Bringing AI technology into core business processes is one of the key challenges for its large-scale industry adoption.
When looking for possible applications of automated decision-making based on machine learning, the retail industry is an ideal example. It is characterized by fine margins and retailers face raised customer expectations as well as extreme pressure from innovative competitors. At the same time, the established players lag in tech adoption. Two important subjects in the retail business that benefit considerably from AI applications are replenishment and price optimization. These deal with the questions of optimal order quantities and price changes. For replenishment, optimal can mean few out-of-stock situations or little waste of perishable products. For pricing, it can be maximization of revenue or profit via the price elasticity of demand.
As these decisions are easier to take if you know what will happen in the future, accurate demand forecasts are of great support. In machine learning terminology, the prediction of demand is a regression task that can be approached by means of supervised learning. However, there are challenges to overcome. For example, most relevant data is time-tagged so machine learning solutions need to be able to include time series analyses into the model. Rare events such as holidays, weather effects, or exceptional promotions require special treatment as well. In addition, to consequently optimize decisions based on estimated demand, it is essential to predict full probability density functions, not just mere point estimators. And eventually, one also needs to find a way to measure success and quality of the decisions made.
Most retailers make millions of ordering and pricing decisions daily. Due to the sheer amount of data that needs to be handled, it is simply impossible for a store manager to do this accurately every day. Therefore, automation is crucial. And machine learning assisted by distributed computing is the tool of choice to calculate predictions and automate decisions at scale.