Driving Collaboration and Optimization at Coupang

The INFORMS Roundtable is the recognized premier destination for top-level leaders in OR/MS practice, and it delivers unique benefits to member organizations, their representatives, and the broader INFORMS community. The Roundtable consists of approximately 50 institutional members, with each member company typically represented by their overall leader of OR/MS activity. The Roundtable convenes three times a year with the goals of enabling member organizations to learn best practices in managing OR/MS teams, supporting the professional growth for Roundtable representatives, and helping the OR/MS profession continue to thrive. The Roundtable also has an important advisory role within INFORMS on all issues related to the practice of OR/MS.This series of articles is designed to share with the broader INFORMS membership valuable information and insights into the practice of OR/MS and advanced analytics. Click here for more information about the INFORMS Roundtable.

Coupang is a technology and Fortune 150 company listed on the New York Stock Exchange (NYSE: CPNG) that provides retail, restaurant delivery, video streaming, and fintech services to customers around the world under brands that include Coupang, Eats, Play, Rocket Now, and FarFetch. 

Coupang offers next day by dawn or same-day delivery of retail orders, and the company has deployed advanced operations research across its entire supply chain. Close collaboration among the company’s Global Operations Technology, Optimization Science, and Fulfillment Operations teams has enabled the in-house development of optimization systems spanning network-level order routing, facility-level picking and packing, and logistics routing. 

These efforts have transformed how Coupang balances quality, speed, and cost at scale, allowing the company to respond quickly and accurately to strategic questions, including where to place inventory, where to position fulfillment capacity, how to allocate orders across the network, how to maximize throughput inside each fulfillment center, and how to plan middle-mile and last-mile transportation.

Leveraging Operations Research

For an operations research scientist, delivering on the promise to move the right inventory to the right place at the right time, then pick, pack, and ship orders in time for dawn or same-day delivery, is a continuous optimization challenge. Coupang’s fulfillment network; fulfillment centers; and transportation and last-mile operations must all work in concert. What began as rule-based allocation and manual planning has evolved over the past five years into a supply chain powered by mixed-integer programming, digital twin simulation, and generalized vehicle routing. This transformation reflects a core directive: leverage operations research to enable, scale, and improve the supply chain.

Coupang operates on a massive scale, serving millions of customers with millions of daily orders. Bringing products to customers’ doors by the next morning or the same day requires a trifecta of quality, speed, and cost efficiency. The business is most successful when an order is delivered on time, intact, and without the customer having to chase it. Processes that once relied on heuristics and manual oversight have reached a scale where optimization drives significant value. 

Replacing “what feels right” with objective functions, constraints, and simulation has become the norm. Teams not only ask, “Does this make sense?” but also, “How do we model this in the optimizer, and when will the simulation results be ready?”

The characteristics of e-commerce add complexity to the supply chain. Coupang’s platform connects millions of customers to a vast catalog. Fulfilling that demand requires a multi-echelon network: fulfillment centers, sorting hubs, and transportation lanes, each with capacity, cost, and service constraints. 

In the past, operations teams maintained priority mappings for each facility – zip code combination around the clock. Facility managers allocated pickers and packers using rules of thumb. Dedicated teams spent hours each day building truck routes by hand. These approaches served growth for a time, but scaling to millions of daily orders required a fundamental shift to rigorous, data-driven optimization.

When designing and operating this system, standard off-the-shelf tools proved too inflexible. There was a persistent gap between the modeled world and Coupang’s business processes, and run times were often prohibitively long because the tools could not exploit problem structure. Close collaboration between Coupang’s Transportation, Analytics, and Operations Research teams enabled the in-house development of optimization engines that provide the right level of detail and flexibility. 

Those efforts have borne fruit in our ability to answer high-impact questions quickly and accurately: Where should fulfillment capacity be placed, and what does that mean for delivery 
speed? How should we allocate orders across the network to minimize cost while preserving service? What are the transportation and capacity needs for the coming days or weeks? How do we maximize throughput inside each fulfillment center?

Fulfillment Optimization

The problem of fulfilling customer orders from a multi-echelon network can be modeled as a constrained multi-commodity network flow problem. Coupang replaced heuristic-based order allocation with a mixed-integer programming optimization system that enables dynamic, cost-optimal routing decisions at scale.

This transition has resulted in significant savings in fulfillment costs while maintaining service quality and speed. The transition was not only technical, but also cultural. Teams now evaluate proposals through objective function analysis and constraint modeling, comparing alternatives with quantitative rigor rather than relying on intuition alone.

