One of the world’s leading luxury car manufacturers.
The manufacturer planned its global upstream production processes and supply chain requirements based on forecasts provided by local sales teams. Thanks to years of industry experience, these forecasts were reasonably accurate – but calculating and delivering these predictions was also a significant drain on the sales teams’ time.
In order to free the sales team for more valuable, customer-facing tasks, the brand asked Faculty to help it automate the car sales forecasting process with AI.
Recognising the reasonable accuracy of past forecasts, we focused on finding an approach that combined the best of human expertise with the accuracy and power of data science; we wanted to go beyond automating the process to deliver something that improved upon it.
Most demand forecasting tools struggle to predict sales for products that generate little sales data, such as those with relatively low sales volumes or short sales histories in a particular market.
To overcome this obstacle, our model ‘fills in’ gaps in the data by learning the relationships between different products and geographies. For example, the model can recognise that two different geographies usually follow similar demand patterns, or if rising sales in one product usually correlates to decreased sales of another product.This allows the brand to make extremely granular predictions about individual products, instead of just making predictions at the category level.
This approach also allowed us to build industry expertise directly into the model. For example, if sales teams knew that sales volumes for certain models always increased in a certain month, we could train the model to consider this in its calculations, instead of the model having to learn this information over time.
Even the best models go unused if they’re not closely integrated into daily operations for sales teams, so we built the model into a dashboard that compares actual sales with target sales and forecast sales. This allowed the global sales team to have a clear understanding of their performance over time, as well as clear expectation of progress toward annual targets.
The resulting model combines the best of data science – the ability to rapidly identify patterns often missed by humans – with the impressive industry experience held by the brands’ sales teams.
When implemented, the model improved overall forecasts of product sales by 50% and provided accurate visibility six months into the future – double the previous quarterly forecasts.
Overall, the model has allowed the brand to save sales teams significant amounts of time, but also transformed the way the brand plans supply chain operations and production.