AI Demand Forecasting Quiz

Answers and explanations appear as you take the quiz.

1. Why can one global forecasting model help when an organization has many related series?
   • It can learn shared patterns across series while retaining series-specific inputs
   • It guarantees the same forecast error for every series
   • It replaces chronological validation with a random split
   • It means one model is trained separately for each series
2. Why are probabilistic (quantile) forecasts valuable for inventory planning?
   • They give a single guaranteed number
   • They express a range and confidence so planners can set safety stock against risk
   • They eliminate the need for any historical data
   • They make forecasts run faster
3. Which is a classic statistical forecasting method that predates modern ML approaches?
   • Temporal Fusion Transformer
   • DeepAR
   • ARIMA
   • TimesFM
4. What is a critical pitfall to avoid when training a demand forecaster?
   • Using calendar features
   • Including promotion information
   • Producing a confidence interval
   • Leaking future data into the training of past predictions
5. Which external signal would most plausibly improve a forecast for ice cream sales?
   • Daily temperature and weather
   • Server uptime logs
   • The CEO's birthday
   • Font choices on the website
6. What did the cited decoder-only foundation-model paper report about its model’s zero-shot forecasts?
   • It guarantees better forecasts for every business series
   • On several public datasets, out-of-the-box accuracy came close to supervised models trained for each dataset
   • It eliminates the need for statistical baselines
   • It forecasts only with fine-tuning on each target series
7. What architecture does the DeepAR paper use to produce forecasts across related time series?
   • A single ARIMA model fitted separately to each series
   • A static lookup table of future values
   • An autoregressive recurrent network trained on many related series
   • A diffusion image model that reads plotted charts
8. Why evaluate a demand model against statistical baselines?
   • To ensure a neural model always wins
   • To make data leakage unavoidable
   • To remove uncertainty from demand
   • To test whether added model complexity improves the target forecast task