Coupang System Design Interview

Table of Contents

What makes Coupang’s interview different from other tech companies Core evaluation areas in the Coupang system design interview Real-time fulfillment and logistics coordination High-scale e-commerce search and discovery End-to-end supply chain visibility Reliability and failure recovery Cost efficiency and throughput Common Coupang system design interview topics Design a real-time order fulfillment system Designing the last-mile delivery system Inventory synchronization across warehouses Search and product discovery Real-time delivery tracking How to structure your answer for the Coupang system design interview Step 1 and 2: Clarify functional and non-functional requirements Step 3: Estimate scale Step 4: High-level architecture Step 5: Deep dive into the most critical components Step 6: Handle failure scenarios Step 7: Discuss trade-offs explicitly Step 8: Scaling and evolution Example: High-level design for Coupang’s Rocket Delivery system Mistakes that cost candidates in the Coupang interview Conclusion

Home/

Blog/

System Design/

Coupang System Design Interview

Learn how to crack the Coupang System Design interview by mastering real-world logistics, inventory accuracy, and SLA-driven architecture. This guide shows you how to think, design, and explain systems the way Coupang expects.

Mar 10, 2026

The Coupang system design interview tests your ability to architect software systems that operate at the intersection of distributed computing and physical-world logistics. Unlike typical system design rounds at social media or SaaS companies, Coupang’s interview demands that you reason through warehouse capacity, courier routing, inventory accuracy, and delivery SLAs, all at national scale in one of the world’s most densely populated markets.

Key takeaways

Logistics-first design thinking: Coupang evaluates whether you can bridge software architecture with real-world constraints like warehouse throughput, courier availability, and last-mile delivery windows.
Inventory accuracy is non-negotiable: Preventing overselling through reservation logic and event-driven synchronization across fulfillment centers is a core interview theme.
Rocket Delivery drives the conversation: Most design questions revolve around Coupang’s same-day and next-day delivery promise, requiring you to design for speed, precision, and cost efficiency simultaneously.
Failure recovery matters more than perfection: Interviewers prioritize graceful degradation, isolation strategies, and fallback modes over theoretical zero-downtime architectures.
Trade-off reasoning signals seniority: Clearly articulating why you chose eventual consistency over strong consistency, or batching over immediate dispatch, carries more weight than the choice itself.

Every year, Coupang delivers hundreds of millions of packages across South Korea, and roughly 70% of those orders arrive within 24 hours of being placed. That is not a marketing statistic. It is an engineering constraint that shapes every system inside the company, from the database schema backing inventory counts to the GPS ingestion pipeline tracking each courier in real time. If you are preparing for a Coupang system design interview, you need to internalize one truth: the systems you design must work in the physical world, not just on a whiteboard.

This guide breaks down exactly what Coupang evaluates, the architecture problems you will face, and the structured framework that produces senior-level answers.

What makes Coupang’s interview different from other tech companies#

Coupang is not a typical e-commerce marketplace that outsources fulfillment to third parties. It is a vertically integrated logistics company that owns its fulfillment centers, delivery fleet, and routing algorithms. This means the system design interview goes far beyond generic discussions of load balancers and microservices.

Interviewers expect you to reason about physical constraints. How many items can a warehouse pick per hour? What happens when a fulfillment center goes offline during a flash sale? How do you assign 50,000 packages to 10,000 couriers across a metropolitan area in under a minute? These are not hypothetical edge cases. They are daily operational realities.

Real-world context: Coupang operates over 100 fulfillment and logistics centers across South Korea, a country roughly the size of Indiana but with a population of 52 million. Population density makes last-mile delivery faster but also increases traffic unpredictability and courier contention.

The interview also places unusual emphasis on cost awareness. Because Coupang owns the entire supply chain, every inefficiency in batching, routing, or warehouse utilization translates directly into margin loss. You will be expected to justify your design choices not just technically but economically.

This combination of software architecture, physical logistics, and cost reasoning is what separates Coupang interviews from those at companies focused purely on digital products. Understanding the core evaluation areas will help you focus your preparation where it matters most.

Core evaluation areas in the Coupang system design interview#

Coupang’s interviewers assess candidates across five distinct dimensions. Each one maps to a real engineering challenge the company faces daily.

