South Korean retailers are moving from recommendation engines to AI-powered shopping assistants, but as personalization becomes more sophisticated, the industry’s biggest challenge may be earning enough trust to access the data that makes AI useful.
South Korea’s retail industry is entering a new phase of AI adoption. The technology is no longer limited to recommending products based on past purchases. Retailers are beginning to deploy conversational assistants that can understand preferences, answer questions, recommend products and guide customers through the shopping journey.
CJ Olive Young, for example, introduced AI shopping assistants at stores with high foreign-customer traffic in July. The kiosks can provide product information, inventory checks, in-store navigation and recommendations based on customers’ skin characteristics and preferences. The system supports eight languages, while a separate AI interpretation service supports 38. Olive Young also plans to use shopping data to understand customer inquiries and behavior.
The same transition is happening at the platform level. South Korea’s Personal Information Protection Commission (PIPC) recently reviewed Naver’s AI Tab, an agentic AI search service that can use information such as service history, interests, preferences, purchase history and other behavioral data to generate more contextual responses.
This creates a new question for Korean commerce: how much personal information should an AI system know to make shopping genuinely better?
Personalization Is Becoming More Powerful
Traditional personalization was relatively straightforward. A retailer observed what a customer searched for or purchased and used that information to recommend something similar. AI can go much further. An agent can combine multiple interactions to understand context and intent. Instead of simply predicting what a customer might buy, it can potentially infer what the customer is trying to accomplish and construct a response around that goal.
That makes personalization more useful, but also more sensitive. The PIPC’s review of Naver’s AI Tab illustrates the problem. The regulator allowed the service to use certain platform-generated information for personalized responses, but required clearer disclosure of the categories of data being used, accessible opt-out mechanisms and safeguards against the inference or use of sensitive information. It also instructed Naver to prevent outputs containing uniquely identifiable information.
The distinction is important. Privacy in AI-powered commerce is no longer only about what companies collect. It is increasingly about what AI can infer from information that has already been collected.
The Data Flywheel Is Becoming a Competitive Asset
For retailers, the incentive to build this capability is obvious. More customer interactions generate more behavioral data. More data can improve personalization. Better personalization can increase engagement, creating still more interactions. That produces a potentially powerful cycle:
Customer interaction → behavioral data → AI personalization → better experience → more interaction.
Olive Young’s plans demonstrate how this is moving beyond the online shopping cart. Its AI assistant can provide recommendations while also generating data around customer questions, dwell time and shopping needs that can inform store operations.
For Korean retailers, therefore, customer data is becoming more than a marketing resource. It is increasingly part of the infrastructure needed to build better AI systems. But that also makes data governance a competitive issue.
The Cost of Getting Data Governance Wrong Is Rising
The consequences became particularly visible with Coupang. Following a data breach affecting more than 33 million users, the PIPC in June imposed a KRW 624.681 billion penalty on Coupang and additional sanctions on its fulfillment subsidiary. The regulator said the breach resulted from inadequate baseline security and management practices rather than sophisticated hacking. It also found problems involving access controls, data deletion and the unlawful collection of online behavioral data from external websites and applications.
The episode is relevant to AI personalization because it illustrates the other side of the data equation. The more valuable customer data becomes, the more expensive poor data governance becomes. This is particularly important as retailers seek to connect information across purchases, searches, preferences and other interactions. The same interconnected data that makes an AI assistant more useful can also make the consequences of unauthorized access or inappropriate inference more significant.
Korea Is Trying to Move Privacy Upstream
South Korea’s regulators are responding to this shift by trying to address privacy risks before AI services become fully embedded. The PIPC’s 2026 policy agenda calls for moving from primarily ex-post sanctions toward preventive oversight and a framework for an AI-embedded society based on privacy trust. It is also developing guidance for advanced AI, including agentic systems, while exploring ways to enable safer data use.
The Naver AI Tab review is an example of that approach in practice. Rather than waiting for a service to launch and addressing privacy problems afterward, the PIPC worked with Naver before launch to establish safeguards around data use and personalization.
For Korean AI and retail startups, this could become strategically important. Compliance may increasingly need to be designed into the product rather than added after development.
Sunay Kumat, an angel investor, told KoreaTechToday that “AI-led personalization will be a competitive advantage initially, but over time it will become basic expectations, and the real winners will be the brands that use AI to create a much better customer experience.”
That observation points to the larger competitive shift. If every retailer eventually has access to comparable AI models, personalization itself will become less distinctive. The differentiator could instead be whether customers trust the system enough to allow it to become genuinely useful.
That means the winning AI shopping assistant may not simply be the one that knows the most about a consumer. It may be the one that makes the consumer understand why a recommendation was made, what information influenced it and how much control they retain.
Privacy could therefore become part of the customer experience rather than a legal notice buried in a policy page.
Korean retailers have strong incentives to make AI more personal. Consumers can benefit from faster discovery, better recommendations and more convenient shopping. But personalization becomes increasingly difficult to sustain when customers feel that the technology knows more about them than they intended to reveal. The emerging equation is therefore more complicated than more data equals better AI. It is closer to:
More relevant data → better personalization → greater convenience → greater responsibility.
South Korea’s retail and technology companies are now testing how far that equation can go. The long-term winners may not be those that collect the most information. They may be those that can turn personal data into useful experiences while giving consumers enough transparency and control to trust the system. As Sunay suggests, personalization may eventually become table stakes. When that happens, trust may become one of the few experiences AI cannot automate on its own.
