80-90% Sales Forecasting Accuracy for a Leading Agro Producer

Goals: Our client aimed to improve production planning and stock management based on the accurate prediction of sales volumes for each customer.

Solution: Infopulse data engineers developed a Machine Learning-based sales forecasting solution with an average accuracy of 80-90% for weekly and monthly forecasts.

Benefits: Optimizes production planning, improves warehouse stocks, space management, and logistics, and significantly accelerates time for sales forecasts from days to a few minutes.

Services delivered: Innovation Services, Intelligent Business, Intelligent Automation, Smart Insights, AI/ML services, Predictive Analytics, BI & Data Analytics, and Advanced Analytics.

Infopulse data engineers developed a sales forecasting solution based on the Prophet model for weekly predictions and the XGBoost for monthly predictions.

The ML-based sales forecasting models implemented by Infopulse introduced a multitude of benefits for our client: 

• Better sales planning: high sales forecast accuracy helps avoid over- or underproduction and minimizes stockout risks.
• Streamlining production and sales processes: accurate sales predictions help optimize the required production volumes and simplify logistics.
• Optimized warehouse stocks and space: minimized remainders, waste, and storage.
• Improving the level of service (if not out-of-stock): ensuring the delivery of required product volumes to customers on time.
• More efficient use of capital and higher profits: as the production is fully coordinated with accurate sales predictions, the resource waste is minimized.
• Faster forecasts: compared with manual sales forecasting, our solution reduced the time for obtaining a sales forecast from a few days to a few minutes.
• Better accuracy of predictions due to automated data processing, choosing a more precise model. The prediction accuracy also improves with each new dataset processed by our models.

Technical Details

Infopulse data scientists began the development of the sales forecasting solution with an in-depth analysis of the data set received from the client. We selected a standard but effective technology approach, using Python and Jupyter Notebook for prediction model development. To save development time and re-ensure the results, we also used Azure ML to train our models with AutoML.

Upon analyzing the obtained results, the team selected the best-performing models, built the analytical reports, and presented results using Microsoft Power BI:

• Based on the model training results, Infopulse data scientists chose the Prophet model for weekly forecasting. 4 of 28 predicted cases showed up to a 90% accuracy, while the remaining 24 cases had 80-89% accuracy.
• As for the monthly forecasting, the best model is XGBoost, producing up to 90% accuracy for 19 of 28 predicted cases, while 4 cases had 80-89% accuracy, and the remaining 5 had an accuracy of less than 80%.