Original practice scenario: A synthetic luxury marketplace imports the same bag from regional catalogs with different names, currency and variant codes. How should identity work?
Read model answer +
Separate a product family from a sellable variant and a regional offer. Similar names or photographs are evidence for review, not sufficient identifiers for merging inventory.
Work through it
Define stable internal identifiers for family, variant and offer, preserving source identifiers and provenance in an explicit mapping table.
Normalize candidate attributes such as material, size and color without discarding the original values; quarantine conflicting mappings instead of merging automatically.
Test renamed collections, reused regional codes, missing attributes and discontinued variants; make corrections reversible with a versioned audit trail.
Tradeoffs to discuss
Strict matching leaves more manual review, while loose matching risks combining distinct goods and misleading availability.
Common pitfalls
Currency conversion does not make two offers identical.
Never transfer stock between variants solely because their display titles match.
Original practice scenario: A hypothetical luxury retailer sees low recorded sales for an item that was unavailable for much of the month. How would you forecast demand?
Read model answer +
Observed sales are constrained by availability and cannot directly represent unconstrained demand. A forecast should express uncertainty rather than treating every stockout day as zero interest.
Work through it
Reconstruct sellable exposure by store, channel and day, separating stockouts, closures, launches and intentional assortment limits.
Compare models using availability-aware features and suitable censored-demand assumptions; retain substitution and wait-list signals as imperfect evidence.
Backtest chronologically across launches and replenishments, evaluating forecast intervals and downstream inventory decisions by item maturity.
Tradeoffs to discuss
Pooling similar products improves sparse estimates but can miss brand or collection effects; detailed models may overfit scarce observations.
Common pitfalls
Do not fill stockout periods with fabricated sales labels.
Avoid using future replenishment outcomes as features available at forecast time.