Real-time fulfillment and logistics coordination#

Coupang’s fulfillment systems must reflect physical reality at all times. When a customer places an order, the system must validate inventory, select the optimal warehouse, and initiate picking and packing workflows, all within seconds. Interviewers expect you to explain how software coordinates humans, robots, and conveyor systems within a fulfillment center.

Strong answers demonstrate understanding of order orchestrationThe process of managing an order's life cycle across multiple services, from validation and payment through warehouse assignment, picking, packing, and handoff to delivery. and how updates propagate with low latency across dependent systems. You should be able to describe how picking queues are prioritized when multiple orders compete for the same warehouse resources.

High-scale e-commerce search and discovery#

Coupang serves tens of millions of SKUs. Search must return relevant, personalized results in under 200 milliseconds, even during peak traffic events like sales campaigns. Interviewers evaluate your ability to design systems that handle indexing, ranking, promotions, and availability filtering without sacrificing latency.

Key considerations include:

Read-heavy scaling patterns: How do you serve search queries that outnumber writes by 100:1?
Inventory-aware ranking: How do search results reflect real-time stock levels across warehouses?
Personalization under load: How do you blend user history with global relevance scoring?

Attention: A common mistake is designing search systems that treat inventory as a static attribute. At Coupang’s scale, inventory changes thousands of times per second, and showing out-of-stock items degrades trust and conversion rates.

End-to-end supply chain visibility#

Every package, truck, and courier in Coupang’s network is tracked continuously. Interviewers assess how you design telemetry ingestion pipelinesSystems that collect, transport, and process high-frequency sensor or location data from distributed sources such as courier GPS devices, warehouse scanners, and delivery vehicles. that handle noisy, intermittent data at massive throughput.

You should be prepared to discuss location streaming, ETA prediction models, and how the system handles GPS signal loss or delayed updates. Observability is not an afterthought at Coupang. It is a core system requirement.

Reliability and failure recovery#

Failures in Coupang’s systems have immediate customer impact. A warehouse management system crash during peak hours can delay thousands of deliveries. Interviewers expect practical strategies for handling partial failures without cascading outages.

Resilient designs emphasize:

Bulkhead isolation: Preventing a failure in one warehouse’s systems from affecting others.
Fallback modes: Serving cached inventory data when the primary database is unavailable.
Graceful degradation: Continuing to accept orders with reduced accuracy rather than returning errors.

Cost efficiency and throughput#

Because Coupang owns its logistics network, inefficiencies compound at scale. A 5% reduction in delivery route efficiency across 100,000 daily routes translates into enormous fuel and labor costs. Interviewers expect you to reason about batching strategies, warehouse utilization, database partitioning, and service decomposition through a cost-aware lens.

The following table compares how Coupang’s evaluation areas differ from those at a typical SaaS-focused interview.

Coupang vs. Typical SaaS/Social Platform: System Design Interview Evaluation Dimensions

Evaluation Area	Coupang Focus	Typical SaaS/Social Focus
Latency Requirements	Low-latency systems for rapid order processing and real-time delivery tracking to meet SLAs	Responsive user interactions for seamless UX; some operations may tolerate higher latency
Data Consistency	Strong consistency required to accurately reflect inventory and order statuses, preventing overselling	Eventual consistency often acceptable, balancing performance and availability
Failure Impact	Failures cause operational disruptions (delayed deliveries, dissatisfied customers); robust fault tolerance is critical	Failures degrade UX but rarely have immediate critical consequences; recovery strategies are more flexible
Cost Sensitivity	High sensitivity; logistics and infrastructure costs must be tightly optimized for profitability	Important but more flexible, especially early-stage; focus leans toward scaling and feature development
Physical Constraints	Must account for warehouse locations, delivery routes, and transportation limits in system design	Primarily virtual infrastructure concerns; minimal direct impact from physical-world constraints

With these evaluation areas in mind, let us look at the specific problem types you are likely to encounter during the interview itself.

Common Coupang system design interview topics#

The problems Coupang asks are not abstract. They map directly to the company’s core systems. Here are the most frequently reported design challenges and what interviewers look for in each.

Design a real-time order fulfillment system#

This is the most common question in the Coupang system design interview. You are expected to design a system that takes an incoming order and moves it through inventory validation, warehouse selection, picking, packing, and dispatch.

