Is your production not made of continuous products? Do you need to forecast sales of a very high number of products? Is your product portfolio highly seasonal?
Are your procurement times so long that you need to know what the final demand figures will be well in advance?
Is the cost of excessive stock a problem? What about lost opportunities due to out-of-stock?
These are some of the questions that the Demand Planner needs to answer on a daily basis: large amounts of data need to be managed (product data, historical data at customer and SKU detail level, sales forecast by customer-SKU, …). More often than not, there may not be the appropriate tools to monitor actual sales and generate accurate forecasts.
The demand planning function is asked every day to reduce the error in forecasting while maintaining a high level of service to the end customer, all the while minimizing the costs of procurement / production / storage.
AN Foresight applies machine learning to demand data to respond quickly to all these questions, producing long / medium / short term forecasts for each portfolio SKU, minimizing forecasting errors and automating the procurement process.
Foresight: our approach
Foresight is a cross-industry application aimed at different markets, such as, but not limited to, manufacturing, retail, pharma and fashion.
Foresight is able to predict demand both in the markets where the time series of products are available – even if they need to be re-built following product phase-in-phase-out logic – and in the markets where, due to the characteristics of the market itself, there are no usable / consistent time series of products, such as in the fashion world where collections are (almost) fully renewed each season.
For each market we have studied and trained specific machine learning models capable of capturing their own specific traits, in order to integrate the upstream and downstream processes for Demand Planning to use, and provide the best forecast possible to act as the starting figure, therefore optimizing the processes of:
- Warehouse management
Another main feature of our Foresight approach is the ability to self-update. In other words, the models are continuously and automatically “refreshed” on individual SKUs, thus keeping them up to date with the latest trends and characteristics of the market.
Foresight is an application for the entire Demand Planning process that boasts three main strengths:
- Data Ingestion: a powerful data collection layer that integrates natively with all business systems without the need to add data warehousing / data marts or other middleware
- Machine learning: state-of-the-art forecasting methods able to identify specific patterns in the data and build accurate forecasts over different time scopes even on broad and varied product portfolios
- Workbench: a collaborative tool integrated in the process that helps to create the best final forecast with the sharing of results with key stakeholders.
Foresight’s analytical engine is able to determine the best model for each product or family of products, leveraging backtesting techniques to fully automate the process.
Users can interact with Foresight based on their skill level – openness is a feature of the solution, even if stability, automation and reliability remain the cornerstones of the engine.
After the pure forecasting phase, users can interact with the results by exploiting a multiuser solution for collaboration, in which each of the interested parties can view and refine the forecasts, based on their information and / or business sense.
Aspects that round up the solution are the ability to perform a workflow with different approval levels as well as time series / SKU hierarchy reconciliation.