40 vs 7 Hours for Analysis & Strategy (Without or With AI)

AI tools are already part of the design process. Not in every task, not in every project, and not as a replacement for familiar tools — but clearly enough that we can no longer treat them as a separate experiment. The real question is how to bring AI into the workflow so it helps, instead of adding another layer of noise.

At Tigre Rossa, we see AI as a working tool. It can speed up research, support early visual exploration, generate more directions at the start of a project, and reduce repetitive production work. But it does not make design decisions on its own. It does not understand a client’s context the way a team does, and it does not know when an image looks striking but fails to work as part of a system.

We decided to explore this through one of our own internal processes: designing a cover system for Tigre Rossa articles.

Why We Needed a System

We wanted to start publishing articles regularly on our own platform: about design, AI, visual communication, internal processes, and the way we work with complex topics. But regular content takes resources. Beyond the writing itself, there is the visual language, the publishing format, and adaptation for the website and social media.

At first, the task sounded simple: we needed covers for articles. But individual images would not solve the problem. If every article required a new illustration from scratch, the process would remain beautiful, but heavy. So we started designing not a set of images, but a system.

The covers had to work in two formats: horizontal for the website and vertical for Instagram and LinkedIn. They had to feel recognisable, connected to the character of the agency, flexible enough for different topics, and easy to refine further. We needed a visual language that felt alive, slightly ironic, and graphic: hand-drawn lines, subtle imperfection, white space, Tigre Rossa’s red and blue accents, and as little visual noise as possible.

From Categories to Visual Types

At first, we discussed whether the covers should be tied to content categories: one style for AI, another for case studies, another for backstage stories or analysis. But that logic quickly felt too rigid. A piece about AI can also be a case study, a process reflection, and a practical breakdown.

So we chose a different structure: not content categories, but types of visual solutions. This led to four directions.

The first

type was narrative illustrations with the Tigre Rossa mascot. In these covers, our Tigra becomes the main character: working with documents, holding coffee, sitting at a desk, or appearing inside a small visual scene.

The second

type was conceptual illustrations without a character. Here, the image is built through objects, symbols, and simple visual metaphors — for example, a mess of documents gradually turning into a clear structure.

The third

type combined photography with illustration layered on top. In theory, it could create a more editorial feeling. But this direction required photo selection, careful integration of graphics, more control over composition, and often extra manual work. It looked interesting, but did not optimise the process. So we removed it from the system.

The fourth

type was minimal graphite icons: one sign or a small set of symbols that communicates the theme of the article.

The final system included three working types: a narrative scene with the mascot, a conceptual metaphor, and minimal icons. We kept what answered the brief: faster production, consistent quality, and visual recognition.

The Mascot as a Test for the Tool

The mascot became a separate challenge. Tigre Rossa already has its own Tigre — with a specific head shape, proportions, expression, blue stripes, red colour, and a friendly, slightly ironic energy. For a person who has seen the brand, these details are obvious. For AI, they are not.

If we ask an image generator to draw a “tiger mascot”, it will draw a tiger. It may be cute and well-rendered. But it will almost certainly not be our Tigre. During testing, the character could change proportions, become too kawaii, turn into a generic mascot, lose brand details, or take on a completely different expression.

That is why, for covers with Tigre, we worked with a reference image and stricter rules. The silhouette, muzzle shape, eyes, stripes, and overall character had to remain recognisable. Over time, it became clear that a prompt in this kind of work is closer to a small design guide than a one-off command. It has a fixed part: style, line quality, colour logic, restrictions, background, and what to avoid. And it has a variable part: the specific story for each article.

One Article, Three Visual Logics

One of our tests was based on the article “40 vs 7 Hours for Analysis & Strategy”. The piece described how a large research process could have taken around forty hours, but was completed in about seven with the help of ChatGPT and NotebookLM.

The visual metaphor was clear: a large amount of unstructured information gradually turns into a working system.

For the narrative type, we could show Tigra surrounded by documents: a desk, papers, charts, arrows, and small doodle elements around the scene. For the conceptual type, we chose a “chaos to structure” composition: on one side, an unstable tower of documents and flying papers; on the other, a calmer area with a laptop, an AI robot, folders, and a sense of order. For the icon type, we used simple symbols: a brain, a light bulb, a magnifying glass, AI, a robot, and a document.

One topic, three different visual logics. That became the value of the system: not every article has to look the same, but each one can still belong to the same visual language.

What We Learned

We tested different tools. Midjourney worked well for exploring narrative illustrations and composition options, but required close control when the mascot was involved. ChatGPT image generation was useful for quick tests of conceptual illustrations, icons, and background removal. Recraft looked promising for contour graphics and icons, but did not give us the level of control we needed for a consistent character.

The generated image was never treated as the final design. It was material that had to be selected, adjusted, and assembled into a complete composition: with a headline, rhythm, balance, format adaptation, typography, and colour.

Sometimes we had to remove extra elements. Sometimes we had to make the line thinner. Sometimes we had to redraw details because AI misunderstood their shape. Sometimes we had to reject the result entirely. Often, the problem was subtler than an obvious mistake: an image could be technically fine, but too noisy for a cover; charming, but off-brand; expressive, but weak next to the headline.

After several iterations, we ended up with a working foundation for future publications: three cover types, style rules, prompts for different scenarios, a clearer understanding of the tools, and criteria for evaluating the results.

AI becomes useful when the task has a frame: format, style, restrictions, and quality criteria. Without that frame, it quickly produces visual noise. With a frame, it helps us explore directions and arrive at workable options faster.

For us, this internal project became a good example of how AI can sit inside an agency workflow. It helped speed up exploration, test several directions, shape prompts, and build the base of a system. But the system itself came from design decisions: which types to keep, which one to remove, when to use the mascot, when to replace the character with a metaphor, how to preserve lightness, and how not to lose recognition.

The work still starts with the task, the context, and the criteria for a good result. Then comes the system. Only after that comes generation.