The critical challenge is preventing overselling. When thousands of customers attempt to buy the same item simultaneously, your system must guarantee that inventory is reserved atomically. Interviewers want to hear about reservation logicA pattern where inventory is temporarily held (reserved) for a specific order during checkout, then either committed upon payment confirmation or released back to available stock after a timeout. and how you handle reservation timeouts, partial stock, and split shipments across multiple warehouses.

Strong designs also address congestion avoidance within warehouses. If 500 orders all require items from the same aisle, how does your system distribute picking work to avoid physical bottlenecks?

Pro tip: Frame your warehouse selection algorithm around multiple factors, not just proximity. Consider current warehouse load, item availability across locations, shipping cost, and delivery SLA. A simple nearest-warehouse heuristic will not impress interviewers.

Designing the last-mile delivery system#

Candidates are often asked to design a delivery routing and assignment engine similar to Coupang’s Rocket Delivery service. This problem tests your ability to reason about courier availability, route clustering, real-time traffic data, and dynamic reassignment.

The key tension is between efficiency and predictability. Optimal batch routing minimizes total distance but introduces delays for individual packages. Real-time reassignment handles disruptions but adds system complexity. Interviewers want to see you navigate this trade-off explicitly.

You should discuss how courier scoring works. A typical model considers proximity to the pickup location, current load, remaining shift time, historical delivery speed, and vehicle type. The assignment engine must run these calculations for thousands of couriers and packages within tight time windows.

The following diagram illustrates how packages flow from warehouse dispatch through courier assignment to final delivery.

Inventory synchronization across warehouses#

Coupang operates dozens of fulfillment centers, each maintaining local inventory state. The challenge is keeping inventory consistent across locations without resorting to heavy global locking that would destroy throughput.

Interviewers expect you to discuss eventual consistencyA consistency model where distributed nodes are allowed to temporarily diverge but are guaranteed to converge to the same state given enough time and no new updates. as a practical approach, supplemented by reservation-based locking for high-demand items. You should explain reconciliation processes that detect and correct discrepancies caused by mis-scans, damaged goods, or theft.

A strong answer covers:

Event-driven updates: Using an event bus like Apache Kafka to propagate inventory changes across services.
Conflict resolution: How the system handles cases where two warehouses both decrement stock for the same item simultaneously.
Periodic reconciliation: Batch jobs that compare physical counts against system state.

Historical note: Coupang’s early inventory systems reportedly struggled with overselling during flash sales, which drove the company to invest heavily in event-sourced inventory architectures that maintain a complete audit trail of every stock movement.

Search and product discovery#

Search systems at Coupang must support high concurrency, Korean language morphology, personalization, and real-time availability filtering. Candidates are expected to explain how search remains fast and relevant during peak traffic.

Your design should include an inverted index layer for text matching, a ranking service that blends relevance with business signals (margin, promotion status, inventory level), and a caching layer that absorbs read spikes. Interviewers pay close attention to how you handle index stalenessThe condition where a search index contains outdated information because updates from the primary data store have not yet been reflected, potentially causing users to see unavailable or incorrectly priced items. and its impact on user experience.

Real-time delivery tracking#

Tracking systems ingest GPS updates from courier devices every few seconds, potentially millions of data points per minute across the fleet. Interviewers assess how you design ingestion pipelines, smoothing algorithms, and notification triggers.

Key design decisions include choosing between push and pull models for location updates, how to handle out-of-order or duplicate GPS events, and how to recalculate ETAs without overwhelming downstream services. The ETA model typically considers remaining route distance, current traffic, courier speed history, and delivery complexity.

Real-world context: GPS data from courier smartphones is notoriously noisy, especially in dense urban environments with tall buildings. Production tracking systems must apply Kalman filtering or similar smoothing algorithms to prevent erratic location jumps on the customer-facing map.

Now that we have covered what you will be asked, let us walk through exactly how to structure your response during the interview.

How to structure your answer for the Coupang system design interview#

A structured approach signals experience and prevents the rambling that derails many candidates. The following eight-step framework aligns with Coupang’s logistics-heavy interview style and mirrors expectations at companies like Amazon and Instacart.

