Original practice scenario inspired by Hermès International Société en commandite par actions's public business context; this is not a reported interview question or a description of its internal systems. A fictional luxury retailer offers a small number of an item online while stores can also sell it. Design a short-lived reservation workflow that avoids overselling without keeping stock locked forever.
Read model answer +
Model availability, reservation, and completed sale as distinct states. A reservation temporarily reduces sellable quantity and must have an authoritative expiry and a guarded conversion to a sale.
Work through it
Choose a stock authority at the location-item grain and atomically reserve only when available quantity is sufficient. Attach a stable reservation identifier.
Make payment completion and reservation conversion idempotent. Resolve a payment arriving after expiry with an explicit compensation or exception policy.
Expire reservations through durable processing and reconcile stock with store updates. Test concurrent buyers, lost responses, delayed payments, and inventory corrections.
Tradeoffs to discuss
Conservative reservations can temporarily reduce conversion, while optimistic promises increase oversell and cancellation risk. Safety stock can absorb delayed store updates.
Common pitfalls
A browser countdown is not the expiry authority.
Reading quantity and decrementing later without concurrency control allows two buyers to reserve the same unit.
Original practice scenario inspired by Hermès International Société en commandite par actions's public business context; this is not a reported interview question or a description of its internal systems. A synthetic retail dataset records sales and inventory. Some products sell out early, so their observed sales stop. Explain why ranking demand by sales alone is biased and propose a defensible analysis.
Read model answer +
Observed sales are limited by both customer interest and available supply. A sold-out product has censored demand; a low total does not establish low interest when it was unavailable for much of the period.
Work through it
Construct availability intervals at product-location level and distinguish stocked exposure from calendar time. Check inventory accuracy and replenishment delays.
Compare sales rates during comparable available periods, with uncertainty for short exposures. Add waitlist or unavailable-page signals only after examining how they are collected.
Report measured sales separately from inferred lost demand. Validate assumptions through controlled stock allocations or replenishment changes where practical.
Tradeoffs to discuss
Additional signals can improve estimates but introduce selection effects. No method recovers exact unmet demand without assumptions about customer behavior.
Common pitfalls
Treating stockout hours as zero-demand observations biases estimates.
Extrapolating one unusually busy hour across a whole week can overcorrect just as severely.
