Real time Demand Sensing & Price Optimization for one of the largest Innerwear Brands

Industry: Apparel

Introduction

One of the largest Innerwear Brand of USA has 3 sales channels – Own Store, Multi brand Outlets (like Walmart, Costco etc) and E-commerce. MBO contributes 86% of overall sales.

As MBO shares POS data in 3-4 month lag and not daily basis, it has been very difficult for our Client to understand Weekly/Daily demand in Category/SKU level. This resulted a significant mismatch in Inventory level. Out of Stock & Overstock became a regular phenomenon in different MBOs, leading to both opportunity loss and brand dilution.

Client’s in house analytics team used to produce mostly descriptive analysis and they wanted us to build an almost real time demand sensing & price optimization engine, based on historical facts & other external factors

Project Journey

Objectives

To start with, Abzooba has clearly defined the following objectives for this project –

  • Weekly demand forecasting for next 1 year at both category and SKU level
  • Sales forecasting with segmentation of promo and non-promo
  • Weekly inventory forecasting at Store- SKU level
  • Price impact analysis and Price optimization modelling implemented at SKU level

Solution overview

In 1st phase, Abzooba’s Data management team and Client’s IT team collaborated to design the DataMart for proposed analytical solution. This involved –

  • Identified of relevant data from the current client BI cube for different kind of modelling
  • Modified the Data warehouse system according to data volume and nature of the data
  • Cleaned the data to remove the noise
  • Prepared data according to model requirement

Big Data Technologies like Hive & Sqoop have been used to manage the data transaction

In 2nd phase, Abzooba started to create the Machine learning models which incorporates all the nuances of data, run different possible models continuously and bring out relevant insights about demand and price. We have executed following steps to build this –

  • Identified relevant business parameters to be incorporated in the model
  • To start with, selected limited number of Store, Category and SKUs for model development
  • Developed of Time series based regression model to define relation between sales and other business parameters (Trend, Seasonality, Promotion, competition etc. are captured)
  • Reiterated of the whole process until certain accuracy is achieved. Added other parameters to improve the accuracy
  • Model evaluation was jointly done by client stakeholders and Abzooba analytics team
  • Ran the models for all other Stores, Category and SKUs
  • Deployed this model on a private cloud (client’s AWS instance)
  • Redeployed these models where output would be automatically recreated with the latest data

Once Model is developed, we have built a connector with Xpresso.VIZ to depict the model output/operational reports. Xpresso.VIZ has following features for this project –

  • Dynamic reporting with ability to change over time
  • Flexible enough to incorporate new metrics quickly
  • Drill down option available into geographical dimensions (region/state/city/store), as well as category dimensions (category/sub-category/SKU)

As stated above, creating a near accurate Demand sensing engine is all about rigorous process of Data management, running multiple algorithms to find out most suitable one and represent them in a most meaningful way to the business users. Following is a high level snapshot –

Business Impact

  • 31 % reduction in loss of inventory from previous year
  • For 5 selected categories, Out-of-Stock scenario has been reduced by 13% than previous year
  • Average accuracy for demand prediction is 89% for selected categories