ORIGINAL COMPANY-CONTEXT PRACTICE
Forecast demand with changing assortment
Original practice scenario inspired by The TJX Companies, Inc.'s public business context; this is not a reported interview question or a description of its internal systems. A fictional off-price retailer sells products that may not be replenished. A model trained on long-lived items is proposed for new short-lived stock. Describe an evaluation and baseline strategy.
Written for independent practice using company context; this is not a verified question from a company interview.
A REASONED APPROACH
Model answer.
Match the evaluation to new-item forecasting rather than repeated observation of familiar products. Assortment changes, stock limits, and short selling windows make a simple random-row split misleading.
Work through it
- Hold out complete items and later arrival periods. Use features known at allocation time, and separate product availability from observed sales.
- Compare with category-level and attribute-based baselines. Score cumulative demand over the intended selling horizon, alongside sell-through and uncertainty where supply constrains observation.
- Inspect performance by category, price band, and item lifetime. State how the model handles unseen attributes and how often it abstains or falls back.
Tradeoffs to discuss
- Pooling categories helps sparse new items but can miss important differences. More complex item embeddings need evidence that they generalize beyond familiar catalog records.
Common pitfalls
- Treating sold-out days as true zero demand biases training.
- Evaluating only items with long histories excludes the difficult cases the proposal is meant to serve.