Sales Forecasting Based On Products Images
A footwear wholesale distributor is supplying a large number of shops all over the country. To maximize the revenue it is crucial to order the right amount of products considering its styles and sizes. The orders need to be placed 6 months ahead and every year available styles vary. The distributor wanted a system that will help him in choosing styles and forecast the demand for each of them based on the historical data about sales in the company’s regions. Also, new models, not seen before, had to be considered.
- Create a forecasting model that can predict the footwear sales based on the historical records.
- Develop a solution that will be able to estimate the future sales of the brand new product in the distributor portfolio.
- Make a forecast for products with no historical record.
- Precise forecasting for distant time-span – 6 months ahead.
We divided our solution into two parts.
The forecasting models that were using the sales records of the products as well as the other, external features to predict the accumulated, monthly demand for 6 months ahead considering sales trends and seasonal fluctuations.
The algorithm that may be used to find similarities between the brand new products and the products with known historical data. We decided to use a Convolutional Neural Network to extract information from products’ images and find the footwear with comparable appearance features and historical sales data and use the forecasting model prepared for it to make an estimation of the new product’s demand.
- Revenue is maximized thanks to optimized order.
- Storage space and is-stores displays are used optimally.
- Reduction of unprofitable and time-consuming sales at the end of the season.
- New styles can be ordered with a high chance of market success.
- Due to the increasing amount of data, every year predictions will be more accurate than before.