– Reading Time: 3 minutes –
With 2.38 million visitors per month, billiger.de is one of the most established price comparison portals in Europe. It is operated by solute GmbH and lists prices for millions of products. We spoke with the CEO of solute GmbH, Bernd Vermaaten, to find out how the portal has succeeded in achieving the enormous increase in visibility.
BV: In recent months, we have consistently removed inferior content from the portal and worked on the expansion of high-quality unique content. In addition, we have prepared a fundamental analysis of the most important topic areas, including mobile phones, televisions and perfumes. Based on these results, we were able to identify the keyword potential. This enabled us to create corresponding URLs and automated content by the online editorial team with the help of complex templates. Another major change was the optimization of meta titles and descriptions.
BV: We at billiger.de are confronted with an ever-growing demand for content. Yet it is impossible to write the articles manually due to the huge product variety of our portal. The manpower costs for this quickly would become unaffordable. That’s why we use the NLG-Cloud at category level for product descriptions with AX potential.
Product categories have to meet three criteria to have AX potential. First of all, the categories must have a high turnover so that the work involved in creating a sentence structure is worthwhile. In addition, many products should be available in this category, as automated content is currently only worthwhile for a certain number of units. And last but not least, it is important that there is a lot of well-maintained product data for the items. The more product data that is stored, the more text modules are triggered in the finished AX training. This is the only way to achieve a detailed and well written text. Completeness and accuracy are also basic requirements. If the data is faulty or incomplete, the generated content will be, too. Categories with bad product data are therefore out of the question for AX training.
Here is an example of a product description from billiger.de that was created with the software of AX Semantics:
BV: We at billiger.de assume that the connection between editorial staff and technology will increase strongly in the future in view of the growing demand for content. This also became clear at this year’s AXCD, where Hybrid Content Editing was presented for the first time as a necessary reaction to this change. Our concrete work steps are as follows:
The Category Management and SEO teams work closely together to identify potential product categories for the NLG-Cloud. Based on the identified categories, the online editorial team creates the next promising project. Text modules with potential added value are identified. The creation of the training is based on a concept that should bring added value to the user and is visually appealing. The AX experts of the online editorial team then start researching the category, linking logic statements and writing text variants in the NLG Cloud. After completion of the training, the API interface to AX Semantics will be activated and all text items will be generated. A QA measure, introduced by us, marks products with insufficient data or too little text (less than 800 words) as “low quality” and thus prevents the automatic upload into the portal. The content that conforms to our quality standards can then be viewed directly on billiger.de without any additional effort. As soon as new data is stored in the database, the API reacts and generates new text or updated versions when the data changes.
BV: We are already very successful in using Hybrid Content Editing, but there is always room for improvement. For us, improvement efforts focus on the optimization of the AX process through closer cooperation among the departments involved. We also have to transfer our old projects to the current software version. In terms of content, the categories populated by AX are about to be expanded. So it is our goal to create less redundancy and more variance through fine-grained content. We are also working on optimizing unstructured, automatically created text. Here we want to exclude the possibility that the features from the product information box will appear again in the continuous text because this repetition would not bring any added value to the user.