Step 1 and 2: Clarify functional and non-functional requirements#

Begin by asking targeted questions that demonstrate domain understanding. Do not ask generic questions. Ask about delivery speed promises (same-day vs. next-day), fulfillment center count and geography, courier fleet model (employee vs. contractor), and peak traffic multipliers during sales events.

Then explicitly state the non-functional requirements you are designing for. At Coupang, these almost always include:

Sub-second inventory validation latency
99.95% or higher availability for order placement
Inventory accuracy above 99.9%
Delivery SLA compliance above 98%
Cost-efficient at 10x current scale

Anchoring your design around these constraints gives every subsequent decision a clear justification.

Step 3: Estimate scale#

Interviewers expect realistic numbers. A reasonable starting assumption for Coupang might be 1 to 2 million orders per day, peaking at 5 to 10x during major sales. With 50,000 active couriers streaming GPS every 3 seconds, that is roughly 17,000 location events per second during peak delivery windows.

Use these estimates to drive decisions about database sharding, message queue partitioning, and cache sizing. A quick back-of-envelope calculation builds credibility.

For example, if each GPS event is 200 bytes:

$$\\text{Peak throughput} = 17{,}000 \\times 200 = 3.4 \\text{ MB/s}$$

That is manageable for a single Kafka cluster, but add order events, inventory updates, and warehouse telemetry and you are looking at a multi-cluster streaming architecture.

Pro tip: Always state your assumptions explicitly and say “I would validate these numbers with the team.” This signals collaborative engineering culture, which Coupang values.

Step 4: High-level architecture#

A typical Coupang system design includes the following core services working in concert.

The services should be decomposed along domain boundaries. Order management handles life cycle state. Inventory manages stock counts and reservations. The warehouse management system coordinates physical operations. The delivery assignment engine handles courier matching and routing. Each service owns its data and communicates through the event bus for cross-cutting concerns.

Step 5: Deep dive into the most critical components#

The interviewer will typically pick one or two components for detailed exploration. Be prepared to go deep on any of them, but inventory reservation and delivery assignment are the most common targets.

For inventory reservation, explain the two-phase approach. When an order is placed, the system issues a soft reservation that decrements available count and sets a TTL. If payment succeeds within the window, the reservation is committed. If the TTL expires, stock is released. This prevents phantom inventoryA state where the system believes items are available but they are already committed to other orders or physically missing from the warehouse, leading to overselling and customer disappointment. while avoiding indefinite locks.

Python

import threading
from datetime import datetime, timedelta
from dataclasses import dataclass, field
from typing import Optional
@dataclass
class Reservation:
    order_id: str
    item_id: str
    quantity: int
    created_at: datetime = field(default_factory=datetime.utcnow)
    ttl_seconds: int = 300
    status: str = "soft"  # "soft" | "confirmed" | "released"
# In-memory stores (replace with Redis/DB in production)
inventory: dict[str, int] = {}          # item_id -> available_count
reservations: dict[str, Reservation] = {}  # order_id -> Reservation
_lock = threading.Lock()
def reserve_inventory(order_id: str, item_id: str, quantity: int, ttl_seconds: int = 300) -> bool:
    """Atomically decrement inventory and create a soft reservation with TTL."""
    with _lock:  # atomic check-and-decrement to prevent overselling
        available = inventory.get(item_id, 0)
        if available < quantity:
            return False  # insufficient stock; reject reservation
        inventory[item_id] = available - quantity  # decrement available count
        reservations[order_id] = Reservation(
            order_id=order_id,
            item_id=item_id,
            quantity=quantity,
            ttl_seconds=ttl_seconds,
        )
    # schedule timeout handler to auto-release if payment not confirmed
    timer = threading.Timer(ttl_seconds, _handle_reservation_timeout, args=[order_id])
    timer.daemon = True
    timer.start()
    return True
def _handle_reservation_timeout(order_id: str) -> None:
    """Reverse a soft reservation if payment was not confirmed before TTL expires."""
    with _lock:
        reservation = reservations.get(order_id)
        if reservation is None or reservation.status != "soft":
            return  # already committed or previously released; nothing to do
        # restore inventory by returning the reserved quantity
        inventory[reservation.item_id] = inventory.get(reservation.item_id, 0) + reservation.quantity
        reservation.status = "released"
    print(f"[TIMEOUT] Reservation {order_id} expired — inventory restored.")
def commit_reservation(order_id: str) -> bool:
    """Convert a soft reservation to a hard allocation upon payment success."""
    with _lock:
        reservation = reservations.get(order_id)
        if reservation is None:
            return False  # no reservation found for this order
        if reservation.status != "soft":
            return False  # already released or committed; idempotency guard
        reservation.status = "confirmed"  # hard allocation; inventory already decremented
    print(f"[COMMIT] Reservation {order_id} confirmed — allocation locked.")
    return True

