Italian retail operator Multicedi has dramatically accelerated its data processing by migrating to Microsoft Fabric, slashing ETL pipeline runtimes from six hours to just two hours. The modernization, detailed in a Microsoft customer story published on May 15, 2026, showcases how the company leveraged Fabric’s unified analytics platform—including SQL Database in Fabric, OneLake, Spark, and Power BI Direct Lake mode—to enable near real-time insights across its retail operations.

Multicedi operates a network of grocery stores and supermarkets throughout Italy, generating large volumes of sales, inventory, and customer transaction data. Legacy infrastructure, built on fragmented on-premises systems and traditional ETL processes, forced long overnight batch runs that frequently crept past six hours—delaying critical business intelligence until well into the next morning. By the time reports reached store managers and category buyers, the data was already stale, limiting the company’s ability to react to stockouts, pricing shifts, or sudden demand spikes.

The move to Microsoft Fabric collapsed this latency dramatically. ETL jobs that once required six hours now complete in roughly two hours, allowing Multicedi to refresh dashboards multiple times per day. Near real-time visibility into sales, margins, and inventory movements is now possible, empowering teams to make faster, more informed decisions.

The technology stack behind the transformation

Multicedi’s new architecture is built around Microsoft Fabric’s unified data lake and analytics engine. At its core lies OneLake, the single SaaS data lake that eliminates data silos by storing all organizational data in an open Delta Parquet format. By landing raw operational data directly into OneLake, Multicedi avoided copying data between multiple systems and simplified governance.

SQL Database in Fabric served as the primary analytical store. Unlike traditional SQL Server or Azure SQL Database, SQL Database in Fabric operates natively within OneLake, allowing tables to be queried via T‑SQL while automatically remaining available to Spark, Power BI, and other Fabric workloads. This convergence meant that Multicedi’s SQL specialists and data engineers could collaborate on the same dataset without complex synchronization.

For heavy transformations—such as deduplicating transaction logs, computing daily aggregates, and enriching sales with product hierarchy data—Multicedi employed Spark notebooks within Fabric. Because Spark jobs directly read and write Delta tables in OneLake, there was no need for extra data movement. The team wrote PySpark scripts that partitioned data by date and store, significantly reducing shuffle and improving job parallelism. These notebooks were orchestrated via Fabric’s built-in Data Pipeline, which replaced legacy scheduler tools.

Power BI Direct Lake mode was the linchpin for delivering sub‑second query performance over billion‑row datasets. Direct Lake allows Power BI to query Delta tables directly from OneLake without importing data into a separate in‑memory model. This eliminated the time‑consuming “import refresh” step that had previously added hours to the reporting cycle. With Direct Lake, Multicedi’s Power BI reports automatically reflect the latest data as soon as Spark jobs finish writing to OneLake.

The story mentions additional components including “Spark, and” at the cut‑off—likely Data Activator, Eventstreams, or Data Factory—all of which integrate seamlessly within Fabric. Eventstreams would have enabled ingestion of real‑time point‑of‑sale events, while Data Activator could trigger automated alerts when inventory drops below thresholds.

From six‑hour batches to near real‑time intelligence

Before the Fabric migration, Multicedi’s ETL follow a classic nightly batch pattern. Each evening, data from stores was extracted via FTP or API calls, loaded into staging tables on SQL Server, transformed using stored procedures and SSIS packages, and finally loaded into a data warehouse for Power BI to import. The entire hop‑from‑source‑to‑report commonly ran past midnight, long after key decision‑makers had left for the day.

With Fabric, the flow was redesigned for speed and simplicity. Raw CSV and JSON files from store systems are now pushed directly into OneLake using Fabric’s built‑in ingestion capabilities. Spark notebooks then execute incremental transformations, processing only the day’s new records instead of full table scans. SQL Database in Fabric hosts curated views and security‑trimmed models, while Power BI’s Direct Lake mode surfaces these models instantly in reports and dashboards.

The 66% reduction in ETL time—from six hours to two—unlocks a cascade of operational benefits. Category managers receive sales dashboards by 8 AM instead of noon. Store managers can track hourly sales against targets by late morning. Supply chain teams spot out‑of‑stock items mid‑day and trigger emergency replenishment orders before customers walk away empty‑handed. Even the finance department, which runs end‑of‑month reconciliation, now gets preliminary P&L figures within hours rather than days.

Business impact: speed, agility, and cost

Faster data processing translates directly into business agility. Multicedi reports that inventory accuracy improved measurably because in‑store scanning and stock corrections are reflected in dashboards within the same operating day. Promotional uplifts are tracked almost immediately, allowing marketing teams to adjust pricing or shelf placement on the fly. Waste reduction, particularly for fresh produce and dairy, saw a noticeable drop because expiry dates are now proactively monitored.

Cost savings are harder to quantify publicly, but the customer story indicates that the move to Fabric’s unified SaaS model eliminated the overhead of managing separate SQL Servers, data lakes, and orchestration servers. Microsoft Fabric’s capacity‑based pricing also means Multicedi pays for a single pool of compute that can be allocated dynamically across SQL queries, Spark jobs, and Power BI—avoiding idle hardware that previously sat powered on 24/7.

Developer productivity and self‑service analytics

Beyond the ETL speed gains, the synergy between SQL Database in Fabric and Power BI Direct Lake lowered the barrier for business analysts. Previously, analysts had to request that IT refresh a semantic model—a process that often took hours and risked breaking downstream reports. Now, reports run directly on up‑to‑date OneLake tables, so analysts can create new measures and visuals without waiting for a batch cycle.

