For many people, starting the day without a cup of coffee is inconceivable – and this explains the popularity of fully automatic coffee machines. Customers can quickly find what they’re looking for in MediaMarkt’s and Saturn’s online shops, where over 350,000 articles are permanently available – including, of course, plenty of fully automatic coffee machines. This huge choice for customers means a huge amount of writing for MediaMarktSaturn’s online team. After all, each product needs a comprehensive blurb to help customers make the right decision. And this job is now done completely automatically using a text robot.

The Saeco PicoBaristo looks great, but does it truly meet my requirements? MediaMarkt’s online shop provides customers with all the information they need. As well as the technical specification and product reviews, the page also contains an extensive description. Everything about the machine’s grinding mechanism, pump pressure and LCD display is explained down to the last detail. At first glance, this service is nothing unusual for a webshop – assuming the text was written by a member of staff.

But in fact, it was written by a robot – although on reading it you’d never guess! The ‘robot’ is a special computer program that the online team began using this year to automatically generate product descriptions. “We always get information about items from the manufacturer. But frequently we only receive technical data and not actual texts, resulting in inconsistent presentation in the online shop,” explains Christian Brandl, Head of Content, who is in charge of automatic product description in the online team.

Texts are automatically written by a new piece of software: a text engine developed by the Berlin-based Retresco, one of the world’s leading providers of AI solutions for the automatic exploitation of content and language. “The more products an online shop contains, the more complex it is to write unique product blurbs. Automatic text generation significantly raises the efficiency of the entire content production process,” says Florian Spengler, who is in charge of the project for MediaMarktSaturn at Retresco.

Natural language is generated automatically

The program works on the principle of automatic natural-language generation (NLG), a branch of artificial intelligence. In a nutshell, NLG is the process of converting structured data in areas such as e-commerce and using it to create reader-friendly texts.

To generate a product description for the Saeco PicoBaristo, pre-formulated sentences must first be defined, which are then individualized with the aid of data. The result is a series of sentences such as: “Another important parameter is the maximum pump pressure of 15 bars.” The text engine also interprets information like this by adding that the pump pressure allows the steam to pass through the espresso powder at an optimum speed to extract as many flavours as possible.

The software then rearranges the resulting sentences in a certain order. In addition, the system analyses variance within a product type. This means that it compares the data of all, say, fully automatic coffee machines and then classifies the items statistically – in other words, it determines whether a certain coffee machine is in the upper, average or lower range regarding a certain parameter compared to other machines. Lower-range devices are advertised as entry-level models, for instance, whereas machines with the best parameters are sold as high-end products.

More than 210,000 product descriptions have already been generated for articles in 427 product groups for the online shops of MediaMarkt and Saturn – and this figure is of course increasing all the time.

Customers receive detailed information and a classification

There are several advantages to automatic text generation: “Our customers receive detailed information about the products and also about their position within the product category. In addition, the blurbs make our products more relevant for internet search engines,” explains Franziska Knabel, who is responsible for the text robot in E-Commerce.

Another advantage is that as the system keeps learning, it delivers increasingly accurate suggestions and even improves its use of synonyms in product descriptions. It’s never too late to learn for neither humans nor machines – and certainly not for text robots! For Christian Brandl and Franziska Knabel, this means their work with the text engine is an ongoing process with no end date. “New product groups are constantly being added,” says the head of E-Commerce. “We’ve set ourselves the goal of generating an automatic text for all our products over the next few months. Our next key goal will be to integrate the services we offer at MediaMarkt and Saturn into the texts.”