A construction materials company faced a high level of stockouts (inventory shortages) of finished products in its Brazilian operations due to limited warehouse space. As a result, they were unable to maintain an adequate level of safety stock, a situation that was further aggravated by the pandemic due to additional demand volatility.
| Industry | Functional Area / Use Case | Technical Use Case |
| Construction Materials | – Supply Chain – Procurement – Inventory Planning – Demand Forecasting | – Data Capture – Machine Learning – Model Combination |
The Challenges
With no available warehouse space and no historical data explaining the new demand pattern, the client decided to leverage its operational flexibility and create a formal response process to mitigate stockout risk. The company’s directors chose to move directly to inventory forecasting rather than building an entire short-term demand-sensing and operational planning cycle, which would have been costly in both time and money. As a result, they succeeded in maintaining a pragmatic solution that was simple to execute and interpret.
How dhauz helped
dhauz developed a highly scalable data framework associated with a pipeline integrating different information sources. Through this, we were able to host and collect data stored in IT systems and daily descriptive reports. The required data provided full visibility of operations over the previous two years on a daily basis, including inventory levels, stock in transit, production plans and schedules, demand and consumption forecasts, existing orders, stockout history, and much more.
With the data ready, the dhauz team trained a Gradient-Boosting algorithm that uses tree-based learning for the supervised task of classifying whether there would be stock available for a given warehouse and product for the following week. To make the results consistent across regions and product families, the algorithm was trained using different time horizons, comparing accuracy, and identifying new variables that explained deviations in specific areas.
Results
After two months, the initiative launched an algorithm with over 80% accuracy and an F1 score above 0.7 for 87 distinct products across the five largest warehouses by volume.
According to the Director of Planning & Supply Chain Fulfillment: “The AI models created by dhauz delivered results far better than expected in inventory forecasting, and we are now setting up a reaction chain based on these results that will improve our performance on an unprecedented scale.”
Send us a message
[formidable id=2]
