Applying artificial intelligence to optimize your supply chain comes with many time-consuming challenges such as model development, model deployment, a user-friendly interface for business users and required organizational change. This article explains how analytics is applied and developed for a client in minimal time.
The case concerns a customer in the logistics industry with several large warehouses to store inventory. To deliver quickly and efficiently, every night a complicated loading process takes place. Due to many different variables, doing so efficiently poses a logistical challenge.
It may vary depending on for example the number of trucks, ordered goods and loading stations. The analytics model we implemented must be able optimize the total loading time of all trucks through the warehouse. In this case, a well-configured optimization model suffices. In other cases, artificial intelligence may outperform optimization models, but that is dependent on other variables such as the data.
There are generally four challenges. The first concerns collecting the right data from different systems and transforming it in such a way that it’s suitable for the use case. Secondly, the (AI) model must be developed. Thirdly, once you have the data and the model, you need to operationalise it and maintain it. And at last, how do I ensure the business can benefit from this? How to make it user-friendly?
Diversified data sources and data format
In this case we used Mendix to rapidly unlock data from the source systems, transform it and complement it. In Mendix, Aiden created a pipeline that collects data via SAP connectors and more simple REST APIs and performs different data transformation steps. As a result, the different data sources are automatically grouped together and each new entry in the data is prepared for further use by the optimization model without manual interference.
Developing the model
The model was developed by translating the current supply chain into operational rules. These rules, combined with the data coming from Mendix, were used by the model to calculate the shortest throughput time. Greenfield Data Enhancing and DataMetric created a piece of software code that transforms the data from Mendix and performs the data sets necessary to continuously search for the shortest throughput time. Any misinterpreted supply chain condition in the model would result in a non-feasible and non-usable model output. As a result, the input from Mendix is automatically transformed to a schedule that is sent back to the Mendix. Because of the little time necessary to operationalize the model, more time could be spent gathering all the supply chain conditions and clearly communicating and validating them with all cooperating parties.
Deployment and lifecycle management
Without operationalisation the model will not deliver the desired value. Simply put, if the model stays on the developer’s laptop, the end-user cannot access it. And what if the laptop gets stolen, has insufficient processing power, or gets hacked? Therefore, deploying the model in a stable, reliable and secure environment is extremely important. Moreover, you need to deal with changes in the input data over time, new data sources may emerge, or if another model seems to work better, you need to revise and re-upload the model. In other words, you must also think about the lifecycle management of the application. How do I ensure it continues to deliver value?
UbiOps allows the data scientist to deploy and maintain their models easily, quickly and without any IT dependency. It takes care of everything you need, such as API management, scalability, security and provides an easy user (developer) interface. With the UbiOps Plug-in for Mendix, it’s also connected to your Mendix app in no-time.
For this use case, the model developed in the previous phase is deployed on UbiOps.
While one may have all the ‘back-end’ aspects sorted out, it often remains a challenge to create a user-friendly interface (front-end) that non-technical users can use for inputs and outputs. In this case, employees of the planning department. Also the front end must allow for easy integration with different back-end systems. Aiden used Mendix to build an agile yet user-friendly interface that allows for intuitive use by the planning department.
Will you differentiate from your competition with Mendix low-code and enhanced data analytics with UbiOps?
If you want to learn more about how to build Mendix applications powered by analytics code or machine learning, please reach out to us.
Eric van der Maten
[email protected]: Partnerships at UbiOps
[email protected] : Account manager Aiden
[email protected]: Owner Greenfield Data Enhancing
Aiden, a full service IT partner within the Benelux, focused on creating value and solutions for complex business challenges. The main focus areas are wholesale, retail, manufacturing and consumer goods where innovative and proven technology such as SAP and Low-code solutions are combined with innovative ML platforms.
UbiOps, the MLops platform to run and manage AI at scale. With UbiOps you can deploy and maintain your Python / R models easily, quickly and without any IT dependency. It takes care of everything you need, such as API management, scalability, security and provides an easy user (developer) interface.