If you want to automate content production with AI, you don’t need a tool right away—you need a precise understanding of your own process. That’s because AI amplifies what’s already there: Good processes become faster, while poor processes produce errors more quickly. We present a five-step, streamlined approach that enables companies to tackle this issue head-on rather than putting it off. You’ll also learn in which areas AI is already creating tangible value—and in which areas humans remain firmly in control.
When scaling AI use cases, attention to detail is key
Marketing and communications are among the areas where companies most frequently use AI—according to Bitkom, 57 percent of AI-using companies in Germany employ the technology in these areas. But there’s a gap between experimentation and production: Camunda’s State of Agentic Orchestration and Automation Report 2026 shows that while 66 percent of the companies surveyed use AI agents, only 11 percent of use cases actually go into production. Fifty-eight percent of respondents cite concerns that uncontrolled agents will exacerbate problems in poorly implemented processes as the main reason.
This is precisely where the flaw in the thinking behind many automation projects lies: they start with the tool and end with the process. Successful projects do it the other way around.
The target image comes before the model
Every automation project begins with a seemingly trivial question: What goes into the process—and what should come out of it? In the case of a content migration, for example, this means: What basic information is stored in the old system, for what purpose was it maintained, and in what format should it be imported into the new system? As long as this target vision remains unclear, any AI will only generate confusion—just faster.
As obvious as it sounds, this question is rarely asked in practice. We often see teams jumping straight into discussions about prompts and models before it’s even clear what would constitute a good result.
Five Steps from Process Understanding to AI Capability
In our projects, a five-step approach has proven effective for making the process fully transparent before deciding on automation.
First, we define the target state: the starting point and end point of the process, including basic information, intent, and output format. Building on this, we break down the phases—the path from the current state to the target state is divided into clearly defined process steps. For each phase, we clarify responsibilities: Who does what? This often reveals just how much implicit knowledge resides in individual minds. Next, we identify the tools and sources—which systems, data, and documents feed into each phase? Finally, we establish input/output gates at the transitions between phases: These define the quality standards an intermediate result must meet in order to be passed on.
Each of these five building blocks is feasible on its own—together, they form the complete picture of the process.
Only on this basis can we seriously determine where AI’s potential lies—and where human involvement remains essential. The “gates” from the previous step serve as natural checkpoints: They mark the points at which a human reviews, approves, or corrects the process before it continues.
Not every phase needs to be automated
The analysis consistently yields a result that contradicts the hype: Not every phase of a content process is suitable for automation. Repetitive steps with clear rules—such as converting structured product data into a target format—are ideal candidates. Phases that require brand judgment, strategic consideration, or legal evaluation remain tasks for humans. Those who have a clear picture of the process can make this distinction with confidence, rather than leaving it to the tool provider or the LLM. AI can certainly help narrow down the decision-making scope, but human judgment and context provide and ensure the value within the process.
Inspire, Create, Scale: Where AI Adds Value in the Content Lifecycle
A helpful distinction adds a horizontal dimension to the five-step process: At what stage of the content lifecycle is the process actually taking place? We distinguish between three streams, each with its own level of automation and a different role for AI.
Inspire encompasses strategy, conceptualization, and ideation—such as content strategies, content design, and product taxonomies. Here, AI serves as a sparring partner: It provides options, research, and structural suggestions, but the final decision rests with humans.
Create refers to the production of individual assets through defined phases, such as landing page creation. Here, humans and AI work together in a co-creation process: the AI produces the content, while humans review it and approve it at each stage.
Scale describes scaling at high volumes—variants, channel derivations, or content migrations via automated pipelines. This is where the level of automation is highest; humans are involved almost exclusively at the input/output gates.
In practice, these three streams blur together faster than you might think: If a team thoughtlessly applies the same automation standards—which make sense in scaling—to an ideation workshop, the collaboration between humans and AI will be calibrated in the wrong place. The practical benefit of this distinction lies precisely here: It helps set the right expectations for each stream, rather than approaching an “Inspire” task with “Scale” speed or leaving valuable scaling potential untapped in “Create” mode.
Real-World Examples: Landing Page Creation and Content Migration
We’ve applied this approach at, among others, a leading manufacturer of power tools and an international technology conglomerate—in two streams that could hardly be more different.
The Create stream focused on traditional landing page creation: Starting with a briefing, a page ready for publication is developed through defined phases, with AI handling text generation and structural suggestions, while approvals at the gates remain in human hands.
The Scale stream involved a content migration: product information from an existing PIM (Product Information Management) system needed to be transferred to a new system—including adaptation to a new taxonomy. It was precisely here that the value of understanding the process became apparent: the real challenge lay not in translating the content, but in clarifying the original intent behind the content’s creation and the quality standards it must meet in the target system.
Both projects began with the same fundamental question: What should the end result be, and what steps lead to it? This clarity from the outset ensured that automation was applied exactly where it was needed in both cases.
Conclusion: Five steps, one complete picture
Anyone who tries to solve everything at once when it comes to content automation usually gets bogged down in theoretical debates about tools and models. Those who instead follow five concise steps—vision, phases, responsibilities, sources, and gates—gain immediate clarity with each step and ultimately keep all options open: the Landing Page Creator for content production, low-code pipelines for recurring workflows, or a systematic content migration for a system change. Those who skip this process end up choosing tools for a problem they haven’t yet understood.
Would you like to know which phases of your content production are suitable for AI support and where you have the greatest leverage? Contact us: We bring the methodology and experience from real-world projects to the table.