We built a model to predict the price of wheat, to support the trading decisions of the commodity team at Tesco. Simplicity and interpretability were key requirements since the traders need to deeply understand the drivers behind each prediction. We will discuss the end-to-end ML solution and demo the plotly dashboard that we use to present the output to the traders.
The Tesco Commodity team is a team of traders responsible for ensuring that Tesco is protected against sudden price movements of key commodities such as wheat and corn.
To inform the team's trading decisions, in the Tesco Data Science team we built a model to predict wheat price movements. Simplicity and interpretability were key requirements: the traders need to deeply understand the predictions in order to incorporate them in their decisions.
The model makes use of a number of different features, from price-related ones to more fundamental factors, sourced through APIs. The whole pipeline runs on the cloud (Azure) and it is integrated with Outlook and Sharepoint to share the model's predictions. The output of the model is presented through an interactive HTML file, generated with plotly, that displays the model's prediction as well as the main drivers behind the prediction. Every week we review the output with the traders to collect feedback on both the model (e.g. whether new features need to be added) and the data visualisation.