Last October, a friend who runs a 40-person marketing agency visited my home office in Graz. He watched me work for about two hours. At the end, he sat back and said, “You’re doing what my content team of eight does. How is that possible?”
The answer isn’t talent. It’s systems. Specifically, AI workflows designed to replicate the function of a department—not by replacing every person in it, but by replacing the coordination overhead, the repetitive production work, and the information processing that eats most of a team’s time.
Let me show you exactly how these work.
Why Departments Exist (And Why That Reason Is Dissolving)
Departments exist for three reasons: specialization, coordination, and throughput. You need a marketing department because marketing requires specialized skills, someone needs to coordinate the work, and one person can’t produce enough volume alone.
AI attacks all three reasons simultaneously. It provides on-demand specialized capability (not expertise—capability). It eliminates most coordination overhead because there’s nothing to coordinate when one person directs the entire process. And it multiplies throughput by handling the production layer.
The result isn’t that departments become unnecessary in all organizations. Large companies with complex operations still need teams. But for companies under 20 people—and especially for solo operators—the department model is increasingly an artifact of pre-AI constraints.
I realized this when I was building my AI content agency. The traditional model would have required writers, editors, project managers, and quality controllers. Instead, I built workflows that consolidated those roles into a single operator position: mine.
The key insight is that a “department” is really a set of interconnected processes. If you can map those processes and identify which parts require human judgment versus human labor, you can rebuild the department as a workflow.
The Anatomy of a Department-Replacing Workflow
Every workflow I’ve built follows the same five-layer structure:
Layer 1: Input Processing. Raw information comes in—customer feedback, market data, content briefs, financial figures. The AI categorizes, cleans, and structures this input. This replaces what a junior team member would typically do: sorting through the noise to find the signal.
Layer 2: Analysis and Synthesis. The structured input gets analyzed. Patterns identified, comparisons drawn, anomalies flagged. This replaces mid-level analytical work—the kind of thing a senior analyst or experienced team member would do, minus the contextual judgment.
Layer 3: Draft Production. Based on the analysis, the AI produces draft outputs—reports, content pieces, presentations, recommendations. This replaces the production work that takes up most of a department’s time.
Layer 4: Human Review and Direction. This is where I come in. I review everything from layers 1-3, apply judgment, add context, make decisions, and redirect as needed. This is the part that can’t be automated—and it’s the part that actually creates value.
Layer 5: Output and Distribution. Final outputs are formatted, scheduled, and distributed. The AI handles the mechanics; I’ve already made the strategic decisions in Layer 4.
When I describe this as AI making things possible rather than faster, this is what I mean. It’s not that each layer is faster. It’s that one person can operate all five layers across multiple functions simultaneously.
Case Study: My Content Department
Let me walk through one workflow in detail. My content operation produces articles, social posts, newsletter content, and book-related material across two languages.
The old way (pre-AI): I’d spend Monday planning the week’s content. Tuesday through Thursday writing. Friday editing and scheduling. Output: maybe 3-4 pieces per week if I did nothing else. Quality was good because I wrote everything personally, but volume was severely limited.
The workflow way:
Input Processing: Every Sunday evening, my system pulls together content signals—trending topics in my space, audience questions from the community, gaps in my content library, upcoming dates or events that need coverage. This takes about 15 minutes of my review time.
Analysis: The AI cross-references these signals against my content strategy, identifies which topics serve which goals (audience growth, authority building, product promotion, community engagement), and suggests a prioritized content calendar. I review and adjust in about 20 minutes.
Draft Production: For each prioritized piece, I write a brief—thesis, key points, target audience, desired action. The AI produces first drafts. This happens overnight or while I’m doing other work.
Review and Direction: I spend 60-90 minutes each morning reviewing and editing drafts. This is where I add my voice, cut the generic parts, insert specific examples from my experience, and make sure nothing reads like committee-produced content. Some pieces I rewrite substantially. Others need light editing. The point is I’m editing, not creating from zero.
Distribution: Final pieces are formatted for each channel and scheduled. The mechanical work is handled; I’ve made all the editorial decisions already.
Output: 12-15 pieces per week across channels, in two languages. Not by working more, but by spending my time on the parts that actually need me.
Case Study: My Research Department
The research workflow was the one that surprised me most. I used to think research was inherently human work—and the judgment part still is. But the collection, organization, and initial synthesis? That’s production work wearing an intellectual disguise.
When I was developing the Subtract to Ship methodology, I needed to process research from multiple fields: behavioral economics, organizational psychology, systems thinking, and startup case studies. The traditional approach would have been a research team spending months on literature review.
