Adidas Markdown Optimization: Two-Year Success with 5-10% Revenue Growth and 12% Profit Uplift per Promotional Campaigns

Key Benefits

  • +10% average selling price
  • +12% standard margin
  • -15% average markdown
  • Demand prediction accuracy 98%
  • 450% ROI for the first year

Brief overview

Adidas continuously investigates new possibilities to utilize the power of ML and AI to stay ahead of the competition. Adidas recognized that its markdown strategies were suboptimal, leading to significant untapped potential in terms of revenue and profit margins. To address this issue, Adidas closely collaborated with acmetric to develop and implement a Markdown Optimization solution. This solution has demonstrated great success in major markets like the US, China, and Europe.

The Markdown Optimization solution empowers Pricing & Trading teams to efficiently identify article-level price elasticities, quickly evaluate thousands of possible markdown strategies and determine the most advantageous markdowns at a country, day, and article level. By automating these processes, Adidas can maximize its sales potential and enhance its profitability in a more systematic and strategic manner.

  • Our solution supports Markdown pricing decisions in 20+ counties 
  • Price elasticity and demand forecast calculated per article, campaign, location and day 
  • The solution is fully automated and handed over to the internal Adidas team.

“During our two-year collaboration ACMetric has provided exceptional results for the Trading Capabilities team at Adidas. ACMetric has become a strategic partner who supports us in key decisions around Forecasting, Pricing, and other data science related questions. 

I can highly recommend working with their team’s top data and decision scientists. They’ll help you, your business, your departments, and people to accelerate achieving your business objectives with their action-oriented and data-driven mindset.” 

– David Westera, Director Trading Capabilities

The problem

Adidas’ Pricing team operates in a very complex environment. Many decisions need to be made as 10.000+ products in 20+ countries need a daily markdown. This implies thousands of pricing decisions per day per person if optimisation is done manually. This was simply infeasible, and hence without data science, optimisation needed to be done on an aggregated level. This heavily restricts the optimisation potential and leaves a lot of value on the table.

Price elasticities differ per country, campaign and product, and hence there is a large collection of different elasticities that have to be measured. Measuring historic revenue uplifts over similar articles didn’t work particularly well, as it was very difficult to isolate various external effects on the price responses, such as economic situation, inflation, other campaigns, inventory limitations etc.

Another notable challenge involved selecting the appropriate markdown from an unbelievably broad collection of different pricing strategies. Exponential growth resulting from thousands of products over tens of countries, tens of discounts and tens of days result in 101.000.000 possible pricing strategies. It was practically not possible to manually find the optimal pricing strategy in this vast and complex space.

Our solution

acmetric Markdown Optimization solution allows Adidas to estimate the Price Elasticities of articles and predict potential sales for the period in question. Having a way to accurately know in advance what uplift in sales giving a certain discount will allow implementing on top of it an optimisation logic. Note, that this logic can be as simple or as complex as the user needs: it can be rule-based, or involve an involved optimization algorithm.
For Adidas, we have created a Causal Forecasting Model to solve this problem. This solution allows to combine two approaches:

  1. Structural Econometrics model to evaluate the effect that price has on sales. Our methods allow us to consider all types of outside effects: from the world economy to sporting events and seasonalities.
  2. Predictive Machine Learning model. Having separated the effects of price into a separate structural model allows us to leverage advanced Machine Learning tools available in Data Science. A wide arrange of models can be used for this step: from Structural to Machine Learning models, to even Neural Networks.

Such a two-step approach allowed us to use the better of two worlds: Econometrics theory, as well as large-scale Machine Learning algorithms.


Why partner with acmetric

Higher Margins
Many products were selling out too early in the campaign due to the discount split logic because market teams wanted to make their targets. Price elasticities allowed for advanced article level decision making

Better Inventory Control
Prices changes provide a new, flexible and relatively quick lever that allows adidas to improve matching stock with demand by making it easier to react to observed in-season demand.

Improved Teamwork
Price elasticities and causal forecasting strongly outperform other forecasting methodologies and facilitated improved coordination between finance, supply chain optimization and marketing.

Future Plans

The successful implementation of acmetric Markdown Optimization solution creates an opportunity for Adidas to enhance more comprehensive applications, such as optimizing full-season pricing decisions throughout the season in the key markets in e-commerce and in-store channels. The Full-season Pricing solution will support the data-driven decision-making process for preseason, in-season and postseason pricing strategies. 

The solution’s crucial ability is to recognise the effects of pricing decisions across different stages. For instance, higher preseason prices enable greater promotional markdowns, and increased outlet pricing permits more moderate in-season price cuts. This integrated approach will increase Adidas’s pricing strategy’s overall efficiency and profitability in a very competitive market.