Design an Amazon-Like Online Store

Problem statement

Design a global e-commerce platform in the style of Amazon: an effectively unlimited product catalog with images, descriptions and SKUs; search and filtering by category, brand and price; user accounts, wishlists and recommendations; shopping carts with a secure checkout flow; and order placement with tracking and email confirmations. The system must survive millions of daily visitors, flash-sale spikes of tens of times baseline, and seasonal surges such as Prime Day or Black Friday, while keeping page loads sub-second and never overselling inventory or double-charging a customer.

The defining property of this system is asymmetry. Browsing is enormously read-heavy: a shopper may generate hundreds of product views, search queries and recommendation impressions for a single order. Yet the moments that matter legally and financially - inventory decrement, payment authorization, order commit - are low-volume, strongly consistent transactions. A strong design therefore does not apply one consistency model or one database to everything; it separates the discovery plane (search, browse, recommendations), the merchandising plane (catalog, pricing, promotions), the transaction plane (cart, checkout, payment), the fulfillment plane (inventory, warehouse, shipment), and the engagement plane (reviews, notifications, loyalty), and gives each its own storage, caching and failure behavior.

Why the problem is distinctive

A social feed can show stale content; a store cannot show a stale price at checkout or sell inventory that does not exist. Conversely, requiring strong consistency everywhere would make the browse path slow and fragile. The interview-worthy tension is deciding exactly where correctness is non-negotiable (inventory ledger, payment, order aggregate) and where freshness-bounded eventual consistency is a feature (search index, recommendation scores, review counts, catalog cache).

Public operating baseline versus design assumptions

Public figures establish the category's scale. Amazon reported net sales of 574.8 billion USD in 2023 and stated that Prime Day 2023 was its biggest ever, with more than 375 million items sold worldwide. Alibaba publicly reported a peak of 583,000 orders per second during its 2020 Double 11 festival. Shopify reported that its merchants sold 9.3 billion USD worth of goods over Black Friday to Cyber Monday 2023. These are cited company figures for context, not requirements for our fictional platform. For capacity planning this answer explicitly assumes: 200 million registered customers, 20 million daily active users, 2 million concurrent browsers at peak, 240 million page views per day, 4 million orders per day, and an 8x flash-sale peak multiplier on reads with up to 60x on order creation for a featured deal. Unless a number is tied to a citation, it is a stated design assumption, target, or budget.

The five planes

1. Discovery: search, category browse, recommendations, personalization.
2. Merchandising: catalog, SKU/offer management, pricing, promotions, media.
3. Transaction: session, cart, checkout, payment, tax, fraud screening.
4. Fulfillment: inventory ledger, reservations, warehouse assignment, shipment, returns.
5. Engagement: reviews, ratings, wishlists, notifications, loyalty, support.

Keeping these planes separate is what allows the discovery plane to degrade (stale recommendations, cached facets) without ever weakening the transaction plane.