Instead, I built a research workflow that processed sources in batches, extracted relevant findings, organized them by theme, identified contradictions and gaps, and produced synthesis documents I could work from. My job became what it should have been all along: thinking about what the research means, not hunting for it.
This same workflow now serves my consulting practice. When a client needs competitive analysis, market research, or trend assessment, the workflow handles 80% of the production work. I add the interpretation, the strategic framing, and the recommendations that require knowing the client’s specific context.
I discussed a version of this in my piece on how I built six books using AI-native methods. The book production was essentially a massive research workflow feeding into a content workflow, with my editorial judgment as the connective tissue.
The Three Workflows Worth Building First
If you’re starting from zero, here’s where I’d focus:
1. Content Production Workflow. This has the highest immediate ROI because content feeds everything—marketing, sales, authority building, community engagement. Build a workflow that handles research, drafting, and formatting while you handle strategy, voice, and editorial decisions.
2. Customer Intelligence Workflow. Collect and synthesize customer feedback, support tickets, social mentions, and market signals. Most businesses drown in customer data and starve for customer insight. A good workflow turns the data into insight automatically, leaving you to make decisions based on what it surfaces.
3. Administrative Operations Workflow. Invoicing, email management, scheduling, document preparation, compliance tracking. None of this requires creativity, but all of it steals time from creative work. Automate it ruthlessly.
Notice what’s not on this list: sales, strategic planning, relationship building, product development decisions. These are judgment-heavy activities where AI assists but doesn’t drive. Don’t try to automate them into workflows—you’ll get worse results than doing them manually with AI assistance.
What Goes Wrong (And How to Prevent It)
I’ve made every mistake here, so let me save you the trouble.
Mistake 1: Building before mapping. I jumped into building AI workflows before I’d properly documented my existing processes. Result: I automated a bunch of things I should have eliminated entirely. The subtraction audit should come before any automation work. Kill unnecessary processes first, then automate the necessary ones.
Mistake 2: Skipping the review layer. Early on, I let some AI outputs go to customers without proper review. The quality wasn’t terrible, but it wasn’t me. My audience noticed. Now Layer 4 is non-negotiable, even when I’m busy. If I don’t have time to review, the output doesn’t ship.
Mistake 3: Over-engineering. My first content workflow had 14 steps, three review stages, and a quality scoring system. It was a masterpiece of process design that took more time to operate than just writing the content myself. I stripped it down to five layers and it’s been working perfectly since. Complexity in workflows is a bug, not a feature.
Mistake 4: Ignoring context decay. AI workflows work great when they have fresh context—your brand voice, current strategy, recent examples. They degrade when that context gets stale. I update my workflow context monthly: new examples, adjusted tone guidelines, current strategic priorities. Skip this and your outputs slowly drift toward generic.
Mistake 5: Forgetting the human bottleneck. The workflow can produce infinite output. You can’t review infinite output. Design your workflows to match your review capacity, not the AI’s production capacity. I limit my content workflow to 15 pieces per week because that’s what I can meaningfully review. I could produce 50, but 35 of them would be unreviewed garbage.
The Economics
Let me put numbers on this, because the economic argument is what usually convinces people.
A content marketing team of four (writer, editor, strategist, coordinator) costs roughly €180,000-€250,000 per year in Austria including Lohnnebenkosten. A research team of two costs roughly €90,000-€130,000. An administrative assistant costs €35,000-€50,000.
My AI tooling costs roughly €300-€500 per month. Call it €6,000 per year. Even if you add €20,000 for my time spent operating the workflows (which is generous), the total is under €30,000 for output that would have cost €300,000+ in traditional staffing.
That’s not a marginal improvement. It’s a structural advantage. And it’s available to any solo founder or small team willing to invest the time in building proper workflows.
The catch is that building time. It took me roughly three months of experimentation to get my core workflows running reliably. During that period, my output actually dropped because I was building systems instead of producing. The velocity principle applies here—move fast through the building phase and start iterating on real output as soon as possible.
Takeaways
- Departments are really sets of interconnected processes—map the processes and identify which parts need human judgment versus human labor.
- Every effective AI workflow has five layers: input processing, analysis, draft production, human review, and output distribution.
- Start with content production, customer intelligence, and administrative operations—these three deliver the highest immediate ROI.
- Always subtract before you automate—kill unnecessary processes first, then build workflows for what remains.
- Design workflows to match your review capacity, not the AI’s production capacity—unreviewed output is worse than no output.
