At a glance
- AI-supported image recognition reliably identifies the properties of products
- Manual effort and error rates are significantly reduced
- Special procedure considerably reduces the effort needed for development.
The business situation of our client
Artificial intelligence (AI) enables process automation that cannot replace humans, but can relieve them of many routine tasks. Our AI solution for the ANWR Group, one of Europe’s leading retail cooperatives for clothing and, among other things, operator of the shopping portal “schuhe.de”, is an outstanding example for this potential. Originally, the company’s employees were required to manually record the characteristics of products on an ongoing basis in order to be able to provide necessary filter options to costumers – for example shoe type, brand or colour – when searching for products. This approach is not only time-consuming, but also error-prone – especially when a large number of new products is introduced during the change of season.
We therefore developed an AI-supported image recognition for shoes, which automatically tags incoming goods based on the product photos alone. For that, we opted for a procedure that saves effort and leads to excellent analysis results at the same time.
The solution for our client
Our model for image recognition is based on a neural network we trained with corresponding image data. Such a training phase can take several weeks or even months if you have to start from scratch – even with specialised hardware.
That’s why we used a pre-trained image recognition model we could tailor to the requirements of the ANWR Group. First, we adapted the existing keywords to the categories, colours and other characteristics of the shopping portal. This step was followed by several test phases during which we fine-tuned the model. This procedure allowed us to save a large part of the training effort.
How data turns into new values
The project essentially shows that even complex AI applications do not necessarily require a lot of effort. Despite the extremely short training phase, the ANWR Group now has a highly specialised model for image recognition at its disposal, saving a significant amount of time when it comes to tagging products. The model is able to determine product characteristics with astonishing precision. The error rate is now lower compared to manual data entry. And: image recognition is not limited to professional product photos; it works just as well with photos taken with a smartphone.
turn your data into value.