Consulting & Innovation
Solutions & Products
Infrastructure & Operations

How Artificial Intelligence is Revolutionizing Goods Receiving in the Mail Order Industry

A look into the everyday life of an online jewelry retailer

Artificial Intelligence in the Shipping Industry
Artificial Intelligence

Artificial Intelligence (AI) is an excellent way to support companies in the area of inventory management and, in particular, in the processing of returns. A look at the day-to-day operations of an online jewelry retailer shows why.

What Problems Do We Face in the Returns Department?

Anyone who deals in jewelry has to manage a large number of small returns: Products that are not returned properly and in their original packaging often do not have an item number. So they have to be entered into the system manually. However, this is only possible if it can be determined without a doubt which article it is. Seasonally changing collections and strong similarities between individual pieces of jewelry make it difficult to identify the right product. A search in the catalog takes a long time and is also extremely error-prone.

How Can the Returns Handling Process Be Made More Innovative?

Like any manual process, the returns processing operation offers significant potential for efficiency gains. A first approach could be to support warehouse staff in narrowing down the possible assignments of incoming items. So-called selection dialogs based on a keyword search for material, object type and size, for example, would be practical. The crux of the matter is that employees need to know the exact keywords. Permanent and conscientious catalog maintenance and intensive training of warehouse staff are essential. Although this approach is helpful, it does not lead to quick results or to a constant improvement of the process.

Object Recognition with Artificial Intelligence Solves the Problem

Let's think one step further: machines must not only help to allocate items, but above all ensure that each item is recognized beyond doubt. Only then will returns processing really be accelerated.

With the help of machine image recognition, a comparison can be made between the catalog images available in the online store and the returns. This means that the item number can be determined and the jewelry can be returned to the system.

So much for the theory. In practice, however, it becomes clear that simple image matching is not enough: reflections of shiny materials, different perspectives and lighting conditions lead to errors in image recognition.

This is where the potential of AI comes into play - specifically, artificial neural networks. Similar to the human brain, these have synapses and nerve cells and are able to simulate parts of human thinking - precisely the parts that are necessary for solving the problem, abstracting and classifying.

Neural Networks Optimize Incoming Returns

The prototype built to solve this problem had the function of recording videos and then analyzing them live. Within a few milliseconds, the system calculates probabilities for each item. The advantage of this approach is that the return is detected even in different lighting conditions, viewing angles and backgrounds because an exact match is not required, only a probable match.

If a previously defined value is reached, the result is considered so probable that it is possible to correctly record the return and also automatically route it to the appropriate location in the warehouse system. If, in individual cases, the system does not arrive at a sufficiently reliable result, it makes suggestions to the warehouse employee, who then selects the correct piece of jewelry by simply clicking on the corresponding image. In this case, the employee continues the process.

Of course, it is conceivable to analyze other characteristics such as weight in addition to optical attributes. This information could be used to draw conclusions about the material. To do this, a scale would have to be integrated and the neural system would have to be given the relevant details it needs to learn which characteristic of the "weight" feature belongs to which article.

Brain Jogging for Neural Networks

In order for the neural network to be built up, it must be trained with images. In this way, it "learns" and can make classifications independently.

The outlined solution can be applied to almost any assortment. Since online stores usually have a large number of photos of their products that the neural network can use as "training material", all the prerequisites for a fast implementation of automated returns are given. Even a change of collection or the addition of new products and entire product groups can be easily handled in this way. The AI creates a considerable time advantage here, because the neural network needs much less time to gather experience than the employees.

Outlook: Application Areas of AI in Companies

Intelligent software and advanced algorithms enable us to hand over tasks to machines that until recently could only be done by humans.

Powerful neural networks, so-called deep-learning systems, are capable of replicating the functioning of networked neurons in the human brain. Deep-learning systems are particularly capable of extraordinary performance in object recognition and natural language processing. Images are recognized correctly even if the objects depicted on them are twisted, partially obscured, or taken in unfavorable lighting conditions. Computers achieve error rates that are only half those of humans.

Neural networks are superior to humans wherever patterns need to be identified in text, video, image or audio files. However, humans are still needed as an instance that tells the neural network what the recognized objects are about, i.e. that makes the assignment. There are areas of application for neural networks in every company, with the entire warehouse and logistics sector benefiting particularly strongly from the increase in productivity.

Pattern identification is an important application area for AI technologies, because in these areas machines can already work more error-free and faster than humans thanks to high computer performance. What is particularly exciting is that neural networks are even capable of discovering patterns that were previously unknown. Artificial intelligence (AI) is particularly suited to taking on repetitive tasks and is therefore able to make processes more efficient and significantly increase productivity in the area of warehousing and incoming returns. Employees, on the other hand, are freed from monotonous tasks and can devote themselves to planning or creative activities.

SCM & Logstics

Supply chain management is changing rapidly - with innovative solutions and tools we secure the decisive logistical advantage for our customers!

platbricks - the Arvato Smart Logistics Plattform

platbricks is a cloud-based platform with which operational processes in logistics can be intelligently planned, controlled and digitally (mobile) executed. Let us show you how platbricks closes your digitalization gaps.

AI in Practice

Be inspired by our application examples for AI. Discover our AI cases from practice.

Written by

Meikel Bode
Expert for AI & Data Driven Business