Demand Forecasting for Leading FMCG Conglomerate

Objective:

One of the world’s largest FMCG brands was looking to revolutionise its demand forecasting capabilities and processes. With large amounts of data and many variables affecting the appetite for products, businesses need solutions that boost their operations and allow decision-makers to make informed data-led decisions.
Aria completes demand forecasting by using an ensemble of transfer learning techniques. Using this insight, we can improve accuracy and adaptability for dynamic market fluctuations leading to reduced costs and time.

Challenge:

Before the partnership, their time-series forecasting methods were underperforming and when we started there was not enough data to train and build the model. Also, granular-level demand forecasting is nearly impossible to build using standard statistical techniques that rely on high data assumptions.
Our team of experts accessed information from pre-trained models which could be applied with estimated weight and values from the business’ historical data and product attributes to aid the retailer’s forecasting.

Solution:

The Model Build Pipeline (below) starts with the raw data and all the variables that need to be considered. In this case, there were over 500 variables and so any that were not needed were compressed and combined. This was then passed to the Temporal Fusion Transformers to provide global demand modelling alongside localised intermittency-adjusted models.
The final step used the Shapley Explanation Model, a game theory approach that generates a highly scalable and explainable forecast. This helps to detail the driving factors behind each forecast, allowing users to feel empowered by their decisions.

The Explainable Simulation (below) explores the front-end scenario simulation where users can change the variables to generate predictions, including product choice, price and discount, promotions mechanic, page position and scenario building. This helps users to see which combination of variables provides the optimal result.
These advancements were seamlessly integrated into the client’s internal data science infrastructure to ensure that progress and optimisation were sustained. This includes the generation of timely and automated weekly forecasts to streamline operations and enhance resource allocation.

“Actionable intelligence allows businesses to truly transform their processes; however, with so many factors involved in successful demand forecasting, it can be hard to know where to start. This is where we come in. Our leading solution provides decision-makers with the tools and insight they need to confidently manage their forecasting, helping them to boost their operations as a result.”

Debonil Chowdhury, CEO and Co-founder of Aria