In the high-stakes world of FMCG, mastering category management has become the cornerstone of sustainable growth and differentiation. Success today demands more than intuition and legacy experience. It requires the ability to anticipate market trends, customize assortments, and extract actionable insights from an ever-growing pool of commercial data. As retail partners and consumers alike expect ever-greater sophistication and relevance, forward-thinking manufacturers are turning to tailored analytical ecosystems to outpace the competition.
With this context in mind, DS STREAM had the distinct pleasure of collaborating with Lorenz Polska, a leader in Poland’s snack segment, to co-create a robust, cloud-based platform that empowers fact-based assortment, pricing, and promotional decisions at scale. This engagement stands as a testament to how strategic analytics, when thoughtfully introduced, can unlock hidden value and future-proof operations across the FMCG sector.
Lorenz Polska operates at the heart of the B2B snack food market, serving all major chains and prioritizing modern trade channels. Their drive to optimize promotions, maximize sales efficiency, and rapidly align the product portfolio with shifting consumer tastes reflects the evolving ambitions of category leaders.
Business challenge
Facing pressures familiar to many in FMCG, Lorenz Polska set ambitious goals:
• Build world-class analytical capabilities within commercial teams.
• Automate complex assortment preparation for diverse retail partners.
• Achieve nuanced understanding of consumer purchasing patterns.
• Drive gross margin improvement across an expanding portfolio.
• Optimize assortment management costs by leveraging process automation.
• Institute robust, data-driven promotion effectiveness measurement.
Lorenz’s journey, however, came with its own unique considerations:
• Organizational focus: With analytics expertise still developing internally and resources allocated to several strategic initiatives, the team had to prioritize wisely where to invest their efforts.
• Data characteristics: Working with datasets of varying granularity, promotional details that were sometimes incomplete, and supplier-provided inputs in different formats created an opportunity to build greater consistency and enrich insights over time.
• Right-sized scope: Rather than aiming for “big data” at all costs, Lorenz operated at a scale that was pragmatic and tailored to their needs—ensuring solutions were actionable, efficient, and fit for purpose.
Solution approach: modular analytics with human touch
Our answer was clear: empower Lorenz with a modular, scalable, and pragmatic data ecosystem, meticulously crafted on Google Cloud Platform (GCP) to avoid unnecessary operational complexity while remaining future proof.
Key Solution Components
• Customer and product segmentation: used K-Means, DBSCAN, and hierarchical clustering to derive actionable segments for agile category management.
• Demand forecasting: deployed ARIMA and LightGBM to embed region and season-sensitive forecasts at the SKU level.
• Price elasticity modelling: Bayesian GLM applied to quantify price sensitivity and guide rational discounting.
• Substitute & complement identification: leveraged correlation and semantic analysis to decode purchase affinities and shelf adjacencies.
• SKU portfolio optimization: applied linear/quadratic optimization to balance margin, reach, and shelf efficiency.
• Integrated data automation: unified multiple sources via automated ETL in Big Query, building intuitive dashboards for business users.
Technology Stack
• Cloud & ML: GCP, Vertex AI, BigQuery ML
• Languages & Libraries: Python, pandas, scikit-learn, LightGBM, PuLP, matplotlib, seaborn
• Workflow: SQL-based ETL in BigQuery, Jupyter/Vertex Notebooks for prototyping, productionized in modular pipelines
High-touch implementation
Rather than delivering a sealed, monolithic “solution”, we embedded knowledge transfer and co-creation throughout the engagement.
• Every module paired technical enablement sessions with business-side workshops, upskilling Lorenz’s team for future self-sufficiency.
• Scalable prototypes allowed for iterative extension, unlocking adjacent opportunities (e.g., shelf analysis, retail audit data integration, advanced pricing).
Impact & results we achieved
• Superior sales forecasts: Machine learning models outperformed sales team estimates, boosting demand planning confidence and accuracy.
• Promotional data enrichment: Integration of LLMs unlocked value from previously siloed promotional datasets, improving campaign ROI measurement.
• Automated assortments change detection: The solution was rolled out to selected retail chain stores, enabling rapid category response in a dynamic retail environment.
• Sustained capability growth: Lorenz’s analytics team now independently expands and adapts the platform, reinforcing operational resilience.
Key insights for FMCG leaders
• Empowerment is everything. Prioritizing internal skill development creates lasting impact well beyond project go-live.
• Flexible, modular architecture wins. Clearly delineated analytical components speed delivery, encourage reuse, and lower change management barriers.
• Pragmatism trumps hype. Right-sizing technology avoids cost overruns and ensures fast, visible wins without unnecessary complexity.
• LLMs unlock data gold. Advanced semantic engines can revolutionize the value drawn from standard sales data.
• Agility yields opportunity: iterative development opens doors to new commercial insights, be it shelf management, trade collaboration, or strategic pricing.
Conclusion
Our collaboration with Lorenz Polska demonstrates that advanced analytics (when delivered with a consultative, category-savvy approach) can transform how B2B manufacturers design assortments, price effectively, and measure promotion impact. By focusing on upskilling, modularity, and pragmatic innovation, Lorenz not only realized measurable operational gains but also secured the in-house capabilities needed to lead Poland’s snack category in an era defined by data-driven excellence.
