Marketing operations are often seen as the glue between strategy and execution. But when the same tasks repeat across multiple tools—like creating landing pages, generating UTM links, drafting emails, and managing registrants—errors and inefficiencies creep in. In a real-world example, Tomoko Tanaka, a regional marketing lead for GitHub in Japan and Korea, turned her event workflow into a fully automated system using GitHub Copilot and GitHub Actions.
What Happened: From Manual Work to Automated Pipelines
Tanaka manages a diverse set of events: recurring webinars for enterprise developers, community meetups in Tokyo, and invite-only executive sessions in Seoul. Each event requires a detailed plan, precise targeting, and consistent follow-up. Historically, this process took days to complete manually—duplicating landing pages, generating UTM-tagged links, drafting emails, and managing project boards.
Each step carried risk: a typo in a campaign name, a missed day of registration updates, or a wrong link could ripple through downstream reports. To reduce these risks, Tanaka began documenting her workflow in plain Markdown, then fed it to GitHub Copilot. Instead of writing code, she wrote down how she did her work—what fields to fill, what labels to apply, and what steps to follow—and used Copilot to generate the actual automation scripts.
Key Facts: The Foundation of the System
- Events are managed as GitHub Issues, with structured forms for title, date, region, and campaign name.
- Labels act as triggers—for example, an event-setup label activates automated workflows.
- GitHub Actions workflows parse the issue body and execute tasks like duplicating landing pages, generating UTM links, and creating email documents.
- The system uses either an API or CLI to interact with external tools—such as an event platform or CRM—without requiring API keys.
- Every workflow change is treated as a pull request, enabling team review, visibility, and version control.
Crucially, the workflow is not rigid. A human review step ensures that campaign names, email subjects, and dates are approved before any automation proceeds. This balance between automation and human oversight prevents errors while allowing flexibility for one-off variations.
How It Works: A Step-by-Step Workflow
The process begins with a conversation in GitHub Copilot. Tanaka types a prompt like, “I want to run a webinar about AI-assisted development in November.” GitHub Copilot then consults a team-runbook file called AGENTS.md, which defines naming conventions, time zones, and email templates. Based on this, Copilot proposes a campaign name, drafts two email versions, and asks targeted questions.
Once the conversation concludes, Copilot creates a GitHub Issue with the correct labels. When the event-setup label is added, a GitHub Actions workflow triggers and performs the following steps:
- Duplicates a past event on the event platform to generate a new landing page.
- Generates a full set of UTM-tagged URLs—each tailored to a specific channel and consistently formatted.
- Creates a Word document version of the invitation email and commits it to the repository.
- Files a request issue with teams responsible for email distribution and tracking.
- Every morning, downloads the registrant list, cleans it, and posts a status update for stakeholders.
- After the event, exports attendees, reshapes the list for CRM upload, tags records, and generates a post-event report.
The entire process—from planning to post-event follow-up—runs on existing tools and infrastructure. No custom software was built. Instead, the team leveraged GitHub’s native features: issue forms, labels, and Actions—all of which are already familiar to developers.
Why It Matters: Automation Without Over-Engineering
Marketing automation platforms exist, but they often fail to adapt to regional or segment-specific variations. For example, a webinar might be delivered in Japanese one month and in Korean the next, with different CRM fields and definitions of a ‘good lead.’ Customizing such tools requires significant time, budget, and coordination.
By contrast, this approach treats workflow changes as pull requests—just like software development. A new event type or region becomes a code change that is reviewed, tested, and merged. This ensures transparency, consistency, and accountability.
Moreover, the use of a conversational interface—now available in the GitHub Copilot app—lowers the barrier to entry. Instead of requiring terminal access, users can now interact with the system through a regular desktop window. This makes automation accessible to non-technical team members.
Limitations and Open Questions
While effective, this system is not a universal solution. It assumes that all repetitive tasks can be broken down into structured inputs and outputs. It also depends on the availability of APIs or CLIs in the tools used—some marketing platforms may lack these capabilities.
Additionally, the success of this workflow relies on consistent documentation and team buy-in. Without a shared runbook, the system may become brittle or inconsistent. The model also does not address dynamic content—such as real-time sentiment analysis or personalized messaging—that requires deeper AI integration.
What to Watch Next
As AI agents become more embedded in workflow tools, expect more examples of “marketing ops as code” in diverse industries. GitHub’s own canvases feature may soon allow even non-developers to design custom workflows through natural language. Meanwhile, the growing use of AI in content generation and analytics could further reduce manual inputs in marketing operations.
For those interested in how AI is reshaping operational workflows, see how AI is accelerating science and improving lives or explore GitHub’s migration of Copilot to Rust for insights into AI infrastructure evolution.
Ultimately, this case shows that automation doesn’t require building new tools—it requires documenting existing processes and empowering AI to execute them with precision and consistency.
Sources & further reading
Featured image: Researchers working on the "R1 humanoid robot" project of the Center for Robotics and Intelligent Systems (CRIS) Humanoid Sensing and Perception lab by European Commission, CC BY 4.0, via Wikimedia Commons. Image source · License