To sustain the pace of development and reduce deployment risk, Coupang built a digital twin platform that allows virtual testing of network designsand operational changes. A dual-track environment – production and simulation – dramatically lowers the risk of each change. The platform has enabled smaller, safer production rollouts and pre-validation of facility and capacity planning modifications. It is being democratized so that non-technical operations teams can independently assess whether proposed configuration changes are operationally safe and economically sound. Looking ahead, the initiative continues to expand through multi-period inventory optimization and zero-authorization configuration management, with the simulation platform as the validation engine.

Facility-Level Optimization

Within the four walls of each fulfillment center, associates pick and pack orders to meet the delivery promises made to customers. Coupang moved from fragmented logic and facility-specific workarounds to a generalized model grounded in operations research. Next-generation picking algorithms jointly optimize picker assignment, packer allocation, and re-bin wall utilization to maximize hourly throughput.

The transformation builds on earlier success in pick routing: the company not only chose better pick routes, but also changed facility workflows. A shift from batch picking to streaming picking architecture, guided by O.R. principles and validated through in-house simulation of end-to-end facility operations, has delivered substantial improvements in throughput and pick efficiency. The algorithms were validated on historical pick-level data and production tests before full deployment.

Beyond the algorithms, the cultural shift has been profound. Facility managers discuss throughput trade-offs and re-bin wall capacity constraints. When proposing new initiatives, the first question managers ask is, “How would we model this in the optimizer, and when will the simulation results be ready?” Infrastructure investments – robust data pipelines, feature flags for safe experimentation, monitoring, and in-house simulation – support this mindset. 

Ongoing work includes intelligent scenario sampling to reduce simulation load while also preserving inference quality. What impresses internal leaders most is how these initiatives 
work together. Fulfillment optimization decides which fulfillment center processes each order and generates optimized shipments; facility-level optimization takes those shipments and optimizes resource allocation and picking sequences. Each system feeds the next, creating a seamless optimization cascade across the fulfillment value chain.

Logistics and Routing

Coupang’s logistics network spans multiple domains, from collecting supplier pallets to delivering large items that may require installation. The Logistics and Routing initiative applies operations research across these domains, turning manual planning into automated, optimization-driven decision-making. The story is a textbook O.R. integration: labor-intensive manual routing has been replaced by algorithmic systems that scale with network growth.

The Milkrun Route Optimization system exemplifies this. It automates middle-mile transportation planning for a complex inbound network. Previously, dedicated operations teams spent hours each day routing trucks. Now, the optimization engine generates cost-optimal routes in a reasonable timeframe. 

Business stakeholders aligned on translating operational constraints into cost-equivalent terms penalize multiple trips to a customer destination, incentivize driver-route familiarity, and penalize variability in daily payout across drivers. These actions are modeled within a generalized Vehicle Routing Problem with Time Windows (VRPTW) and capacity constraints. Exception handling that once took hours now takes minutes, enabling rapid replanning when operations change.

The same platform addresses seemingly disparate use cases through a unified VRP framework. Objectives such as maximizing fixed truck utilization, minimizing excess cost for remaining demand, and maintaining service commitments are incorporated via a feature-flagging system. 

This system serves both a large, inbound logistics network and post-purchase installation dispatch. The logistics network involves thousands of pallets that are collected daily from suppliers and delivered to fulfillment centers via cross-docks. In post-purchase installation dispatch, technicians are sequenced for appliance installations of varying difficulty. Future enhancements will include integration with customer-facing apps for real-time ETA updates and back-end analytics for driver roster optimization to maximize customer surplus.

A number of choices help explain Coupang’s ability to scale and evolve: a centralized view of operations that connects network, facility, and logistics; combined teams of operations research scientists and engineers who own the full path from model design to production systems; and executive sponsorship that has made operations research a strategic differentiator rather than a supporting function.

Operations research has moved from a peripheral tool to a foundation of how Coupang operates. The transformation spans network routing, facility operations, and resource allocation. Beyond technical achievements, O.R. has catalyzed a cultural shift toward analytical thinking. Operations teams evaluate proposals through quantitative analysis; engineering teams design systems with optimization in mind; and business stakeholders understand trade-offs through discrete-event simulation. Impact has exceeded initial projections, and the commitment continues: multi-period, multi-echelon optimization; zero-authorization config management with simulators as the validation engine; and customer-facing ETA integration with back-end driver roster optimization remaining on the roadmap.

With operations research embedded across the entire fulfillment value chain, from strategic network design to tactical route planning, Coupang has demonstrated that O.R. – when embraced at the highest levels and implemented with technical excellence across network, facility, and logistics – can transform a company’s operational positioning and competitive advantage.