Previously, F.A.Z. approached the challenge of choosing which articles to place behind the paywall retrospectively, by looking at past records and figuring out through trial and error what articles perform best based on experience. The new AI tool relies on different sources such as the conversions of all previous paid articles, including some metadata like author, department and date of publishing. All texts are analyzed with Google’s BERT technology to provide more accurate comparability. BERT is Google’s neural network-based technique for natural language processing (NLP). BERT stands for Bidirectional Encoder Representations from Transformers. In short, BERT can help computers understand language more similarly to the way humans do. Nico Wilfer, Chief Product Officer, adds: “In line with our editorial rules, the algorithm makes the suggestion but does not decide automatically which articles are placed behind the paywall. The element of human choice is an important role in our considerations when implementing Artificial Intelligence in our products.”

As well as overall subscription growth, F.A.Z. tracks the quality of the AI tool and checks the variance of the suggestion against the final conversion rates. This tracking measures the quality of the tool’s predictions. Results to date show that the AI tool is very effective at predicting future article conversions, with almost half of all predictions nearly 100% accurate. Wilfer says that F.A.Z. ‘s editors are very interested in the use of an AI service that supports their daily work and are learning how to make the most of the tool.