An international pharmaceutical company operating in several dozen countries faced challenges in effective production planning. Our customer used 100% of its production capacity, making the choice of products they manufactured critical for financial success. Long delivery times for products’ components, sometimes stretching up to 6 months, posed an additional difficulty.
Upgrading Production Planning Accuracy
We recommended an AI-based machine learning engine to predict sales for up to six months ahead, as to plan production effectively. In the first stage of the project, our team spent a considerable amount of time on understanding the business and data from our customer. The customer, a wholesaler, provided sales data that was not in line with the actual demand for products in the market. Numerous data types and proxies were indicated and then collected.
In the second stage, our team concentrated on further analysis of the data, development, and testing of models. Several models were shortlisted and presented to the customer. Models operated on various levels of generality such as group of products, type within a group, and individual SKUs.
The results provided by the models used were visualized in PowerBI, providing product managers support in decision making on production order placements. Further, we worked on developing an active monitoring system. The proposed system is meant to confront predicted values with those provided by product managers and alert group managers, if inconsistencies are significant.
Performance of the model was dependent on the level of generality it operated on, as data at lower levels of generality were more volatile. The metric used to evaluate performance of the model was its accuracy versus accuracy of product managers in sales prediction. On average, in 7 out of 10 cases (SKUs) performance of the model was higher or equal to that of product managers. On higher levels of generality, performance of the model was even higher compared to that of the product managers.
Implementation of the models provided instant support for product managers in production planning and improved accuracy of predictions by 9%.
accuracy of production planning
lower risk of significant errors in predictions
instant support for product managers