Research & Insights
Data-driven research from the invent.sale team
The Stochastic Forecasting Engine: 4 Competing Models per SKU
Complete breakdown of the StochasticProjector engine: polynomial, sinusoidal, exponential, and logarithmic models compete for every product. Weighted R² with exponential decay, P10/P50/P90 Poisson confidence bands, demand cliff detection, and stockout prediction.
5-Type Seasonality Classification: How the Algorithm Distinguishes Peak, Off-Season, and False Signals
Complete seasonality classification algorithm: 52-week profiling with exponential year-over-year decay, low variability detection (±30% threshold), adjacent peak merging, year-boundary wraparound handling, and 5 types for every product-week combination.
Dead Stock Detection & Cascade Markdown Optimization: 4 Criteria, 4 Discount Tiers
4-criteria dead stock filter (7 days without sales, not off-season, stock ≥ P10, balance > 0), frozen capital calculation, cascade markdown system at 50/25/15% by seasonality type, and release ratio percentile analysis on real retail chain data.