The co-creation initiative between Fiera Milano’s Data & AI Office, the operational arm, Nolostand and Elevate, evolved a rigid inventory tracking process into an agentic ecosystem. This systemic shift redefined how stock levels are monitored, forecasted, and utilized. The result is immediate transparency within a high-volume logistic environment, providing tools that bridge the gap between physical assets and digital intelligence.

A physical mandate, a digital strategy - scenario

Fiera Milano operates one of the largest exhibition districts in the world. Its parent company, Nolostand, focuses on set-up services for exhibitions, such as: design, construction, and fitting of thousands of square meters of exhibition space. This operational arm manages a massive inventory - roughly 900,000 records tracking everything from structural components to furniture. The mandate is dual: Fiera Milano provides the strategic digital infrastructure, while Nolostand ensures the physical readiness of events ranging from the Milan Furniture Fair to the upcoming Winter Olympics.

The friction of legacy - challenge

The challenge was not a lack of data, but a lack of accessibility. The inventory information was trapped within a legacy ERP system and extracted via massive, cumbersome Excel files.

The existing workflow contained specific systemic vulnerabilities:

• Manual dependency and bottlenecks: The analysis relied heavily on specific individuals who understood the raw data structure. If a strategic question arose ('How many chairs of this type do we have for December?'), it triggered a manual, time-intensive extraction process.
• Reactive rather than predictive: Because the effort to retrieve current status was so high, the system was used primarily to check past records rather than to forecast future needs or optimize stock rotation.
• Semantic opacity: The raw data lacked business context. A column header in the database did not necessarily match the natural language a business user would use to describe an item, creating a translation gap that only technical staff could bridge.

The objective - expected outcome

The goal was to transition from this high-friction, manual process to an intelligent layer that could:

• Democratize access by allowing non-technical users to query the database using natural language.
• Accelerate decision-making by automating the retrieval and visualization of stock trends.
• Enhance predictive capability by moving from static reporting to AI-assisted forecasting.
• Preserve legacy infrastructure by building an intelligent layer on top of existing systems, avoiding a costly 'rip and replace' scenario. Retaining the ERP was a strategic application of the 'maximum value, minimal disruption' principle. AI acts as a 'resilience layer,' extending the lifespan of legacy investments and allowing the company to self-fund next-generation systems with today’s savings.

The Agentic Data Layer - the solution

The project was executed through a 'laboratory' approach - a rapid, agile Proof of Concept (PoC) delivered in under three months. Working with the Google Cloud ecosystem, the team developed a platform of Agentic AI tailored to the specific lexicon of exhibition logistics.

The solution is an intelligent system structured around four specific agents:

• Forecasting Agent: Analyzes historical data to predict stock requirements based on square footage sold for upcoming events.
• Substitution Agent: Identifies available alternative products when primary stock is depleted, ensuring operational continuity.
• Stock Control Agent: Provides instant visibility into current inventory levels across different categories.
• Flexible Reporting Agent: Allows users to ask complex questions ('Show me the consumption trend of wood panels in Q4') and receive immediate visual graphs and data tables.

Efficiency, autonomy, and semantic alignment - the outcome

Delivered in 3 months by a joint team of 5 from Fiera Milano, Nolostand and Elevate, the project drove a shift from manual data wrangling to strategic data interaction: From extraction to interaction The 'Excel bottleneck' has been removed. Users no longer wait for files to open; they ask questions. The system interprets the intent, retrieves the data from the Data Lake, and presents the answer. This has freed up the R&D and operational teams to focus on planning rather than data cleaning. Semantic data restructuring: A crucial, invisible outcome was the semantic mapping of the warehouse. The team didn't just dump data into an AI; they tagged and structured the metadata so the AI could understand the difference between a 'chair' and a 'stool' in the context of a fair. This effectively captured the tacit knowledge of senior employees and embedded it into the system. Enhanced governance and scalability The system provides a blueprint for broader application. By proving that AI can bridge the gap between a legacy ERP and a modern user interface, Fiera Milano has established a replicable model. The architecture is designed to scale, with plans to expand into a broader 'Data Catalog' agent that will democratize data governance across the entire organization.

Lessons learned: a model for pragmatic AI

The project established an operational model for introducing GenAI in traditional sectors.

Co-designing for semantic precision: The most critical phase was not coding, but defining the 'prompt catalog' and data tags with the business users. This ensured that when a user asked for 'wood,' the system retrieved the correct material codes. Collaboration was the method for ensuring accuracy., Agility as a risk mitigation strategy: By choosing a 3-month PoC timeline, the organization minimized risk and maximized focus. This 'laser' approach allowed them to validate the technology on a specific use case (warehouse) before considering wider adoption, turning skepticism into advocacy.

Key insights for your organization

The experience with Fiera Milano and Nolostand provides a replicable roadmap:

• Democratization requires translation: To make data accessible to everyone, you must build a layer that translates technical database logic into business language.
• Don't wait for perfect infrastructure: You don't need to replace your legacy ERP to get modern insights. An intelligent agentic layer can act as the bridge, unlocking value from old systems immediately.
• Start with high-friction tasks: Identify where your smartest people are doing the most boring work (like opening zip files). That is where AI delivers immediate ROI.
• Structure is a prerequisite for intelligence: GenAI is not magic. It requires a well-governed, semantically tagged data foundation to function reliably in a corporate environment.
• Evolution is iterative: Start with a 'laboratory' mindset. Prove the value in one vertical (inventory), then scale the logic to others (finance, HR, governance).