Back to log
🐙

I Gave an Ads Manager 37 Python Scripts. She Outperformed the Engineer.

case-studyai-agentsbuild-log

I built 37 Python scripts to automate Google Ads audits. Placement analysis, exclusion management, report generation, the whole pipeline. Months of work.

Then I had to hand them off to Nicola.

Nicola is an ads manager. Not a developer. She'd never opened a terminal before, never run a Python script, never used Git. And I needed her to run this system independently while I moved on to other projects.

I was worried it would be intimidating. CLI tools, git branches, API credentials. This is the stuff that makes non-technical people feel like they're going to break something.

What actually happened

She ran 3 scripts in her first 35 minutes.

Not guided. Not hand-held. She opened Warp (an AI-native terminal), checked out her isolated workspace branch, and started running reports. Within that first session, she'd generated a full weekly performance report across multiple client accounts and a detailed account-level audit.

By the end of February, she'd generated 27 reports. I had generated 14. She was outproducing me on the system I built.

She found something I missed

One of her first reports flagged a critical issue: a client account (Account B) was showing zero conversions. The campaigns were running, money was being spent, but conversion tracking was broken.

I hadn't caught it because I was focused on placement audits, not conversion monitoring. She caught it because she was actually looking at the numbers without assumptions about what "should" be working.

That's not a script running. That's an operator thinking.

The setup was simpler than I expected

The onboarding for Nicola was four steps:

cd ~/projects/agency/google-ads
git checkout operator-workspace
git pull
claude

That's it. She works on an isolated git branch so she can't accidentally touch production. Read-only scripts run freely. Anything that writes to a client account requires --dry-run first and explicit confirmation.

The 3-hour training covered:

  • Hour 1: What the terminal is, what the API does, why this saves time
  • Hour 2: Running the actual report scripts, reading the CSV outputs
  • Hour 3: Supervised exclusion deployment (with --dry-run safety net)

The safety rules were simple: always dry-run first, never edit production without approval, commit everything to GitHub, stop if something breaks.

What surprised me

I expected the technical barrier to be the hard part. It wasn't. The hard part was my own assumption that non-technical people would find this scary.

Nicola didn't find it scary. She found it empowering. The terminal wasn't a wall. It was a door. She could pull performance data across 17 accounts in 2 minutes instead of 30. She could generate audit reports that used to take a full morning. She could see patterns across accounts that are invisible when you're clicking through the Google Ads UI one account at a time.

She didn't just learn to use the scripts. She started creating her own. 20 new scripts on her branch, adapted for her workflow, generating reports for client proposals and performance tracking.

The uncomfortable realization

I spent months building an automation system. I thought the value was the code. The 37 scripts, the API integrations, the placement categorization engine.

The actual value was making an ads manager dangerous. Someone who already understands the business, already knows what good performance looks like, already has relationships with the clients. She didn't need to learn Python. She needed a faster way to do what she already knew how to do.

The scripts were the tool. She was the weapon.

What this means for teams

If you're building AI tools or automation systems and your plan is "the engineer will run it," you're leaving value on the table. The people closest to the work will always see things the builder misses.

The setup:

  • Isolated branches per operator (can't break production)
  • Read-only scripts run freely, write scripts need approval
  • AI terminal (Warp + Claude) so they can ask questions as they go
  • CSV outputs they can open in Excel if they want familiar territory
  • 3 hours of training, then they're independent

Nicola is now the primary operator of the Google Ads automation system. I built it. She runs it better than I do.

The best automation systems aren't the ones with the most code. They're the ones where the right person is using them.

Share