Logo_w_o_bg.png
Menu Board

Dynamic pricing

Our dynamic pricing solution is currently implemented for one of the largest international sportswear manufacturers. It will be realised within the client's in-season trading platform for Europe and the USA, which has had over 10 million expected incremental net sales in 2020. The product is expected to increase sales and reduce dependency on sales campaigns.

Want to read more about how we did this and what we can do for you? Scroll down for more details or contact us!

The case

What is the optimal price to set for an e-commerce article? One of the largest international sportswear manufacturers has to answer this question thousands of times every season. The initial price of an article is typically motivated by value, competitors and the underlying costs behind the article. However, how do costumers react  when you change the price to a lower or higher amount? Would sales change? If yes, then by how much? Enough to compensate the changing margin? What articles should you consider changing the prices for if you want to optimize profit, revenue, stock management or a combination of the three? We built a solution that suggests prices and helped our client to find answers to these four questions.

Solution

Together with our client we developed an econometric model based on historical transaction data to infer the price elasticity of demand: the effect of price on the number of sold articles. To ensure accurate estimation of price elasticity, a large variety of additional data sources that contain explanatory factors behind sales were included in the model. Seasonal data allowed us to disentangle yearly, monthly, weekly and even weekday effects from price specific effects. Similarly, marketing and campaign data made it possible to differentiate the impact of a chosen price from end-of-season or Black Friday sales hypes. Other relevant data sources included for comparable reasons are macro-economic data, stock and inventory data, website traffic data and more. Not all articles had enough historical data to properly estimate their price elasticity. Therefore, we used article-specific information, such as product category, gender and colour, to cluster articles. This allowed us to provide insights on articles with little available data by relying on similar articles with a lot of data.

 

Interesting to note is that estimation of price elasticities should not rely on minimisation of prediction errors through some machine learning methods. Price elasticities are obtained as functions from the parameters of an economic model; hence one must focus on minimising the parameter estimate errors. This implies that a generic machine learning algorithm is not well suited for the problem. What does provide insights is an econometric modelling approach based on careful considerations regarding chosen variables and their relationships to sales. Please have a look at our blogpost for a more detailed and technical explanation of the intricate differences between the two approaches and why differentiating between them is important.

Impact

The solution is currently being implemented in our client’s in-season trading platform for Europe and the USA, which has had over 10 million expected incremental net sales in 2020. It is expected to increase sales and reduce dependency on sales campaigns. The solution is general in terms of business applications and flexible, allowing for additional restrictions on price, stock and predicted sales. The resulting tool is user-friendly and it is planned to be used in different formats, ranging from almost fully automated to a more supplementary role.