Improved Resource Allocation Based On Footfall Forecasting In Stores Using Neural Networks

Improved Resource Allocation Based
On Footfall Forecasting In Stores
Using Neural Networks

The Challenge

A leading electronics retailer in the Middle East invests in various forms of online advertising to increase footfall in their stores. The management wants to understand the impact of the different investments. They also want to forecast what footfall they can expect on a day, given the marketing spend, so that they can staff the stores with the appropriate level of staffing to ensure good customer service, forecast sales and plan in-store promotions.

Hypothesis

  • Footfall can be forecasted based on weather, marketing spend and demographics within the catchment of a store.
  • Footfall can be attributed to marketing campaigns.
  • Customer demographics in store catchment can be used to predict successful in-store promotions.

Execution

  • Facebook/Instagram data scraped to gain a measure of the level of engagement on a daily basis.
  • Web traffic, app traffic and Facebook/Instagram engagement was incorporated with a lag in predicting footfall.
  • Media spend was translated to expected reach.
  • Neural forecasting models forecasting models, ARIMA and Holt-Winters were developed.
  • An ensemble of all base forecast models was trained.

Outcomes

  • 17% saving in staff cost on shop floor.
  • The forecasting model was deployed in production and used to optimize marketing spend and right size the deployment of staff on the shop floor.
  • Promotion recommendation based on the spend and engagement online.

Project Highlights

14%

increased efficiency of marketing campaigns