NEXTAIL

A discount was two different decisions

At IE Exponential Learning in Madrid, my four-person team worked with Nextail footwear data covering 4.2 million units from June 2019 to February 2020. We wanted to know when a first markdown would change how a product sold. A blanket recommendation to discount would have ignored the possibility that some shoes already sold at full price and others needed a different list price altogether.

We grouped products by price, discount, sales and season using hierarchical clustering. The groups gave us a reason to treat products differently. In the inelastic group, 77% of units sold at full price. In the elastic group, 84% sold on discount.

Make the call at product level

The groups informed the pricing strategy. For an individual product, we built a random-forest classifier to predict whether sales would rise after the first discount. We compared the seven days before and after the price change. The classifier reached 75% accuracy in the programme evaluation.

Our demo-day proposal paired that prediction with the elasticity groups. Hold the discount where full-price sales already looked strong. Revisit the list price where a markdown was unlikely to solve weak demand. We delivered the analysis and model as a team project. It ended at programme evaluation, before any production test of revenue impact.