Enterprise platform for dynamic product demand predictions in retail

点击翻译按钮获取中文摘要

During the current Covid-19 pandemic, businesses require more support than ever; in particular, retail organisations are experiencing major disruption. The rapidly evolving conditions and government regulations are making it increasingly hard for retailers to adapt their stock and strategy quickly enough to meet their customer's demand for high-priority products, such as consumer staples and medicines. During a crisis, it is especially important for these organisations to be as efficient as possible. Accurate and timely demand predictions are key to minimising product shortages and maintaining adequate volumes of stock, resulting in a drastic, positive impact on their customers and on the financial robustness of the organisation itself.This project will expand the causaLens platform. In its current form, it is the leading time-series platform with unique technology that leverages the latest research into causality to autonomously build dynamic models that adapt to new data. The delivered product will contain technology specifically tailored for consumer demand applications, including the capability to load new data on a fixed schedule, to discover and update models in an online fashion, to support small data scenarios, and to provide predictions across multiple locations.The main innovation developed will be the ability of the deployed machine learning models to update their parameters as new data arrives, immediately reflecting changes in the environment and providing more accurate demand predictions. In a time of crisis, the assumptions about the drivers of a product's demand are likely to change. Demand is no longer able to be reliably predicted using historical data from a time of normal business operations. Therefore, the system will need to operate in a small data scenario as only recent data will provide the greatest benefit in providing accurate predictions. This will be achieved by implementing the latest research into causal algorithms, which have the ability to leverage causal relationships discovered in the data to learn from vastly fewer points than traditional machine learning algorithms. Additionally, the system will allow the user to combine data from various retail locations to predict demand at each one. A system that is capable of learning this new environment and providing accurate predictions in this way has never been used before.