found that does not automate. Which is the real answer to how one person runs a hundred accounts. You do not remove the work. You remove everything that was never judgement in the first place, and then you find out how much judgement you actually have.