Miniconf is a company specializing in the design and production of children's clothing. Like many manufacturing businesses, it must plan production well in advance and coordinate its supply chain effectively by defining quantities and timing for supplier orders. In this context, having reliable sales forecasts is essential to optimize production, reduce resource waste, and promote a more sustainable production model. The project, developed in collaboration with AIRIC, aimed to assess the applicability of Artificial Intelligence and Machine Learning techniques to automate and accelerate the demand forecasting process, supporting the company's experts in forecasting activities without replacing their decision-making role. The main challenge was to develop a predictive model capable of estimating sales for new products in the campaign, even in the absence of product-specific historical data, by leveraging information from previous campaigns. The model was also designed to generate high-granularity forecasts, down to the single combination of item, color, and size, progressively updating its estimates based on the new sales data collected throughout the campaign.
Context Analysis and Data Collection
The first phase focused on acquiring and analyzing historical company data in order to understand its structure, quality, and characteristics. The activity was carried out in close collaboration with the Miniconf team, combining domain knowledge with analysis of the available data so as to identify the most relevant information for the development of the predictive model.
Deliverables:
- Cleaned and validated dataset.
- Presentation of the data structure and characteristics.
Definition of the Framework and Pre-processing Pipeline
In this phase, the input and output of the forecasting system were defined along with the pre-processing pipeline needed to prepare the data. The goal was to build a structured and reproducible process, capable of progressively integrating new available data and keeping forecasts up to date throughout the commercial campaign.
Development and Validation of the Forecasting Model
Once the evaluation metrics had been defined, different types of predictive models and various feature combinations were tested with the goal of identifying the most accurate and flexible solution. Performance was assessed through backtesting on historical data. The main challenge in this phase was creating fine-grained forecasts capable of estimating demand for specific items in different color and size variants, supporting operational decisions with a very high level of detail.
Deliverables:
- Demo of the developed model.
- Technical documentation and solution code.
Model Refinement and Live Testing
In the final phase, the model was validated in a real operational setting, alongside the forecasting process already used by Miniconf. The comparison between the forecasts generated by the Artificial Intelligence model and those produced using the traditional method made it possible to identify the solution's strengths and weaknesses. Based on the results obtained, model refinement activities were introduced through feature engineering techniques and the integration of domain knowledge provided by company experts. Ongoing collaboration with the Miniconf team proved to be a key element in progressively improving the system's performance and strengthening its ability to update forecasts in response to new data collected as the campaign evolved.
The project showed how historical company data can be a strategic asset for developing predictive models that support production planning. The solution developed makes it possible to generate forecasts at any point during the commercial campaign, reducing the effort required for forecasting activities and supporting decision-makers with timely, fine-grained estimates. The approach developed has the potential to optimize production processes and contribute to a more efficient management of the forecasting process.
Co-Director
AIRIC
Nicola
Gatti
Project Manager | Senior AI Research Engineer
AIRIC
Tommaso
Bianchi
AI Research Scientist
AIRIC
Tomaso
Castellani
AI Research Scientist
Gianmarco
Genalti
Senior Business Analyst
Miniconf
Francesca
Regina
Business Analyst
Miniconf
Michele
Scipioni
Digital and Customer Marketin
Miniconf
Lara
Loddi
General Management
Miniconf
Leonardo
Basagni
Business Analyst Senior Manager and Controller
Miniconf
Alessio
Gori
