Previous Case Study: Restos Plaisirs
Next Case Study: iBreastExam



SBI is a Canadian family-owned business that designs, builds and markets residential heating products that are sold throughout the world. The reputation of their products goes well beyond our borders. In addition to Canada, the United States and Europe, they are also available in Brazil, Colombia, South Africa, and Australia through a distribution network that is constantly growing.

SBI previously relied exclusively on its sales team’s forecasts to estimate the organization’s future sales each year. This method obviously involves many biases since forecasts are often based on salespeople’s instincts or on incomplete or even inaccurate information. Hence, SBI wanted to implement an AI model that would help reduce the importance given to the sales team’s predictions, especially since enough data was available for the algorithm to generate the predictions itself.



Products on which sales forecasts are generated


Points of sale



SBI’s objective was to stop relying exclusively on the forecasts of its sales team and to be able to predict the quantity of products it needed to manufacture for the inventory more accurately. Therefore, the challenge was to identify the variables that had a real impact on heating appliance sales (e.g., unemployment rate, house construction starts) as well as to find the right combination of parameters that would train an AI model to make predictions that would be as good as, if not better than, those made by the sales department.



Vooban developed a sales prediction algorithm based on machine learning techniques. The model was deployed on the Microsoft Azure cloud platform and runs automatically every week. It makes two types of predictions simultaneously: the number of upcoming orders and the number of transactions that will be completed successfully.

Each time the code runs, the algorithm retrieves the most recent data related to the products being analyzed from various internal sources using a database. The sales forecasting model is also trained on a range of external data, such as from Statistics Canada, that has a significant impact on fireplace sales (e.g., weather). In addition, SBI has access to an interactive dashboard that allows it to visualize the sales forecasts issued by the algorithm, which it can then compare with its sales team’s forecasts.

To ensure the reliability of the AI model, we performed a series of retrospective tests. During these tests, we first trained the model on data from previous years and then tested it with another year for which actual sales data was available. This allowed a comparison between the model’s predictions and actual sales for a large selection of products while ensuring that the model’s predictions were reliable.

The implementation of an AI model made all the difference for SBI. Before that, their team could only predict their yearly sales with about 50% accuracy. Our algorithm now manages to achieve an 85% accuracy rate. This increase can also be seen on the short term. If the predictions are made a month in advance, the model shows 93% accuracy, whereas the sales team had a rate of 85%.