Multicedi also embraced Fabric’s Copilot features to accelerate development. Writing complex Spark transformations is now assisted by natural language prompts, and Power BI reports can be generated from conversational questions. This democratization of data engineering allowed the retailer to shift from a centralized data team of five to a hub‑and‑spoke model where “citizen data practitioners” in merchandising and logistics build their own models.

Implementation timeline and lessons learned

The migration, which began in late 2025, was completed in under four months. Key milestones included:

  • Discovery and design (4 weeks): Mapping source systems, defining OneLake ingestion patterns, and creating a security model with Fabric workspaces for each business function.
  • Pilot migration (6 weeks): Moving the most business‑critical datasets—sales, inventory, product master—to OneLake and building initial Spark transformation pipelines.
  • Parallel run (4 weeks): Running legacy and Fabric pipelines side‑by‑side while iterating on Direct Lake model performance.
  • Switchover and optimization (2 weeks): Phasing out legacy schedules and fine‑tuning Spark partitions, Delta table optimizations, and Power BI aggregations.

One lesson Multicedi highlighted was the importance of optimizing Delta tables from the start. Using Z‑ordering on frequently filtered columns (e.g., store_id, date) and compacting small files after each incremental load prevented query performance degradation. The team also leaned heavily on Fabric’s Monitoring hub to identify Spark job bottlenecks and adjust cluster sizes accordingly.

Security and compliance in a retail context

Handling sensitive retail data—including customer loyalty information and employee metrics—required robust governance. Multicedi used Fabric’s OneSecurity model, which applies Microsoft Purview data protection policies consistently across OneLake, SQL endpoints, and Power BI. Row‑level security implemented in SQL Database in Fabric automatically flows through to Direct Lake reports, ensuring that store managers see only their own store’s data. Auditors can trace data lineage from source to report without additional tools, a significant improvement over the previous fragmented logging.

How Multicedi compares to similar retail transformations

Multicedi’s story mirrors a broader industry trend of brick‑and‑mortar retailers embracing real‑time analytics. Competitors using legacy data warehouses typically struggle to refresh dashboards more than once per day, often waiting 8–12 hours for full loads. By adopting Microsoft Fabric, Multicedi leapfrogged to a near real‑time posture that rivals modern e‑commerce platforms. The two‑hour ETL window is particularly impressive for a retailer that runs hundreds of physical stores with varying internet connectivity and POS system formats.

Compared to other Fabric customer stories—such as those from airline maintenance or logistics—Multicedi’s emphasis on SQL Database in Fabric and Direct Lake stands out. Many organizations still rely on SQL Server mirrored databases or import‑mode Power BI, both of which add latency. Multicedi’s decision to use SQL Database in Fabric as a Direct Lake‑compatible analytical store, rather than just a development database, exemplifies the architectural clarity Fabric enables.

What’s next: real‑time eventstreams and AI

The Microsoft customer story hints at Multicedi’s future plans. Having achieved a two‑hour ETL cycle, the retailer is now exploring Eventstreams to ingest point‑of‑sale data in real time, cutting latency further to minutes. Combined with Data Activator, the company envisions a “zero‑delay” store environment where shelf sensors and transaction logs trigger automatic reorders, dynamic pricing, and personalized coupon delivery via mobile app.

Machine learning is also on the roadmap. With curated datasets already in OneLake, data scientists will use Fabric’s Synapse Data Science modules to build demand forecasting and promotion optimization models. These models can be registered with MLflow and served directly via SQL Database in Fabric or Spark, allowing real‑time scoring within the same pipeline that feeds Power BI reports.

Critical reception and lessons for Windows enthusiasts

Although Fabric is a cloud analytics platform, its relevance extends to the Windows ecosystem. Power BI Desktop, the primary development tool for Fabric reports, runs natively on Windows 11 and Windows 10. Direct Lake mode optimizations, including columnar segment caching and vertical autoscaling, benefit from the latest Windows performance features like GPU‑accelerated rendering and DirectStorage. IT administrators managing Windows workstations for data teams can take advantage of Fabric’s integration with Microsoft Entra ID and Group Policy to enforce secure access to OneLake short‑cuts.

For Windows‑centric organizations considering a similar modernization, Multicedi’s story offers several takeaways:
- Embrace the lake‑first architecture: Land raw data into OneLake and let services like SQL and Spark coexist without duplication.
- Adopt Direct Lake mode early: It eliminates the import bottleneck and works best with well‑partitioned Delta tables.
- Invest in incremental transformation logic: Full refreshes are a relic; Spark in Fabric makes incremental ELT straightforward.
- Leverage Fabric’s unified security: OneSecurity reduces the complexity of managing separate access lists for each tool.

Conclusion

Multicedi’s achievement—cutting ETL from six hours to two—demonstrates the tangible speed and agility gains Microsoft Fabric brings to traditional retailers. By combining SQL Database in Fabric, OneLake, Spark, and Power BI Direct Lake, the company transformed its data platform from a nightly bottleneck into a near real‑time resource. The shortened data latency has already improved inventory management, promotions, and shelf availability, while the unified SaaS architecture reduced operational overhead. As Multicedi pushes toward true real‑time stream ingestion and AI‑powered recommendations, its journey serves as a blueprint for any data‑driven enterprise looking to modernize analytics on the Microsoft stack.