Coupang
multi-surface recommendation and personalization engine
Coupang runs an in-house ML platform that powers recommendation, search, and ads across all its apps, with more than 600 ML projects and more than 100,000 workflow runs a year, adopted by all its major ML groups.
Objective
Personalize every surface of the app (home, search, ads) and industrialize the productionization of models to support both revenue growth and operational efficiency.
The deployment
Coupang runs an in-house foundation, the Coupang ML Platform, which industrializes the model lifecycle end to end: exploration, training data preparation, development, and productionization. This foundation powers the recommendation and personalization surfaces of the app (home feed, search, product ads) across Coupang shopping, Coupang Eats, and Coupang Play, as well as price forecasting and search query understanding via language models. According to the engineering blog published in September 2023, all of Coupang's major ML groups use one or more of the platform's services, with more than 100,000 workflow runs over a year across more than 600 ML projects. Coupang presents AI as a long-term lever for revenue growth and margin expansion.
Results Proof C
Coupang's official engineering blog (interested source, infrastructure figures) and the public company's financial results for measuring scale. Coupang does not publish a conversion gain attributed precisely to recommendation; what is proven is an industrialized, durable deployment (more than 600 ML projects, more than 100,000 runs a year) and a scale confirmed in financial results. C out of caution, absent a quantified per-surface lift.
How it works
Documented architectureThe stack in detail
- infra Coupang ML Platform in-house MLOps platform that industrializes exploration, data preparation, training, and productionization (more than 100,000 workflow runs a year, 600+ projects)
- llm Modeles de recommandation et de ranking proprietaires in-house models that personalize the home feed, search, and ads across Coupang, Coupang Eats, and Coupang Play
- llm Ko-BERT Korean variant of BERT integrated by Coupang for search query understanding
- outil Modeles de prevision de prix in-house price forecasting models fed by the same platform
How it runs, concretely
For ops teams-
1Data preparation data team
The platform prepares the training data from customer behavior per surface.
-
2Model training data team
The ML groups develop and train the recommendation, search, and price models via the platform.
-
3Productionization AI
The models are deployed on the app's home feed, search, and ads.
-
4Feedback loop AI / data team
Clicks and purchases feed back into training to adjust the recommendations.
Browsing and purchase behavior per surface. If the feedback data (clicks, purchases) is not clean and fresh, recommendations drift.
How your customers perceive this type of use
Sourced studiesLe paradoxe est documente des deux cotes : 71% des consommateurs attendent des interactions personnalisees et 76% sont frustres quand elles manquent (McKinsey, 2021), mais 75% declarent ne pas acheter aupres d'organisations auxquelles ils ne confient pas leurs donnees (Cisco, 2024). La « creepy line » est localisee : messages recus quelques secondes apres une recherche et suivi de localisation sont les pratiques qui mettent le plus mal a l'aise (Periscope by McKinsey, 2019).
Acceptance conditions
- La confiance dans le traitement des donnees precede l'achat : 75% ne achetent pas sans elle (Cisco 2024)
- Un cadre legal protecteur rassure : 59% des consommateurs disent que des lois fortes sur la vie privee les rendent plus a l'aise pour partager des informations dans des applications IA (Cisco 2024)
- La personnalisation elle-meme est attendue quand elle est consentie : environ la moitie des consommateurs (US 55%, UK 52%) disent s'inscrire souvent ou parfois a des services personnalises (Periscope by McKinsey 2019)
Red lines
- Le message declenche quelques secondes apres une recherche ou un achat : deuxieme ou troisieme cause de malaise selon les pays (Periscope by McKinsey 2019)
- Le suivi de localisation percu comme de la surveillance : 40% de malaise en Allemagne et au Royaume-Uni (Periscope by McKinsey 2019)
- Le mesusage des donnees personnelles par l'IA, devenu la premiere inquietude des consommateurs, a 53% et en hausse (Qualtrics 2025)
Sources: McKinsey & Company 2021 · Periscope by McKinsey 2019 · Cisco 2024 · Qualtrics 2025
How to replicate
Inference, not sourcedData prerequisites
- Customer behavior logs per surface
- Clean, fresh feedback data (clicks, purchases)
- Catalog and search signals
Org prerequisites
- An ML industrialization platform (training, deployment, monitoring)
- ML teams per domain
- A measurement loop on engagement and conversion
Possible stack
- In-house or managed MLOps platform
- Recommendation and ranking models
- Language models for search
The plan, step by step
- Step 1Choose a single surface (for example the home feed) and audit the available behavior logs.Deliverable: Clean training dataset on the chosen surface
- Step 2Build the feedback loop (clicks, purchases) and the reproducible training pipeline.Deliverable: Versioned data and training pipeline
- Step 3Train a first recommendation model and deploy it in an A/B test on a fraction of traffic.Deliverable: Model in production on a limited share of traffic
- Step 4Measure engagement and conversion against the existing heuristic, iterate on the features and the ranking.Deliverable: Quantified review and winning model generalized on the surface
- Step 5Industrialize (monitoring, regular retraining) and extend to a second surface such as search.Deliverable: Reusable MLOps foundation and second surface underway
First step: Set up a clean feedback loop on a single surface (for example the home feed) before extending to search and ads.
The same pattern, proven in other industries
Function of Beauty
personalized product formulation by algorithm from a quiz
Henkel
genAI content supply chain for one-to-one personalization at scale
Hims & Hers
AI-driven personalized treatment routing on intake data (MedMatch)
Sources
- S1 Meet Coupang's Machine Learning Platform Interested party archive pending
- S2 Accelerating Coupang's AI Journey with LLMs Interested party archive pending
- S3 Coupang, Inc. Q3 2025 Earnings Release Primary archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.