For delivery assignment, walk through the scoring model. Each courier receives a score based on weighted factors:

$$\\text{Score} = w1 \\cdot \\text{proximity} + w2 \\cdot \\text{capacity} + w3 \\cdot \\text{SLA_urgency} + w4 \\cdot \\text{historical_speed}$$

The engine evaluates all eligible couriers, ranks by score, and assigns optimistically with conflict resolution for race conditions.

Step 6: Handle failure scenarios#

This is where many candidates fall short. Coupang interviewers specifically probe failure modes because logistics failures have direct customer impact. Prepare concrete answers for these scenarios.

Failure Scenarios and Mitigation Strategies

Failure Type	Impact	Mitigation Strategy
Warehouse System Outage	Order processing halts, shipment delays, inventory inaccuracies, revenue loss	Real-time health dashboards, disaster recovery plan with backups, failover systems, and routine drills
Courier GPS Signal Loss	Inaccurate shipment tracking, delivery delays, misrouted packages, reduced customer satisfaction	Cellular triangulation or RFID as backup, manual location reporting protocols, predictive delivery analytics
Inventory Database Failure	Inaccurate inventory data, overselling, stockouts, fulfillment errors	Fault-tolerant multi-node architecture, data replication via Apache Kafka, Kubernetes for automated failure recovery
Kafka Cluster Degradation	Delayed or lost messages, disrupted microservice communication, potential data loss	Monitor producer/consumer telemetry, limit excessive connections, enforce proper resource allocation and configuration reviews
Flash Sale Traffic Spike	Website slowdowns or crashes, lost sales, customer frustration, reputational damage	Load balancing, caching strategies, scalable infrastructure, performance testing, and CDN traffic distribution

Attention: Do not say “we would just add more servers.” Coupang interviewers want to hear about isolation patterns, circuit breakers, and fallback behavior. Show that you understand the blast radius of each failure mode.

Step 7: Discuss trade-offs explicitly#

Trade-off reasoning is the strongest senior signal in a Coupang interview. For every major decision, articulate what you are gaining and what you are giving up.

Common trade-offs to discuss:

Batching vs. immediate dispatch: Batching reduces delivery cost but increases average delivery time. Coupang balances this by batching only within SLA-safe windows.
Strong vs. eventual consistency for inventory: Strong consistency prevents overselling but introduces latency. Eventual consistency is faster but requires reconciliation and compensation logic.
Centralized vs. decentralized routing: A centralized engine produces globally optimal routes but is a single point of failure. Regional engines are more resilient but may produce suboptimal cross-region routes.

Pro tip: Use the phrase “the tension here is between X and Y” to frame trade-offs. It signals structured thinking and makes your reasoning easy to follow.

Step 8: Scaling and evolution#

End your design with a forward-looking discussion that demonstrates long-term systems thinking. Coupang is actively investing in several areas that make excellent closing topics.

Demand forecasting with ML: Pre-positioning inventory based on predicted regional demand reduces delivery distance and time.
Warehouse robotics: Automated picking and sorting systems change the throughput characteristics of fulfillment centers and require different coordination patterns.
Edge-based routing: Running routing algorithms closer to regional hubs reduces latency and improves resilience against central system failures.
International expansion: Designing for multi-region, multi-currency, and multi-regulatory-framework operation from the start.

These topics show that you think beyond the immediate problem and understand how systems evolve under business pressure.

Example: High-level design for Coupang’s Rocket Delivery system#

Let us walk through a condensed example that ties together everything discussed above. This is the type of coherent, end-to-end answer that earns strong hire signals.

