Executive Summary
To improve demand planning, AMSM launched an initiative to introduce AI and machine learning-driven forecasting across its goods wholesaler portfolio.
The objective was to modernise forecasting processes, improve prediction accuracy, and reduce the financial impact of overstocking, stockouts, and product expiry, while providing a strong foundation for future data-driven and AI/ML innovation.
AMSM wanted to introduce AI forecasting capabilities while maintaining its existing analytics architecture and avoiding a full cloud migration of its data estate.
Working with Eunoia, AMSM implemented a scalable forecasting platform built on Databricks, integrating advanced machine learning capabilities with its existing SQL Server analytics environment.
The platform uses data engineering pipelines, Google’s TimesFM forecasting model, and Databricks MLflow for model monitoring and lifecycle management.
Within three months, AMSM established a robust forecasting framework capable of generating product-level forecasts while maintaining compatibility with its existing analytics infrastructure.
Key Results
Although the forecasting initiative is still evolving, AMSM has already observed several notable improvements.
- Stronger data-driven decision-making
Automated forecasting enables teams to review expected demand across large numbers of products simultaneously, supporting more informed planning. - Proactive stock and procurement planning
Improved visibility into future demand allows procurement teams to anticipate inventory requirements and reduce reactive decision-making. - Improved cross-team alignment
Shared forecasting insights encourage better coordination between sales, procurement, and supply chain teams. - Greater responsiveness to market changes
Continuous model monitoring allows forecasts to adapt to shifts in consumer behaviour and market dynamics.
Most importantly, the initiative has strengthened organisational confidence in AI-supported decision-making, positioning AMSM to expand its use of advanced analytics in the future.
Company Overview
Name: Alf Mizzi & Sons Marketing Group
Industry: FMCG goods wholesaler
Headquarters: Malta
Size: 500+ employees
Alf Mizzi & Sons Marketing Group (AMSM) is one of Malta’s most established FMCG importers and distributors. Founded in 1915, the organisation represents more than 200 local and international brands and employs approximately 500 people.
The group supplies a wide range of retailers and food-service businesses across Malta and Gozo. Managing such a diverse product portfolio requires careful demand planning to ensure the right products are available at the right time while minimising excess inventory and waste.
The Challenge
Demand forecasting in the wholesaler industry presents several operational complexities.
Product demand can fluctuate significantly due to factors such as promotions, seasonality, supply chain disruptions, and changing consumer behaviour. Additionally, historical sales data can be influenced by stockouts, returns, or irregular market events, making accurate forecasting difficult.
AMSM recognised that improving forecast accuracy could help address several business challenges:
- Reducing overstocking and understocking
- Minimising product expiry and write-offs
- Preventing lost sales caused by out-of-stock situations
AMSM already operated a mature SQL Server OLAP-based data warehouse that supported enterprise reporting and analytics. The objective was therefore to introduce AI forecasting capabilities without fully migrating the organisation’s data estate to the cloud.
This required an architecture that could:
- Integrate with the existing data warehouse
- Support large-scale machine learning workloads
- Remain cost-efficient and manageable for internal teams
Another important consideration was organisational adoption. As an early adopter of AI-driven forecasting, AMSM understood that success would require continuous experimentation, stakeholder involvement, and gradual trust-building.
The Solution
To address these challenges, AMSM engaged Eunoia to evaluate multiple architectural approaches and present several alternatives, including projected running costs and implementation considerations. After reviewing these options, AMSM selected Databricks as the platform for data engineering, machine learning, and model management.
Hybrid Data Architecture
Rather than migrating the entire data warehouse to the cloud, Eunoia implemented incremental data pipelines that replicated the required datasets from the on-premise SQL Server environment into Databricks. This hybrid approach allowed AMSM to benefit from advanced AI capabilities while preserving its existing analytics infrastructure.
Machine Learning Development
Using PySpark within Databricks, Eunoia developed data pipelines and trained multiple forecasting models using AMSM’s historical sales data. Various models were tested against agreed accuracy benchmarks, with results carefully explained to business stakeholders to ensure transparency and understanding. Google’s TimesFM foundational model was selected as the production forecasting model due to its strong performance across AMSM’s datasets.
Forecast Integration
Forecast results are written back into the OLAP data warehouse and surfaced through AMSM’s existing BI dashboards. This design ensures business users can review forecasts alongside actual sales figures within familiar reporting tools, making the insights easier to interpret and act upon.
Continuous Model Monitoring
Performance monitoring was built into the solution from the outset. Using Databricks MLflow, AMSM tracks model performance metrics over time and can determine when retraining is required. This ensures forecasting models remain accurate as consumer behaviour and market conditions evolve.
Knowledge Transfer and Collaboration
To accelerate delivery, AMSM outsourced the data engineering and machine learning development to Eunoia while maintaining active collaboration through weekly working sessions.
Eunoia also supported AMSM in building internal capabilities by introducing new technologies and programming languages such as PySpark. This approach ensured that AMSM teams could participate in technical discussions, understand the models being developed, and gradually prepare to take ownership of operational support.
Key Results
Although the forecasting initiative is still evolving, AMSM has already observed several notable improvements.
- Stronger data-driven decision-making
Automated forecasting enables teams to review expected demand across large numbers of products simultaneously, supporting more informed planning. - Proactive stock and procurement planning
Improved visibility into future demand allows procurement teams to anticipate inventory requirements and reduce reactive decision-making. - Improved cross-team alignment
Shared forecasting insights encourage better coordination between sales, procurement, and supply chain teams. - Greater responsiveness to market changes
Continuous model monitoring allows forecasts to adapt to shifts in consumer behaviour and market dynamics.
Most importantly, the initiative has strengthened organisational confidence in AI-supported decision-making, positioning AMSM to expand its use of advanced analytics in the future.
“Working with Eunoia has transformed our approach to forecasting. The collaboration has already delivered measurable improvements, and the transition has been smooth and rewarding”.
– Bryan Saliba
Senior Information Systems Manager
