“Just start, but with a clear focus”

Mr. Carstens, in June 2023 we talked about how AI is finding initial applications in packaging development. What has specifically changed since then and what impact does that have on your daily work?

Moritz Carstens: AI is now indispensable in our daily work, and we now use it quite naturally in various project phases. The models have become significantly more powerful, and we have built up internal know-how. This allows us to move more quickly from the first idea to well-developed drafts. Experimenting with different styles, for example, is more efficient and diverse thanks to generative AI.

We can also optimize routine tasks such as visualizing prototypes or market research - AI is a huge help there. We also use AI to simulate small market research studies. For example, we create heatmaps or have designs evaluated from the perspective of target groups, which works surprisingly well. Data protection is no longer a big problem. We have developed our own secure tools and "Ways of Working" as well as a "Code of Conduct" that allow us to work safely with the new tools for our clients.

How has the understanding and acceptance of AI developed in your teams or with your customers?

Carstens: It has matured significantly. Initially, there were reservations - some designers were concerned that their own creativity would be displaced by technology. However, these fears have subsided with practical experience. We strongly believe in the interplay between humans and machines - in 'human intelligence' combined with 'artificial intelligence.' Designers become initiators and curators. The machine takes over the elaboration and reproduction across various touchpoints. This 'brand intelligence' that emerges helps our customers improve the quality and efficiency of brand management. So, AI is a powerful tool, but not as intelligent without humans.

We also observe growing acceptance among our customers. We still conduct workshops to show what AI can do and where it can be used, but many now increasingly use AI tools themselves. The adaptation happened incredibly quickly.

Which new AI technologies or tools do you currently consider particularly relevant in packaging development?

Carstens: Generative AI models are currently particularly relevant - both in the image and text areas. On the creative side, we use generative image AI (such as Firefly or Midjourney) to generate visual ideas in different styles for various touchpoints within minutes, which previously would have required days of illustrations or photoshoots. I also find language models equally exciting. With the right model, we can specifically adapt tonalities or create translations - it works impressively well.

In addition, we are observing AI-supported analysis tools that are worth their weight in gold for packaging designers. They can, for example, search through huge databases of existing packaging designs, materials, and sales figures and learn from them. This way, we receive data-driven suggestions on which design elements work well with certain target groups or how we can optimize a layout before we even print it as a prototype.

Specialized packaging AI platforms are also on the rise. Some providers can already generate print-ready PDFs or Indesign files at the push of a button - provided a scalable design system is in place. This is still in its early stages but has enormous potential to automate routine tasks in artwork creation.

Overall, I consider the combination of generative creative AI and analytical AI to be the most relevant - tools that can both generate new ideas and directly test or further develop them. This way, we cover the entire development process more efficiently.

In the first interview, you emphasized that data quality is crucial. Have you now developed specific strategies for data collection or networking?

Carstens: Absolutely - good data is essential if you want to use AI successfully. Therefore, we have invested a lot in our data strategy over the past two years. We primarily feed our AI tools with self-collected and curated data or use models based on royalty-free material.

It is also important to us to network the data. Silos do not help AI. Therefore, we try to link various data sources - such as design assets with consumer feedback or production data. An example: we analyze how certain design decisions - such as a color choice or type of material - affect sales figures or consumer behavior. Such networked data gives AI context and enables it to provide more informed recommendations. We also consistently pay attention to data quality in terms of relevance and timeliness. It is better to use fewer but reliable data than to indiscriminately pour everything into the AI.

What role does collaboration between the various stakeholders - such as marketing, production, or logistics - play in AI-based packaging development today?

Carstens: More than ever, all involved disciplines must work hand in hand when AI comes into play. Why? Because AI can propose holistic solutions that consider all departments - but only if we also feed in the corresponding information from all stakeholders. Admittedly, this is still a bit of a future vision, because large corporations are usually not yet set up that way. But the journey will go exactly in that direction - I am sure of it.

Where do you currently see the biggest hurdles in using AI in packaging development - technical, organizational, or regulatory?

Carstens: Technically, generative AI in the packaging sector still faces limitations. While we create impressive renderings and designs with AI, turning them into a print-ready packaging file is labor-intensive. An AI-generated layout usually needs to be cleanly transferred by designers into our graphic software. There is a lack of precision, such as exact cutting patterns or print-specific settings - AI generally does not deliver a directly usable production PDF. This also complicates the transferability to different packaging formats. In short: it still doesn't work without human refinement.

Organizationally, the challenge is managing the change. AI brings different workflows - we need new skills in the team, must adapt processes, and rethink decision-making paths. Another hurdle is data acquisition and sharing: companies often have valuable data (for example, from market research or production quality) that would be useful for AI models. But they are in different departments or not easily usable due to data protection reasons. Breaking down this silo mentality is a task that many organizations are still working on.

What developments do you expect in the next 12 to 24 months? Are there topics or areas of application that are currently still under the radar but have great potential?

Carstens: Many software providers - whether Adobe or specialized packaging software - are already working on seamlessly embedding AI functions into design environments. This means: A user might not even notice that they are using AI - for example, when layouts are automatically adjusted or image content is generated with a click of the mouse. This convergence of AI and common tools will further simplify and spread the application.

Additionally, I expect significant progress in multimodal AI systems. These can simultaneously understand and generate images and texts. For us, this could mean that an AI analyzes a packaging image and immediately provides us with suitable improvement suggestions or marketing texts - a kind of 360-degree support.

There are indeed a few exciting fields under the radar. One is the fully automatic creation of variants: Today, manual work is still needed to adapt a design to ten different sizes or languages. In the future, AI systems could autonomously generate hundreds of artworks in different formats and language versions with a stored design system. There are already initial approaches for this, but they will probably only reach market maturity in the coming months.

Relatively unnoticed, but very interesting, is also the connection between AI and smart packaging. I think of packaging that uses sensors or digital printing to display individual messages - here AI could generate content in real-time that is tailored to the user or situation.

What advice would you give to companies planning to start using AI in packaging development now - what should they pay attention to, where should they start?

Carstens: My most important advice is: just start, but with a clear focus. Companies should first identify where AI has the greatest leverage in the packaging process. It is essential for me to think about the framework conditions early on and to develop a stance as a company. Many of our customers now look at their brand core to derive an AI stance from it. If I focus on real people, it may not be advisable to let AI represent people - not because it doesn't work, but because it may contradict one's own ethics. This must be defined individually: how and where do we want to use AI and above all why?

In recent years, we have conducted many workshops and have been able to empower numerous customers. Furthermore, companies should organize their data base. AI learns from data - the better the data, the better the results.