Our client, a major international fast-food business, needed a way to predict how well new products would sell at launch. Demand planning for the sale of new menu items is always challenging as new products have no historical track-record that can be used to benchmark performance. With our client, accurate demand planning is absolutely imperative to both avoid the wastage and cost of over-estimating demand, and the opportunity cost of under-estimating demand.
We hypothesised that sales of any product depend primarily on three key factors; Product attributes (ingredients and price), marketing investment dedicated to the product, and date of sale (to account for seasonal trends).
Data was consolidated from disparate sources and a Machine Learning methodology was used to learn from historical data to predict sales quantities from product recipes, marketing and pricing inputs. Various modelling methods were trialled including Linear Regression, Gradient Boosting Machine (GBM) and Random Forest before settling on Random Forest, which was the best performer.
The Advanced Analytics model that we created was able to predict sales quantities with an average error of 18% on new products. Key sales drivers were identified (along with their significance), which provided valueable business insight that improved internal decision making in regard to optimising future products and marketing spend. We also deployed the best predictive model into a user-friendly dashboard that allows business users to flexibly configure any new product scenario and estimate sales for the product. The dashboard planner also delivered confidence estimates around a sales estimate to meet the business’ required confidence levels.