Computing Baseline in Retail: Data Science Perspective

Computing Baseline in Retail: Data Science Perspective

Retail Analytics is becoming major contributor to data science domain since evolution of e-commerce. Many E-commerce giants like Amazon,  Flipkart deal with promotional activities to expand their business. With people driven market, promotion effectiveness is a means by which retailer’s get to know how good/bad are their products are performing.

Computing Lift is one such metrics which determines the promotion effectiveness. Problem is that there are multiple ways to compute it and on multiple factors. eg. Margin or Sales. Most of the retailers are specifically looking at the difference in promotional and baseline (non-promotional) sales. Promotional sales are quite easy to identify as it’s simply the sales of items during the promotion period like “New-Year Sale”. However, computing baseline is when things start to get tricky for retailers.

So there 2 quick tricks which I will share in this post. Keep in mind this tricks apply for individual products, which can later be scaled to its parent sub-category/category.

Trick 1: Compute Sales prior to promotion.

It is quite a simple approach wherein promotional lift is measured based on sales right before the promotion considering same duration as of the promotional period. But it has one drawback, it doesn’t bring seasonality and trend into consideration.

Note:

  • 1. Seasonality means people’s purchase behavior during certain period like Christmas Sale, New Year Sale etc. In this period sales will bounce high
  •  Trend means certain fixed pattern in sales which occur repeatedly.

Trick 2: Sales average for last a year(53 weeks) eliminating historical promotional periods.

By removing sales during promotional weeks and computing average of remaining week’s sales, baseline can be computed. Although, it considers a year data, it lacks in efficiency when it comes to historical sales behavior. But, it can be effective measure for those products which have been launched 1-2 years prior.

If, particular product has historical data of sales for more than two years [2.1 years /25 months is minimum criteria to compute seasonality and trend in time series], computing baseline becomes easy,  as we can easily capture trend and seasonality and compute Baseline accordingly.

Example:

Once we know baseline for product X1 is 10 units during 1st week of Feb, and if Actual sales were 15 units, we can say it made a Lift by 5 units for that week.

By: Gaurav Chavan

 

Leave a Reply

Your email address will not be published. Required fields are marked *