... is revolutionizing the way returns are received and managed
Artificial intelligence is becoming more popular, and for many companies that already have used the innovative technology to optimize their products, it is now indispensable. Those optimizations range from smart refrigerators that automatically reorder food to toothbrushes that analyze tooth brushing techniques and show potential for improvement. Not only does artificial intelligence affect new product variations but it can also simplify entire processes, resulting in significant cost savings in the long run. The following is an example of how online and mail order return processes can improve.
What problems have to be overcome in the return receipt?
Commonly in returns, the returned item may be missing the original article number or other pieces of needed information. Matching the item with the article number can be time consuming and error-pron. For example, those who trade jewelry have to manage a large number of small-sized return items: So the returns have to be manually posted to the system. However, this is only possible if it can be determined beyond doubt which item it is. Seasonally changing collections and substantial similarity between individual pieces of jewelry make it challenging to identify the right product. To search for an item in the product catalog takes a long time and this manual process also opens the doors widely for errors. Thus, Arvato Systems has created a joint POC with a jewelry company to solve the problems within the return process.
How can the processing of returns become more innovative?
Like every manual process, the processing of returns offers significant potential for the increase of efficiency. A first approach could be to help the warehouse staff to narrow down the possible classifications of incoming articles. So-called selection dialogues would be useful with function based on a keyword search, for example, the material, object type, or size. The problem with this approach is that the employees must know the exact keyword. Hence, permanent and conscientious catalog maintenance and intensive training of warehouse staff are essential. The approach might be helpful, but it does not lead to fast results or a constant improvement of the process.
Object detection with AI solves the problem
To really improve the processing of returns, machines not only have to help to allocate the articles but overall they have to make sure that each piece is clearly recognized. With the help of machine learning based image recognition, a comparison of the images displayed in the web store and the actual returns can be made. Thus, the item number is determined by the machine and the piece of jewelry can be returned. Unfortunately, this theoretical approach is not sufficient enough to be used in real life. The reflections on the glossy materials, different perspectives, and lighting conditions lead to errors in image recognition. To make the theoretical approach applicable for real-life processes, AI and specifically artificial neural networks are needed. Similar to the human brain, those artificial neural networks have synapses and nerve cells and can simulate parts of rational thoughts which are the parts necessary to solve the problem of separating and classifying.
Neural Networks Optimize the Return Receipt
The prototype Arvato Systems built to solve this problem had the function of capturing video and then analyzing it live. Within a few milliseconds, the system calculates probabilities for each article. The advantage of this procedure is that the return is recognized even in different light conditions, angles, and backgrounds because only a probable match is needed.
If a defined value is met, it is possible to record the returns correctly and then also automatically route the item to the appropriate location within the storage system. In the case that the system does not achieve a sufficient result, it makes suggestions to the warehouse employee, who then selects the right piece of jewelry with a simple click on the corresponding image. In this case, the employee continues the storing process. An analyzing of material or other additional features such as weight is also possible. To do so, a scale would have to be integrated and the neural system needs to be provided with the required product weight details. The in-house built prototype is transferable to almost every product or merchandise.
A neural network in inference mode is supposed to classify a previously unknown object ...
... and decides on the class of rectangle (classification) based on the knowledge learned in training mode, although neither alignment nor color has previously occurred (abstraction).
Training for Neural Networks
To build the neural network in the first place, it has to be trained with pictures. In this way, it "learns" and can independently make classifications.
The solution outlined is transferable to almost every assortment. Since online shops usually have a large number of photos of their products, which can use the neural network as a "training material," all the conditions for a quick implementation of the automated returns receipt are given. Also, a change of collection or the addition of new products and entire product groups can be handled easily in this way. AI creates a significant time advantage because the neural network takes much less time to gain experience than the employees.
Outlook: Application areas of AI in companies
Intelligent software and advanced algorithms enable us to transfer tasks to machines that, until recently, could only be done by humans. Robust neural networks, so-called deep-learning systems, can simulate the functioning of networking cells of the human brain. Deep learning systems are exceptionally capable of exceptional performance in object recognition and natural language processing. Images are recognized correctly, even if the objects displayed on them are twisted, partially obscured, or shot in unfavorable lighting conditions. Computers achieve error rates that are only half those of humans.
Neural networks are superior to humans wherever patterns in text, video, image, or audio files have to be identified. However, the human being is still needed as an instance that tells the neural network what the detected objects are, i.e., what they do. Every company has applications for neural networks, with the entire storage and logistics sector benefiting greatly from the increase in productivity.
The identification of patterns is an essential area of application of AI-technologies because, in these areas, the machines can already work faster and faster than humans, thanks to their high computing power. Most exciting is that neural networks are even able to discover patterns that are not yet known. Artificial Intelligence (AI) is particularly suited to take on repetitive tasks. It is, therefore, ready to make processes more efficient in the field of warehousing and returns and to increase productivity considerably. The employees, on the other hand, are freed from monotonous tasks and can devote themselves to planning or creative activities.
Are you interested to learn more? Please contact Meikel Bode to discuss how AI can improve your return processes.