Modernizing Retail Analytics with Microsoft Fabric

Our client is one of Poland’s leading fashion conglomerates, with a robust portfolio of menswear and accessories brands. Focused on modern tailoring and timeless style, the company plays a significant role in shaping Polish menswear, blending heritage with contemporary design across both retail and e-commerce channels.

The executive board decided to transform the business by embracing modern machine learning technologies. A previously conducted in-depth analysis (available here) laid out a clear, phased roadmap for the company’s digital reorganization and implementation of a Data Monetization Strategy based on years of collected data.

The company needed to implement a new data platform, as its existing architecture lacked full reporting capabilities. Report creation required several hours (sometimes more) of analyst work. Any change in analytics requirements meant reprocessing data, and even with the same calculation scope, separate versions often had to be rebuilt—making the process cumbersome and time-consuming.

Definitions of various data points and KPIs were unclear and inconsistent. For instance, the sales department reported and understood sales data differently than logistics. Data came from diverse sources: a SAS system, ERP, a sales database, standalone APIs used for counting in-store traffic, and Excel tables.

From Algomine’s perspective, this represented classic legacy-system challenges:

• No formal documentation of data sources existed. This gap was mitigated thanks to client experts whose knowledge proved invaluable. Nevertheless, there were occasions when reality didn’t match their recollections, requiring workaround solutions.
• All data was siloed and needed to be connected, unified, and pre-processed to a “common denominator.”
• We conducted system and business process analysis to verify whether reported data aligned with actual workflows and KPIs—for example, one division excluded advance payments in sales reporting, while another included them.
• Multiple data sources had to be integrated: ERP, sales databases, shop-traffic APIs, Excel-based reports, and metadata dictionaries required harmonization.

The project began in January and ended in August.

During the first month, we conducted intensive analyses and workshops to diagnose pain points, issues, gaps, and the overall status quo of the reporting ecosystem.

Within three months, we delivered a concept and design for the new data platform: architectural draft, implementation plan, required data sources, and validation of data quality—consistency, cleanliness, and correctness.

Initial datasets were tested and verified across divisions. As a result, we uncovered critical misunderstandings and deficiencies that shaped the final solution.

A needs analysis and business process audit prompted a revision of the initial design. Originally, orchestration was to be pipeline-based using Spark notebooks. However, due to Spark’s notebook invocation overhead—even with session reuse—the process was slow. Algomine’s project leader proposed reversing the workflow: initiating processing from within notebooks. This change reduced data processing time approximately fivefold.

To manage data from diverse sources, we adopted a metadata-driven approach: definitions of structures, column names, primary keys, and sources were stored in metadata tables. Dependencies between datasets and sources were also captured in metadata. Initially, this added complexity, but over time it greatly simplified onboarding new data into the platform—a flexible, adaptable approach enabling new data layers to be added easily.

Our team recommended building a centralized data platform based on Microsoft Fabric—leveraging components integrated under the unified Fabric environment (including OneLake as the shared data lake, Lakehouse, semantic modeling, and integrated Power BI).

As part of the solution, we implemented a three-layered data architecture within the platform:

• Raw layer – containing unprocessed, source-system data in its original form.
• Standardized layer – where data is cleaned, structured, and business definitions are applied (e.g., naming conventions, KPIs, keys).
• Aggregated layer – enabling the combination and correlation of data across different domains (e.g., business domains, departments), allowing for cross-functional analysis within a single platform.

Data Governance was a crucial element, supported by a toolset for rational metadata management: how data is acquired, its structure, catalog membership, and owner responsibilities.

All data sources were interfaced into Fabric via gateway ingestion. We implemented a 1:1 medallion structure: each source system received its own Lakehouse sandbox—for ERP, each API, each Excel source—mirroring the medallion architecture (bronze/silver/gold) within separate lakehouses. Excel report imports were automated, requiring disciplined, standardized file structures and placement in the appropriate lakehouse location.

Today, data from these systems synchronize smoothly within the Fabric platform, enabling consistent reporting and semantic model building based on client needs.

A key feature is Direct Lake mode: Power BI can connect directly to OneLake (Fabric Lakehouse) in Delta/Parquet format without data import, enabling semantic models to be processed within Fabric and data to remain in storage. This is more performant than traditional DirectQuery, eliminating the need to load data into Power BI and significantly boosting model efficiency.

Once the data platform was in place, our BI team developed dashboards in Power BI covering sales performance, in-store customer traffic, and product-level analytics. They created a modular reporting model, allowing users to assemble reports from ready-made components.

All dashboards are connected to the data lakes and platform components built during the project. The system reports directly from the Fabric Lakehouse, and its responsiveness remains very high even at large data volumes—dashboards are available almost instantaneously.

The dashboards are designed to monitor sales by brand, region, and product, while tracking KPIs. They also provide budget analyses and comparisons of planned versus actual sales performance at the level of each individual store. Additionally, sales can be estimated per product, shop, or region.

Examples of available metrics include:

• items sold,
• year-on-year changes,
• multi-year comparisons,
• sales per square meter,
• profits and costs,
• store visits versus total shopping-centre traffic,
• daily transactions,
• product returns.

Inventory data is also fully integrated into reporting. Stocks can be analyzed by collection, year, and rotation, and projected against sales forecasts. This perspective enables detailed per-product analysis and provides management with a clear view of whether sales goals and KPIs are on track to be achieved.

The introduction of the Fabric platform created a single source of truth: with data centralized in one platform, preparation processes became uniform across the organization—ensuring consistent definitions of metrics such as “sales” across departments.

The final step toward full optimization is phasing out Excel-based reports entirely. This will minimize report preparation time to the point where the reporting interface, scheduled at specific intervals, can deliver real-time lakehouse data immediately.

The Fabric implementation dramatically improved reporting and analysis quality. It accelerated data processing and report generation several times over.

The result: better business visibility, higher decision quality, and faster go-to-market responsiveness.

From an infrastructure standpoint, on-premises archiving is feasible. Fabric’s storage cost is advantageous, and compute is decoupled from file type: you can store Parquet or Delta files, and choose compute independently. SQL, notebooks, and Python workflows all integrate smoothly—making Fabric highly flexible.

These changes position our client to roll out machine learning and the previously crafted Data Monetization Strategy. The platform supports a shift from reactive reporting to proactive, precise demand forecasting and planning.

Ready to Scale Your Data Analytics?

If capacity limits are slowing down your AI and Data Analytics journey, our expert team is here to help.
Let’s connect—schedule a meeting and see how Machine Learning and AI can unlock new opportunities for your organization.