← All case studies
Case study · Demand planning
SKU-level forecasts that cut markdowns and stockouts together
A North American specialty retailer replaced static seasonal curves with confidence-banded demand signals across 2,400 SKUs.
- Industry
- Specialty retail
- Profile
- Omnichannel retailer · 180 stores
- Timeline
- Rollout in 64 days · Redshift + Oracle
82%
forecast accuracy
−15%
markdown rate
+18%
inventory turns
Summary
Merchants were choosing between markdown pain and stockout pain every quarter. Predicta delivered SKU-region forecasts with explicit confidence bands merchandisers could trust.
The challenge
Legacy demand planning assumed stable seasonality. Promo lifts and micro-regional weather shocks were handled with gut feel. Buyers lacked a single source of truth before lock deadlines.
What we deployed
- 01Connected Redshift sales history, promo calendars, and Oracle inventory positions into one feature pipeline.
- 02Trained SKU-level models with confidence bands surfaced in the merchandising workflow — not a separate BI tool.
- 03Automated weekly what-if scenarios for promo depth and allocation shifts.
- 04Synced approved forecasts back to Oracle via existing batch windows — no rip-and-replace.
AWS RedshiftOracle RetailPredicta demand engineMerchandising Slack bot
Outcomes
- Demand forecast accuracy improved from 70% to 82% on top-volume SKUs.
- Markdown rate fell 15% year-over-year in the first full season post-deploy.
- Inventory turns improved 18% without increasing stockout rate on hero SKUs.