AI prototype restructures 10,000 pages
by “Swiss Post”
Searching becomes more difficult the more content there is: The limits of manual web management
Large company websites are constantly growing and with them the complexity.
They have to cover many different services, target groups and content. Swiss Post’s homepage currently comprises around 10,000 pages in four languages.
Industry comparisons show that maintaining all these pages corresponds to at least six full-time positions, which would have to be spent exclusively on keeping the content up to date.
The problem is that only 20 pages account for 80% of the traffic.
The invisible content load: Outdated, untraceable, irrelevant
The problem is that just 20 pages generate 80% of the traffic. The rest is barely used, outdated or hidden deep within the structure. Although a lot of information is still correct in terms of content, its presentation seems outdated or it is so difficult to find that it is hardly accessible to users.
How AI reorganizes 10,000 pages: The PoC in detail
Together with Swiss Post, we have implemented a proof of concept (PoC) that demonstrates how artificial intelligence can organize content in a structured way, update it automatically and make it available in a user-friendly way.
To this end, 10,000 pages and 5,500 images were vectorized – i.e. converted into a digital, mathematically describable form – and transferred into a searchable knowledge network. This now makes it possible to make queries in natural language.
However, user feedback has suggested that a static-looking website leaves a more competent impression on some customer segments than the response from a chatbot, even if it is correct. The consistency of the information is obviously valued more highly here.
The next step: converting vectorized data into finished web pages
Based on this insight, the PoC was further developed so that pages in the corporate design are automatically created from the vectorized data using a dual LLM architecture.
The quality is checked via a QA interface and verified by unit tests (evals). The solution enables a hybrid UX in which AI provides direct answers, while classic elements such as filters or navigation continue to provide guidance.
A scalable AI system for future-proof content management
The PoC shows how a cumbersome information archive can be transformed into a dynamic platform. An experiment has resulted in a scalable AI system that can be transferred to other companies.
The solution can be seamlessly integrated into existing IT infrastructures and is already equipped for future technologies such as Microsoft NLWeb. For Swiss Post, this represents a strategic opportunity to further develop its website into a service universe by 2027.
Take-Aways. Learnings. Challenges.
Take-Aways. Learnings. Challenges.
Take-Aways. Learnings. Challenges.
Take-Aways. Learnings. Challenges.
Take-Aways. Learnings. Challenges.
- Traditional content maintenance is not scalable, as the effort and benefits are disproportionate.
- AI does not replace editorial offices, it creates a platform that automatically keeps content up to date and makes it controllable.
- AI doesn’t just make websites more efficient – it fundamentally changes how users experience information.
Norbert Kremer, Lead Services
Interested in finding out more about the project and Panter?
Talk to Norbert and find out how years of experience and expertise can help your projects move forward.
