Use Cases

How best to tailor stock levels for revenue optimization?

Case Study
Topic:Demand Prediction

Business Context
Retailers lose more than $1 trillion a year from inadequate inventory control for out-of-stock and overstocked merchandise, according to an IHL Group research report.

An ecommerce retail intelligence provider was looking to predict future demand for specific products that its customers sell.

Data Science question

How best to tailor stock levels for optimal pricing? The challenges retailers face include:

  • Should they keep products in stock until the price goes up or should they part with particular products now to make way for items that might sell better?
  • How does the retailer prioritize each item in stock, pinpointing when to increase price to slow sales?
  • Which products are trending as the most popular now?
  • Which product categories are excelling now, which would open the door for multiple item cross-selling and up-selling?

Solution: Automatic predictive modelling
With the amount of detail that feeds into each pricing decision, Firefly intelligent preprocessing was able to incorporate key decision-making factors into selecting the best methods at each stage of preprocessing features, including engineering, embedding and stacking. Key retail factors were taken into account, such as seller rankings, daily prices per product, number of items sold and competitor pricing. Firefly concurrently trained many different algorithms and created hundreds of models. The final ensemble, the golden model, is combination of complementary models to provide the best coverage.

Results: Reduce excess inventory, increase revenue
With accurate predictive models, retailers could precisely pinpoint which products to promote and how to price them at each specific time. Lowering stock and returns maximizes their return on investment.