AUTOMATED INTEGRATION OF PRODUCT DATA

The Austrian online marketplace offers small and large domestic retailers access to the digital market to sensitize consumers to local brands and providers. Extensive and detailed product data is required for both stationary and onlineCommerce retailers to be successful on the Austrian digital platform.

The online offering is to be expanded rapidly and the onboarding of retailers and manufacturers is to be speeded up in order to be able to offer more articles on the Austrian online marketplace in a targeted manner. The suppliers' product data are to be automatically integrated into the target data structure of shöpping.at and the quality of the product data is to be improved. These steps are intended to accelerate the growth of shöpping.at and increase the conversion rate.

Process automation was a central success factor for the further increase in efficiency of shöpping.at. This is only possible with state-of-the-art software in this area. For this reason, the Austrian online marketplace considers the cooperation with Onedot to be very useful, in particular the possibility of using artificial intelligence and machine learning for data processing via Onedot in order to expand the desired process automation with this promising technology.

HOW ONEDOT HELPED

shöpping.at enables suppliers to make the extensive product catalogues available in their own formats. Onedot read in these different formats, corrupt data sets were identified and sorted out. Since the supplier data was sometimes not very structured, attributes had to be extracted from the product texts first. Then the product data was assigned to online categories on shöpping.at. The system thus ensures that online users can find the products more easily. The Onedot AI proposals were checked by the shöpping.at data team and incorrect assignments were overridden. This feedback helps the software to learn and in the future to perform such categorisations more accurately.

The available and extracted attributes were automatically mapped to the target attributes of shöpping.at. Since each supplier designates its own attributes differently, one mapping had to be made per supplier. To ensure that search filters in the online shop can be displayed uniformly and consistently, value lists, numbers and units, among other things, were standardised. In order to display the different configurations of the product in a user-friendly way, product variants were created based on the relevant attributes, such as size or color.

THE RESULT

The supplier onboarding took place in batches. A batch depends on how many new products the suppliers have provided at that time. A representative batch contains about 100'000 products or more for the setup of the marketplace. The four-level categorization achieved an accuracy of 94% in two iterations. Compared to manual mapping, the Onedot software's machine mapping offers the possibility to transfer more attributes. These attributes can then be used for additional filters and further simplify product searches. On average, suppliers provided only 26% of the filter attributes in a structured form. By refining the data, the Onedot software was able to increase the fill rate of the filter attributes from 26% to 86%. 97% of the products could be combined in possible product variants. This high-quality product data provides the basis for a successful customer experience on the marketplace, better conversion and more sales.