Requirements: Same-day delivery for orders placed before a cutoff time. Accurate inventory with no overselling. Dynamic courier assignment based on proximity, capacity, and SLA urgency. Real-time tracking with ETA updates every 30 seconds. The system must handle 2 million orders per day with 10x peak capacity during sales events.

The request flow begins at the API Gateway and proceeds through four phases.

Phase 1, Order Placement: The Order Management Service validates the request and calls the Inventory Service to create a soft reservation. If the reservation succeeds, the order is persisted and a payment request is initiated. An OrderCreated event is published to Kafka.

Phase 2, Fulfillment: The Warehouse Management System consumes the OrderCreated event and assigns the order to the optimal fulfillment center based on item availability, geographic proximity to the customer, and current warehouse load. It generates a pick list and enqueues the task for warehouse workers or robotic systems. PickComplete and PackComplete events are published as the order progresses.

Phase 3, Delivery Assignment: The Delivery Assignment Engine consumes the PackComplete event and runs its scoring algorithm against available couriers. It considers proximity (weighted heavily), current package load, remaining shift time, and the order’s SLA deadline. The winning courier receives the assignment on their mobile app, and a DeliveryAssigned event is published.

Phase 4, Tracking and Delivery: The courier’s app streams GPS coordinates to the Tracking Service via the Ingestion Gateway. A stream processor applies Kalman filteringA mathematical algorithm that uses a series of noisy measurements observed over time to produce estimates of unknown variables that tend to be more accurate than single measurements, commonly used to smooth GPS trajectories. to smooth location data, updates the courier’s position in a time-series store, recalculates the ETA, and pushes updates to the customer through the Notification Service. Upon delivery confirmation, a DeliveryCompleted event closes the order life cycle.

Real-world context: Coupang’s actual system processes the transition from order placement to courier dispatch in under 30 minutes for many items. This tight cycle requires every service in the chain to operate with sub-second internal latency and near-zero queuing delay during normal operations.

This design balances speed, accuracy, observability, and cost. Each component can be scaled independently, failures are isolated by service boundaries, and the event-driven architecture provides a complete audit trail for debugging and compliance.

Mistakes that cost candidates in the Coupang interview#

Even technically strong candidates fail Coupang interviews by making avoidable errors. Here are the most common pitfalls.

Ignoring physical constraints. Designing a warehouse system without considering pick rates, aisle congestion, or shift schedules shows a lack of domain awareness. Coupang is a logistics company first.

Treating inventory as a simple counter. Inventory at Coupang involves reservations, holds, returns, damaged goods, mis-scans, and multi-warehouse splits. A naive increment/decrement model will draw immediate follow-up questions you cannot answer.

Skipping failure scenarios. If you only describe the happy path, interviewers will assume you have not operated systems at scale. Proactively address what happens when things break.

Over-engineering early. Do not start with a microservices mesh, service discovery, and distributed tracing before you have established the core data flow. Begin simple and add complexity in response to specific constraints.

Attention: Coupang interviewers often ask “what happens if X fails” as a follow-up to every component you describe. If you cannot answer confidently for your own design, it signals that you have not thought deeply enough about reliability.

Understanding these pitfalls prepares you to present a design that is both technically sound and operationally credible. Let us wrap up with final guidance on pulling everything together.

Conclusion#

The Coupang system design interview sits at a unique intersection of distributed systems engineering and real-world logistics. The two most critical things to demonstrate are that you can design for inventory accuracy under concurrent load and that you can reason about physical delivery constraints, courier assignment, and SLA enforcement as essential system requirements. Every architectural decision should be justified through the lens of cost, reliability, and operational reality rather than theoretical elegance.

Looking ahead, Coupang’s continued investment in warehouse automation, ML-driven demand forecasting, and potential international expansion will make these system design challenges even more complex. Candidates who can speak to how their designs evolve with autonomous delivery vehicles, cross-border logistics, and real-time dynamic pricing will stand out.

If you can walk into the interview and design a system that a Coupang engineer would recognize as operationally viable, not just architecturally sound, you are already ahead of most candidates.

Written By:

Areeba Haider

Free Resources

blog

Amazon System Design Interview Questions

blog

The top 6 system design interview mistakes to avoid

blog

What is Redis? Get started with data types, commands, and more