RetailLoop
RetailLoop's five brands were each forecasting demand in separate spreadsheets, leading to chronic stockouts on bestsellers and overstock on slow movers. Aivora AI built a unified demand forecasting pipeline that blends historical sales, seasonality, and promotional calendars into a single SKU-level model feeding directly into RetailLoop's replenishment system. The result was fewer stockouts on top sellers and meaningfully leaner inventory across the board.
-44%
Stockouts on top 50 SKUs
-19%
Excess inventory carrying cost
+15 pts
Forecast accuracy (MAPE improvement)
10 weeks
Time to first production forecast
Wherethingsstood
RetailLoop runs five direct-to-consumer apparel and home goods brands out of a shared warehouse and fulfillment operation, managing demand planning for roughly 400 active SKUs. Each brand's merchandising team maintained its own forecasting spreadsheet, updated manually on a weekly cadence, with no shared view of cross-brand seasonality or promotional overlap.
The result was predictable: bestsellers routinely sold out during promotional pushes because reorder points hadn't accounted for the promotion's expected lift, while slower-moving SKUs piled up in the warehouse because nobody had flagged the seasonal drop-off in time. Inventory carrying costs were climbing even as stockout-driven customer complaints rose in parallel.
RetailLoop needed a forecasting system that could ingest data from all five brands, account for promotional calendars and seasonality per category, and hand off clean reorder recommendations to the existing replenishment workflow without requiring merchandisers to become data scientists.
Whatwebuilt
Aivora AI began with a two-week data audit across all five brands' sales history, promotional calendars, and existing inventory management system exports, standardizing everything into a single warehouse schema. The forecasting pipeline itself combines a gradient-boosted baseline model per SKU category with a promotional-lift adjustment layer trained on RetailLoop's own historical promotion data.
The pipeline runs nightly, refreshing forecasts for all 400 SKUs and pushing reorder point recommendations directly into RetailLoop's replenishment system via API, with a merchandiser-facing dashboard showing forecast confidence and the key drivers behind each recommendation. Merchandisers can override any recommendation, and overrides are logged to improve the model over time.
The build took ten weeks from data audit to first production forecast, with a four-week parallel-run period where the new system's recommendations were compared against the existing spreadsheet process before merchandisers fully switched over.
Howtheengagementran
Discovery
Two-week audit of sales history, promotional calendars, and inventory data across all five brands.
Prototype
Built the SKU-category forecasting models and promotional-lift layer, backtested against 18 months of history.
Pilot
Four-week parallel run comparing model recommendations against the existing spreadsheet process on two brands.
Rollout
Rolled out nightly forecasting and API-driven reorder recommendations across all five brands and 400 SKUs.
“We used to find out a bestseller was out of stock from a customer complaint. Now the reorder happens before we'd have even noticed the problem ourselves.”
Marcus Feld, VP of Merchandising, RetailLoop