01 · Manual stock decisions
Inventory decisions required teams to compare sales, stock, product, and store information manually. The same network could experience stockouts in one location and excess inventory in another.
Customer work · Retail and consumer
A national lifestyle retailer connected SKU-store forecasts to inventory decisions
BluePi built a recurring forecasting and recommendation cycle for ordering, reordering, distribution, and redistribution. In one category, same-store sales grew 11 percent, the safety-stock factor moved from 2.5 to 1.5, and raw-material planning extended to nine months.
Forecasts became recurring inventory actions
The retailer managed ordering and inventory balancing across a large product and store network. BluePi connected forecasts to the actions planners needed to take: order, reorder, distribute, or move stock between stores.
02 · SKU-store planning cycle
An AWS-based forecasting system produced demand estimates by SKU and store across short planning horizons. The workflow converted those estimates into recommendations for reordering and redistribution.
03 · Sales and safety stock
In one category, same-store sales grew 11 percent. The safety-stock factor moved from 2.5 to 1.5, and raw-material planning extended to nine months.
From historical sales to a SKU-store recommendation
The system joined sales, stock, product, and store data, generated SKU-store forecasts, converted them into ordering and redistribution recommendations, and connected planner action to later inventory outcomes.
Forecast at the decision grain
The models produced forecasts for individual products and stores so portfolio averages did not hide weak recommendations at specific locations.
Match the horizon to the action
Separate 14-day and 30-day forecasts supported reordering, redistribution, and decisions with longer supply lead times.
Measure the inventory outcome
Teams compared recommendations with later sales, stockout days, fill rates, and excess inventory to identify where the model, planning rule, or execution process needed attention.
Where this pattern fits
This pattern fits retailers that manage inventory across many stores and products but still rely on manual analysis for reordering and redistribution. A useful starting point is one representative product group, a fixed set of stores, and a planning cycle with measurable stock and sales outcomes.
Case details
Open a section to review the customer problem, implementation, business change, and architecture.
01The starting point
The retailer needed to balance product availability against excess inventory across a distributed store network.
- Manual inventory planning: Ordering and redistribution depended on repeated analysis and individual judgment.
- Stock imbalance: The same product could be unavailable at one store and overstocked at another.
- Different planning horizons: Procurement, reordering, and redistribution required forecasts over different periods.
- Limited traceability: Teams could not easily connect a stock outcome to the forecast and recommendation that preceded it.
02What BluePi built
BluePi connected source data, forecasting, recommendation logic, planner review, and post-live measurement.
- Planning data in Amazon S3: Historical sales and master data entered through CSV or SQL extracts transferred by SFTP.
- Feature preparation: Sales, store, product, inventory, calendar, and other relevant signals were cleaned and combined.
- SageMaker forecasting models: Python models generated demand estimates by SKU and store.
- Multiple forecast horizons: Separate 14-day and 30-day models supported different inventory actions.
- Redshift result storage: Forecasts and analytical outputs were available for comparison and reporting.
- EC2-based analysis: Teams could review forecasts, recommendations, and later results.
- Inventory recommendations: The system supported ordering, reordering, distribution, and redistribution.
03Results
The retailer moved from manual inventory analysis to a recurring forecasting and recommendation cycle across multiple product categories.
- 11 percent same-store sales growth in one category
- Safety-stock factor reduced from 2.5 to 1.5
- Raw-material planning extended to nine months
- SKU-store recommendations supported recurring reordering and redistribution
- Post-live reviews tracked sales, stockout days, order fill rate, demand fill rate, and excess inventory