Packaging Size Determination via Machine Learning
The processing time needed to fill orders is critical for e-commerce warehouses. For this reason, it is necessary to automate some warehouse processes to save valuable time. One activity that can be automated is the selection of package sizes. Through machine learning, the process of predicting which sized package is needed to fill an order can be continuously improved for more efficient work.
How Are Package Sizes Currently Selected?
Currently, employees in our warehouse choose the appropriate package size to fill an order based on instinct and experience. First, they estimate what size package they need, then confirm the size by scanning the package or pressing a button. The manual selection process is both time-consuming and relies on an employee’s experience. New employees need to develop a knack for packing before they can be successful at predicting what size parcel they need.
How Can we Improve this Process? By Using Machine Learning!
Machine learning can add significant value across the e-commerce field. Possible areas of application include fraud detection and optimising product recommendations. In our case, this innovative technology could also be used to predict the right sized packaging for our orders. The goal would be to develop a machine learning model that independently determines the correct package size based on the products that need to be packed in the order.
How Can We Achieve this Goal?
We’re currently working on a concept for a machine learning system that could predict packaging sizes. One idea is that the machine learns which combinations of products fit in which sized packaging based on existing data from past orders. The machine identifies patterns in our data that can be used to predict the parcel sizes needed for future orders. The model would be based on a neural network that is trained using supervised learning. The underlying model could use TensorFlow and run as a microservice. CSV files containing information about the order’s weight and – if available – the volume of the products in the order would be used to feed data into the model. If the order includes a fragile product, the system would need to know, as fragile products take up more space due to the additional packaging they need. Based on this information, the model should predict which size package is needed to fill the order.
The Four Steps of Implementation:
1) Storage
To begin with, we need to collect the data related to allocating package sizes. An abundance of good data forms the basis for an accurate prediction. We’ve already begun storing data, and keeping records of the different packaging sizes needed to fill various orders. We’re still working on the process and are in the planning stage.
2) Analysis
In the future, we’ll be able to analyse the weight and volume of past orders to better predict the package sizes we need to fill future orders.
3) Processing
Next, we will build a machine learning model, test it, and fix any errors. Finally, the data could be sent as a CSV file to the microservice, which would report back the correct packaging sizes.
4) Application
When the microservice predicts the package size needed to fill an order, our colleagues in the warehouse will no longer have to guestimate the package size they need themselves. If the microservice suggests the wrong size, the employee would manually correct the suggestion so that the machine learning model could learn from the mistake.
Conclusion: a Fully Automated, Improved Process
This innovative system could replace a tedious manual process with a fully automated, self-improving process. This could massively increase efficiency in the warehouse, saving valuable time and costs. Additionally, machine learning models could also make predictions regarding warehouse stock schedules or the volume of packages couriers should expect.
If you want to be a part of more exciting projects like this, you know what to do!
