How We Use Claude Code to Run a Marketing Agency
A behind-the-scenes case study of how EBCD uses Claude Code to automate client reporting, prospect scraping, video generation, and more. Real tools, real numbers.
We run a marketing agency with 13 active clients, a 3-person core team, and zero full-time developers. Our secret weapon costs $20/month and runs from the terminal. It's called Claude Code, and it has fundamentally changed how we operate.
This isn't a review. It's a behind-the-scenes case study of exactly how we use Claude Code at EBCD — what we've built, what it replaced, and what still needs a human.
The Workspace: An Agency Brain in a Folder
Everything starts with a file called CLAUDE.md. Think of it as the agency's operating manual that Claude Code reads at the start of every session. It contains our workspace structure, every command we've built, API configurations, video style rules, client information, and workflow documentation.
When I start a new session, I run /prime. Claude Code reads the context files, understands who I am, what clients we serve, and what tools are available. No re-explaining. No context loss between sessions. It picks up where we left off.
This workspace approach means Claude Code isn't just answering questions — it's operating inside our business systems with full awareness of our processes.
What We've Actually Built
Client Reporting: 4 Hours to 45 Minutes
Every week, each client gets a performance report. Before Claude Code, that meant logging into Meta Ads Manager for each account, pulling screenshots, copying numbers into a template, writing commentary, and formatting the deliverable. Four hours minimum for 13 clients.
Now I run /report [client-slug] --type weekly and get a formatted report with spend, leads, CPL, trend analysis, and actionable recommendations. The tool pulls data from Meta's API, crunches the numbers, and outputs a client-ready document. I review, adjust commentary where needed, and send. Total time per client: under 4 minutes.
Prospect Scraping: Manual Research to CRM-Ready Leads
Finding potential clients used to mean hours on Google Maps, manually checking business details, copying information into spreadsheets, and formatting for CRM import.
The /scrape-prospects command scrapes Google Maps for ICP-matched businesses in any metro area and outputs CRM-ready CSVs. 200+ qualified prospects per location in under 10 minutes. The system routes contacts to the correct GHL account automatically — agency prospects go to our main CRM, referral partners go to our Alkalino sub-account.
Video Ad Generation: From Concept to Rendered Video
Our video pipeline generates ad creatives from script to final render. We define beats (scene segments), the tool generates voiceover scripts, pulls b-roll from our custom library and Pexels, and renders through Remotion. One command produces multiple video variations ready for split testing.
We generate territory-specific videos, ad hook tests, and organic content — all following strict style rules (no on-screen text, no captions, 5 clips per beat, cuts every 3 seconds) that are enforced by the CLAUDE.md configuration.
Dashboard: Real-Time KPIs for Every Client
Claude Code built our entire client dashboard — a Flask + Chart.js application with role-based authentication. Clients log in and see their own metrics. Our team sees all clients with comparative KPIs. It runs on Railway for $5/month and includes standup alert automation that flags underperforming campaigns.
The /report command and the dashboard sometimes pull data through different paths, which means the occasional discrepancy. That's a real limitation — not everything is perfectly unified. But having both tools beats having neither.
Blog and Content Generation
The /generate-blog-post command creates SEO-optimized blog posts following our exact style guide, keyword placement rules, and internal linking strategy. Posts are staged in a drafts folder for human review before publishing. Every post on ebcd.io goes through this pipeline.
Landing pages for territory-specific campaigns get generated via /landing-page — Spanish-first, mobile-optimized, with GHL form integration already wired up. What used to take a day of design and development takes 20 minutes.
Email Sequence Automation
We built a 6-email Spanish-language nurture sequence for our water treatment leads. Claude Code wrote the copy, structured the drip logic, and generated the API calls to create the sequence in GHL. The trigger fires automatically when a contact gets tagged.
Competitor Intelligence
Automated scraping of competitor Instagram profiles and Meta Ad Library entries. We monitor creative rotations, messaging shifts, and spend patterns across competitors in each client's market. What used to be a quarterly manual exercise is now a weekly automated report.
What AI Coding Tools Can't Replace
For all the AI coding tools available to marketing agencies, there are clear limits:
- Strategy calls with clients. AI can prepare the data and talking points, but the relationship and judgment calls are human.
- Creative direction. The tool generates variations, but deciding which hook resonates for a specific market requires experience and intuition.
- Handling exceptions. When a client's Meta account gets flagged, or an API changes its authentication flow, or a campaign needs emergency pausing — that's still manual.
- Quality control. Every output gets reviewed. Reports get proofread. Blog posts get edited. Video scripts get watched. The AI is the first draft machine. Humans are the quality gate.
The Numbers
| Metric | Before Claude Code | After Claude Code | |---|---|---| | Weekly reporting time | 4+ hours | 45 minutes | | Prospect research per metro | 3-4 hours | 10 minutes | | Blog post creation | 3 hours | 30 minutes (including review) | | Landing page build | 1 full day | 20 minutes | | Clients managed per person | 4-5 | 8-10 |
The tool costs $20/month. It runs from the terminal. There's no enterprise sales process, no per-seat licensing, no implementation timeline. You install it and start building.
Why This Matters for Other Agencies
If you're running a small agency and thinking about AI marketing tools, start with the operational bottleneck — the repetitive work that keeps you from scaling. For most agencies, that's reporting and data management. Build one automation, see the time savings, then expand.
The agencies that figure this out in 2026 will operate at 3-5x the efficiency of those that don't. That's not a prediction — it's what we're already living.
Frequently Asked Questions
What is Claude Code and how much does it cost?
Claude Code is Anthropic's command-line AI coding assistant that runs in your terminal. It costs $20/month as part of Claude Pro, with no per-seat licensing or enterprise sales process. It reads your project files, understands your codebase, and can execute commands, write code, generate reports, and automate workflows. Unlike ChatGPT or other chat-based tools, it operates directly in your file system with persistent context across sessions.
Can a non-technical person use Claude Code for their agency?
You need basic comfort with the terminal (command line), but you don't need to be a developer. The learning curve is roughly 1-2 weeks for someone who has never used a terminal before. The key is building a CLAUDE.md file that teaches Claude Code your specific workflows, then using slash commands that abstract away the complexity. Once configured, running a report is as simple as typing /report client-name --type weekly.
What marketing agency tasks can AI automate in 2026?
Based on our experience, the highest-ROI automations for marketing agencies are: client reporting (4+ hours to 45 minutes), prospect research (3-4 hours to 10 minutes per metro), blog content creation (3 hours to 30 minutes including review), landing page builds (1 day to 20 minutes), and competitor monitoring (quarterly manual task to weekly automated report). Strategy, creative direction, client relationships, and quality control still require humans.
Is Claude Code better than ChatGPT for marketing agencies?
They serve different purposes. ChatGPT excels at one-off content generation and brainstorming. Claude Code excels at system-level automation — it reads your entire workspace, remembers your processes, and executes multi-step workflows. For agencies that need repeatable, consistent outputs (reports, landing pages, blog posts following specific templates), Claude Code's workspace approach is significantly more efficient than pasting prompts into ChatGPT each time.
Want to see the actual workspace in action or talk through how to set up something similar for your agency? Book a 30-minute call — happy to walk through what we've built and what translates to your setup.