All about the SMART Demand Forecast System Update

The functioning of the entire supply chain begins with demand forecasting. How much to order to satisfy consumers, and how to avoid freezing inventory, prevent write-offs, and ensure proper customer service are questions that supply chain managers ask themselves every day when building forecasts for each SKU. To ensure their high accuracy, there are solutions such as SMART Demand Forecast, which are based on machine learning and artificial intelligence algorithms.

The SMART business team works daily to improve the SMART Demand Forecast functionality. Our focus is always on improving the financial performance of our customers’ business and providing the best user experience, so in this release, we have developed the table personalization functionality, optimized the promotional campaign import process and cross-browser compatibility, accelerated the interface and Data Science processes, and implemented the technical system optimization. Read the article to learn more about all system updates.

Product documentation has been updated

The Security and Environment Policy documents, that describe the operation and interaction of environments, standards and approaches to the security component of the product, have been finalized. The documents regulate the interaction of SMART Demand Forecast with the data of our customers.

The project management structure has been updated. We are meticulous in creating and keeping product technical documentation up to date in order to provide customers with the opportunity to familiarize themselves with documents regarding business processes, solution architecture, terminology, etc.

Table personalization functionality has been developed

The SMART business team has added table personalization functionality with the ability to select the number of displayed rows, customize the list and column sequence.

This customization allows users to change the appearance of the tables while working with the demand forecasting system, which in turn greatly improves the user experience.

In other words, if the nuances of your work require customization of the interface, you can move or hide the information as needed.

The process of importing promo campaigns has been optimized and the loading speed has been increased

We always care about ensuring fast and reliable operation of the system. Therefore, we have worked hard to reduce the waiting time when importing promo campaigns and their subsequent validation in SMART Demand Forecast.

Business rules and process of basic Data Health Check have been developed

When integrating data into the system, it will automatically check the downloaded data for critical errors that will interfere with the main function of the solution: preparing a forecast. In this release, we have added a check for:

• metadata
• data integrity
• compliance with the reference structure

System interface performance has been improved

To prevent repeated calculations, saving the results of executed functions has been implemented in the system. Thus, there was a significant increase in interface performance when loading system pages.

Cross-browser compatibility has been ensured

The correct display of the interface and work of the system functionality, regardless of the installed browser, has been set up.

The functionality of forecasting to the higher aggregation levels has been developed

Previously, the processes for forecasting at the lowest level of aggregation were implemented in SMART Demand Forecast. To cover a wider range of business tasks, the team worked out approaches to generating a model that allows forecasting at higher levels of time, product and business aggregation. Weekly and monthly aggregation levels have already been implemented in the system.

The concept of Time Series forecasting by models has been validated

The system works with one model for forecasting, LGBM. To develop a multi-model approach for different business requirements, a Deep Learning model has been developed for time series forecasting. Work on the implementation of this concept into the system will be included in release 2.0.

Accelerated Data Science processes In release 1.2, a number of frameworks and approaches were tested to reduce the execution time of data science processes, which will affect the cost of commercial operation of the SMART Demand Forecast system. The process of parallel scoring of several complex scenarios was considered, which in turn will speed up the provision of user forecast results.

The system has been technically optimized

An important aspect of the IT solution is to provide users with stable and fast system operation, given the constant processing and analysis of large amounts of data, so the SMART business team carried out a technical optimization of SMART Demand Forecast, including:

• Adaptation of ICacheValidator methods
• Development of the process of incremental loading of data into the system database
• Testing and modification of measurement unit conversion processes.
• Optimization of performance for executing stored procedures in the application’s database
• Automation of the Feature Selection module operation

The factor selection module has been optimized according to the input conditions. Two running modes have appeared: manual and automatic. Manual mode is required for the pilot stage and finer tuning of the solution for the customer. Automatic mode is necessary for comprehensive use with minimal human time spending.

• DS processes code has been refactored

The SMART business team refactored and optimized the step of generating a planned data set and importing the library in Databricks notebooks, and also reduced the size of the Docker image. Thanks to this, the code became cleaner and more understandable.

For more information about SMART business solutions and services, please call +38 (044) 585-35-50, or submit your request here.