Recommendations were provided to change the discount offered under each planned promotion, as and when suited, for a month-long period
A large apparel retail company wanted to evaluate the impact of promotions on sales and accordingly optimize its upcoming promotional spends. The presence of multiple simultaneous and back-to-back promotions, and the absence of a process to capture promotional data across departments, posed challenges in developing a forecasting model for promotional sales.
In order to identify areas of improvement for the upcoming promotion calendar and recommend the right discount which maximizes bottom line benefit, a time series model was built. The model predicted the promotion sales.
• Input data collected included, nature of promotions (global, specific), discounts offered, transaction style, department, sales, return volume, and product information such as department, style, category, color, size etc. The promotions were categorized into 3 types: Full Price, Non-Full Price and Entire Purchase promotions across discount categories
• Various regression and times series models were built to forecast sales (units & revenue), identify the baseline and isolate the impact of promotions. Prophet Model was selected to split the sales (units sold) into individual basic Time Series components (seasonality, trend, etc.) and Promotional components. Unit, revenue and margin lift values were calculated in scenarios with and without discounts
• The model helped to forecast promotion sales for the upcoming month based on the cadence planned by the promotion planning team and simulated the promotion sales forecast for all possible discounts
• A recommendation framework was developed to determine the discount to be offered for upcoming promotions by comparing the forecasted margin for simulated discounts against those as per the original promotional plan