Demand Forecasting of Merchandise for a eCommerce Company

Project Description

Forecast the demand using prior sales data on various categories e.g. sporting goods, baby and toddlers etc.

The Sales data contains:

  • Sales past data for variant SKUs
  • Contains different number of sold SKUs in past
  • Contains inventory product id
  • Contains SKU id
  • Contains color, brand, discount, dimensions columns to calculate color index, brand index, discount band

Demand calculation:

Demand is forecasted by taking sales value of product, counts of product sold earlier, brand index, discount band, price band, topic values (deal text data), seasonal index and color index. Most of these values are a value of strings but we converted them to numbers by using index using moving averages (in case we have sales data for all the 52 weeks for that brand or color).

Topic modeling and forecasting and validation:
Topic modeling is done by using text mining using tools like R, Pyhton and SAS using the text information from deal text data, like color, brand, shape, size etc. The solution looks at clustering similar words into different cluster (we call them topic1, topic2 etc), this is done by latent drichlette allocation.

The prediction window is for 7-15 days, the prediction monitored everyday using web based tableu account and the model is refreshed in every 3 months.

Some Benefits of Forecasting:
  • The benefit of having a forecast model is to ensure higher customer satisfaction in terms of delivery time and availability of product.
  • The forecast model also helps the company to ensure to have to correct amount of inventory at disposal.
  • The liquidation cost is also reduced by a large amount and hence saves lot of cost.