From stockouts to happy customers: Proven solutions for time series forecasting in retail
summary
Retail forecasting solutions center on concrete fixes: handle censored demand with seasonal imputation or by passing count out-of-stock days as features and evaluate only on unconstrained periods; solve cold starts with metadata and image embeddings plus LGBM/boosted trees that natively accept missing values; tame promotions via cleaning, decomposing trend/seasonality with a stats model and fitting residual spikes with tree boosters; deploy a model zoo (100+ configs) and an ensemble optimizer to combine models—approach proves robust in production.
AI-written summary · about
chapters (15)
- 0:00Retail forecasting challenges
- 1:11Why retail forecasting is complex
- 3:14Retail forecasting in supply chains
- 5:23Constrained demand impact
- 7:49Solutions for stockouts
- 9:32Evaluating forecasts without stockout bias
- 11:19Cold start problems in retail
- 14:07Model choices for cold start
- 15:55Diverse product portfolios
- 18:47Ensembling models
- 21:14Handling promotions
- 23:41Statistical and ML models
- 24:43Hierarchical data challenges
- 25:47Four years of retail forecasting
- 26:09Closing remarks and Q&A
related talks
22mFrontiers of Machine Learning from the Lens of Retail
30mAI in Retail - Advancing Promo Demand Forecasting
38mOptimizing Retail Forecasting Through AI Models
22mRonald Menich, Chief Data Scientist, Predictix, LLC
40mMulti-Series Forecasting at Scale with StatsForecast
35mBuilding a successful inventory planning solution with Amazon Forecast
59mPredictability Check: Demand Forecasting
08mMachine Learning for Bangladeshi Retailers