Inventory Forecasting — Ganbo (by vivo)
Visual: demand forecast charts, SKU segmentation and replenishment alerts.
Client Overview
Ganbo, a fast-growing e-commerce brand backed by vivo, operates B2C marketplaces and flash-sale channels. Rapid SKU churn and promotional volatility made accurate forecasting and prioritized replenishment essential to reduce excess inventory and avoid missed sales.
Challenge
Seasonal promos, flash sales, and channel-specific demand made simple moving-average forecasts unreliable. The business needed a data-driven approach that could factor promotions, lead-time variability, supplier constraints, and SKU-level seasonality to drive replenishment.
Solution — Multi-horizon Demand Forecasting & Prioritization
We built multi-horizon forecasting models combining time-series models with gradient-boosted feature-rich learners for SKU-level demand, plus a downstream prioritization engine to recommend replenishment action considering supply constraints and promotion plans.
Core modules
Approach
- Feature engineering for promotion intensity, channel mix, and seasonality.
- Cross-validation with holdout windows to validate forecast accuracy across promo cycles.
- Pilot with top 2k SKUs and iterate on lead-time modeling before scaling.
- Integrate with procurement systems to create suggested POs and monitoring loops for supplier performance.
Technology stack
Implementation — How We Rolled Out
Phase 1 — Pilot & Feature Scope (Weeks 1–8)
Identify high-impact SKUs, define features, and run backtests over previous promo cycles.
Phase 2 — Model Ensemble & Validation (Weeks 9–18)
Train ensembles with covariates, measure forecast bias and RMSLE, and implement attribution for promo lift.
Phase 3 — Prioritization & Procurement Integration (Weeks 19–28)
Integrate optimization engine that recommends PO quantities under budget and lead time constraints; pilot with procurement team.
Phase 4 — Scale & Monitor (Weeks 29–36)
Scale to full SKU set with automated retraining and monitoring for model drift and supplier performance.
Impact & Results
30%
Reduction in stockouts on promoted SKUs
18%
Reduction in excess inventory (slow-moving SKUs)
20%
Improvement in forecast accuracy (MAPE) for pilot SKUs
2 months
Time to production-ready forecasts for pilot set
Qualitative outcomes
- Procurement could prioritize POs with clear expected demand and budget trade-offs.
- Marketing and operations aligned better on promotion planning with visible demand forecasts.
- Supplier SLAs improved due to better lead-time planning driven by forecast insights.
Client Testimonial
Key Highlights & Learnings
- Promo-aware features and backtests across promo cycles are essential for reliable forecasts.
- Start with high-impact SKUs; scaling to the long tail requires careful cost-benefit analysis.
- Integrate procurement and marketing calendars into the forecasting loop for alignment.