Replenishment System Optimized by ThinkCol’s AI Capabilities

ThinkCol forecasted the customer demand for the supermarket branch of a well-known comprehensive trading company in Hong Kong. Prediction models as well as visualization dashboards were created to optimize replenishment amounts for various SKUs.

A key challenge faced was ensuring accuracy. ThinkCol performed feature engineering by aggregating SKU historical sales, category sales, global sales across all stores, days since last sales, launch dates, seasonality, and external data, such as weather and holidays. These features were then fed into different models to make accurate predictions for each grocery item. Hyperparameter tuning and error analysis were done to ensure the validity of the forecast results. The models are also auto-retrained once new data are available to keep the results updated. With balanced consideration for the demand forecast, inventory stock, and supplier ordering restrictions, ThinkCol created an algorithm to optimize the final replenishment decisions - whether to restock certain groceries.

ThinkCol also helped to visualize the model’s results through a Power BI dashboard so as to assimilate it into the retail replenishment process and make the forecast more readable, user friendly for marketing and operation staff. On the dashboard, actual sales, forecasted sales, and error rate were all shown.

With ThinkCol’s assistance, this demand forecast project is a precedent for the organization’s future internal machine learning projects. ThinkCol is creating a real-time pipeline according to MLOps best practices to refresh the models with new data and to generate new weekly predictions automatically.