SELF-LEARNING CUSTOMER ORDER DATA ENTRY

TRANSPORT & LOGISTICS SECTOR

BRIEF

Manual processes cost companies time and money. Employees often spend half their time doing repetitive tasks and data entry. For one of the most prestigious transport & logistics companies in Europe, our Belgian team created a proof of concept on how to automate their manual order entry process. The company receives 300+ customer-specific orders daily in the form of native PDFs that are manually entered into their system. Our goal was to automate this manual process which can lead to a significant decrease in operational cost.

Within a short period of time, our team of software engineers and data scientists created a proof of concept in the form of a live demo. By extending it to 4-5 months, we can further develop the proof of concept and turn it into a real-life solution with a 60% automation level. This solution will not only improve work efficiency but also significantly reduce costs. In order to reach the maximum automation level, this project requires our support in the areas of:

Project Management
Frontend development for processing Order PDFs
Backoffice development for updating the algorithms/ model
Data structuring and preparations
Algorithm development and training
Testing and improving

Demos-self-learning-data-entry-application

THE CONCEPT

300+ customer-specific orders in the form of native PDFs are manually entered into the system daily.
The idea is to create an application that loads PDF orders and detects as much information as possible using an intelligent self-learning algorithm. It means that it can fill in the required fields with data from the PDF, business logic, and historical data. Once the user is satisfied, the application either generates an export file or calls an API which will be used for further processing in the actual order entry system.

CUSTOMER CHALLENGE

As we further develop the application, the real challenge is to make sure that the application can recognize texts inside of images or unstructured PDF documents. Lastly, the application needs to be able to read and understand e-mails and one-time orders of infrequent customers. All of this requires intensive training, testing, and improving the algorithms/ model so the application can process information accurately.

MEDIAAN IN ACTION

During the development of the proof of concept until the development of the actual solution, the following techniques are used:

Angular
Python
Microsoft Azure
Docker
Scrum/ Agile

SOLUTION IN DETAILS

The proof of concept is able to

Collect data and label examples to reach the maximum automation level
Convert PDFs to text
Apply business logic and use historical data to fill in required data-fields
Provide a graphical user interface to display PDF’s while processing it
Aid in annotating data that is automatically recognized or suggest likely options based on data formats
Generate exports files or call API’s as output to other systems
Send training data back to self-learning algorithm

To reach the maximum automation, the application needs to be enhanced so it can

Recognize texts inside images/unstructured PDF documents
Read and understand e-mails
Continuously trained to ultimately result in a self-learning algorithm for detecting data in the PDFs
Make a technical/automated connection to Order Entry/ERP applications
Accurately process one-time orders of infrequent customers

RESULTS

Within a short period of time, our team of software engineers and data scientists created a proof of concept that shows the benefits of automating the manual order entry process. By extending it to 4-5 months, our team can further develop this proof of concept into an actual solution that will lead to a 60% automation level. This solution will not only improve work efficiency but also significantly reduce costs.

WANT TO SEE MORE?