Claude Code for Growth Marketing: 5 Systems That Run Themselves

By Frederik Baun, conversion copywriter and founder of Words Sounds Visuals

I run five of these systems in my own agency today, each with a real number attached to it, not a hypothetical.

Claude Code for growth marketing, in my agency, runs as five separate systems. Each one does a single job. Each one has a real number attached to it.

I run a cold outreach engine, this blog, a lead database, a thumbnail pipeline, and the scheduler that keeps the other four running on time. Every one of them started as a single task I got tired of doing by hand.

I wrote up the three-layer architecture behind all five in how I use Claude Code for marketing. Directives say what to do, orchestration decides how, and code does the work the same way every time.

Growth marketing lives or dies on volume and follow-through more than on any single clever idea. These five systems are what let me run that volume without adding headcount.

What Claude Code for Growth Marketing Looks Like as Five Systems

A system, in this list, runs on a schedule or a trigger without me starting it by hand. A one-off prompt doesn’t qualify, no matter how good the output is.

Each system below has a directive that says what it does, a script that does the work, and a number that comes from watching it run.

The Cold Outreach Engine Running 44 Inboxes

My cold outreach system sends from 44 inboxes spread across 22 domains. That keeps the volume on any one domain low enough to protect deliverability.

A shared suppression list sits underneath all 44 inboxes. A prospect who unsubscribes or replies once stops getting emailed from any of them.

The system also writes a daily report on send volume and reply status. That’s how I see a launch’s health without opening a single inbox.

I read that report each morning. I decide whether a sequence needs a change, then let the rest of the day run on its own.

Each launch draws from its own subset of those 44 inboxes, so one campaign’s send pattern stays separate from another campaign’s warmup. I rotate which inboxes carry which launch, and the system tracks that rotation so I don’t have to hold it in my head.

The Blog That Drafts, Gates, and Publishes Itself

This blog runs on a content calendar, a writer directive, and a publish gate. A scheduled job handles each step without me starting it by hand.

That job drafts the next week’s planned posts every Monday morning, before I’ve opened my laptop.

Every draft has to pass a gate that runs a copy check for banned writing patterns and a fact check against live sources. The gate records a pass hash before anything can go live.

A second scheduled job checks each morning for a gated post whose publish date has arrived, then deploys it.

The Monday job stops after drafting and gating. The title, the featured image, and the deploy decision stay mine until I’ve read the draft myself.

The Lead Database That Tracks Who’s Already Been Contacted

One database holds every lead across every source I use: Upwork, cold email positive replies, and website opt-ins. A suppression list and a verified-email cache sit on top of it as views, so the database stays the single source of truth.

Before any new outreach list goes out, the system checks it against that one database first. That single check is why the same person doesn’t get a cold email from me twice, months apart, from two different campaigns.

Those three channels, Upwork, cold email positive replies, and website opt-ins, are the only ways a contact enters the database. That rule keeps a stray forwarded email in my inbox from turning into a phantom entry in the database.

The Thumbnail Stack That Costs About $0.80 a Video

My YouTube thumbnail system runs three tiers of image models: a primary tier, a fallback, and a specialist third option for harder concepts. The primary tier costs me around $0.20 per image, so a four-image pack for one video comes to about $0.80.

I feed the system two reference photos chosen for the emotion the concept calls for, matched against the mood of the video itself. That one choice, matching the reference photo to the concept’s emotion, moved the finished thumbnail more than which model generated it.

The primary tier handles most concepts on its own. The fallback steps in when a face comes out wrong, and the third tier adds a trained style on top for the harder concepts.

I reach for that third tier only a few times a month, on the concepts where the first two tiers miss the reference photo’s expression.

The Scheduler Holding the Other Four Systems to a Clock

Every scheduled job across these five systems runs through the same mechanism: a LaunchAgent logged into my Mac’s session so it can reach the credentials the jobs need.

Three of those jobs run the blog: one drafts on Monday morning, one checks for publish-ready posts every day, and one pulls ranking data into a spreadsheet each week.

The outreach engine and the lead database run on their own jobs underneath the same scheduler. Building the scheduler once means every one of those jobs runs on time by itself, so remembering to start them stopped being my job.

These jobs run on LaunchAgents instead of cron, because I hit the same silent failure enough times to stop trusting cron with anything that touches credentials. A job that worked fine under a LaunchAgent would fail without warning under cron.

How to Use Claude Code for Marketing When You’re Starting From Zero

Start with one task you already do the same way every time, written down as plain-English instructions. You don’t need five systems on day one.

Pick the task costing you the most hours this month. For me, that used to be checking a cold email suppression list by hand before every send. It’s now the first check the outreach system runs on its own.

Write down what the task requires. Show Claude Code the shape of the job once, and let it turn those steps into a script. The instructions stay the plan, and the script becomes the part that runs itself.

What This Means for Your Growth Marketing

Strategy calls, client relationships, and judgment calls about which prospect to prioritize still take a person, the same way they did before any of this existed.

What five running systems replace is the busywork sitting between a decision and a result: checking a suppression list by hand, remembering to publish on a certain day, pricing out a thumbnail one image at a time.

Five systems running today, each solving one repeated task, are still just five scripts underneath.

The part that doesn’t automate is knowing your own business well enough to see where the busywork lives. I keep the tools I’ve built along the way at the tools hub.

The blog system above runs on the same skill files I released publicly in the five marketing skills for Claude Code, so you don’t need my exact stack to start writing better-briefed drafts today.