Can Claude Code Automate Ad Operations? Results of Our Experiment
Hello, this is the AD x AI Lab team at Ad Innovation.
Entering 2026, the evolution of AI agents has completely shifted from the "consulting via chat" phase to the "autonomously completing tasks" phase. At the center of this is "Claude Code," a terminal-resident agent provided by Anthropic.
While this tool has already become an indispensable partner for engineers, just how practical is it in the "field of ad operations"?
We conducted a thorough verification to see how much of our actual operational workflow we could automate. To put it briefly, it was a shock that rewrites the definition of an operator.
What is "Claude Code" in the first place?
What makes it decisively different from traditional ChatGPT or Claude (Web version) is that Claude Code "autonomously performs file operations on the PC, executes terminal commands, and integrates with APIs."
As of the latest version in 2026, the following features are particularly powerful:
MCP (Model Context Protocol) Support: Capable of direct dialogue with Google Ads and Meta Ads APIs.
Agent Teams Feature: Multiple AIs, such as "analysis specialists" and "creative generation specialists," form a team to perform parallel tasks.
Claude Ads Skill: The emergence of a specialized skill that completes over 190 ad audit items in just a few minutes.
Using these, we tested the following three scenarios.
Verification 1: Cross-media report creation and "anomaly detection"
Until now, we spent 30 minutes to an hour every morning going back and forth between Google, Meta, TikTok, and Twitter (X) management screens, transcribing data into spreadsheets, and checking day-over-day performance.
What we did
We gave Claude Code a single instruction: "Fetch yesterday's performance for all media via API, identify campaigns where the CPA has worsened by 20% or more, and send them to Slack with a factor analysis."
Results
Time required: Approximately 2 minutes
Accuracy: Zero transcription errors, as it references raw data directly via MCP.
Surprising point: It didn't just report numbers; it returned operator-level insights as a set, such as, "The CPA deterioration in this campaign is caused by a drop in CTR for a specific creative. I recommend adjusting the bidding strategy or replacing the creative."
Verification 2: "Mass generation to submission" of responsive ads
The most time-consuming parts of ad operations are creating creative variations and the submission process. We conducted a verification based on the workflow practiced by Anthropic itself.
What we did
We integrated with the Figma API to break down the elements of existing high-performing creatives. We instructed Claude Code: 'Create 50 banner variations reflecting the features of the new product and complete the layout in Figma.'
Results
Time required: 30 minutes → 30 seconds (for variation generation only)
Practicality: Tasks that previously required manual copy-pasting were completed in a single batch, including file generation. We were able to execute draft submissions directly via the Google Ads API.
Professional perspective: To avoid 'AI-like' expressions, we referenced a CLAUDE.md (configuration file) that trained the AI on our company's past tone and manner, successfully mitigating the risk of brand damage.
Verification 3: 190-item 'Account Audit'
We automated the 'comprehensive configuration check' previously performed by veteran operators.
What we did
We imported the latest 'Claude Ads' skill available on GitHub. claude ads audit --platform google
Results
Check items: 190 items, including missing negative keywords, broken URL links, duplicate match types, and budget imbalances.
Result: A PDF report was generated in just 5 minutes. This is a task that would take a human half a day to do manually.It was the moment we completely shifted from the job of 'finding mistakes' to the job of 'making decisions' based on AI reports.,
Conclusion: Can ad operations be automated?
After trying it out, our answer is 'YES, but only with 'co-creation' as a prerequisite.'
What has changed
Elimination of simple tasks: Report creation, routine submissions, and primary analysis no longer need to be done by humans.
Order-of-magnitude improvement in speed: The figure of 'analysis that took 3 hours now finishing in 7 minutes' is by no means an exaggeration in 2026.
The 'human role' that still remains
What became clear through this verification is that AI cannot function unless it is provided with a "purpose" and "context."
What is the client's true challenge?
Which metrics should be prioritized for improvement?
Where is the "line that must not be crossed" for the brand?
I am convinced that defining these elements and steering the powerful engine that is Claude Code is what the advertising operations professional of 2026 and beyond will look like.
Conclusion
At Ad Innovation, we are quick to incorporate such cutting-edge AI agents into our practical work, providing our clients with "fast, deep, and accurate" marketing support.
If you are looking to "reduce operational man-hours with AI" or "spend more time on essential strategy planning," please feel free to consult with us. Let's build a new form of advertising operations together, walking alongside AI.
A brief promotion
Ad Innovation is currently developing and operating our proprietary AI advertising system, "RakuAd AI". It is a platform that integrates a natural language interface via Slack, cross-media management, and automated reporting.
We are currently accepting pre-registrations and requests for advance demos ahead of the release. Why not be among the first to experience a new form of advertising operations that walks alongside AI?
If you are interested, please contact us from the link below.